https://www.listennotes.com/e/642868ade20b4d71b928680e1ab0620d

You're watching TBPN. Today is Tuesday, June 3rd, 2025. We are live from the TBPN Ultra Dome.

Still gets me, John.

The Temple of Technology. The Fortress of Finance. The capital of capital. We have a great show. We're going through a bunch of stuff. We're covering Co2's public market updates. We're going through some news about Nike, some news about Nvidia earnings. we're going through we have a couple great guests today and we're going to take you through the timeline of course.

Of course.

We also have a new gong cam which I think we should show off right now. Jordy why don't you give it a hit?

Let's do it.

Let's do it. Let's check it out. The production team has been cooking. Check it out. Oh yeah we have the cam. The whole horizontal cam is really really something really something.

There we go. Yeah, I didn't realize that it would be horizontal. Oh, it's great, but it really is just about extremely chaotic. You want to see the vibration?

You want to see the vibration? It's swinging. It's swinging. Every time you switch to ramp.com, you earn the right to hit the gong. Time is money, save both. Easy use corporate cards, bill payments, accounting.

And a whole lot more, all in one place.

Go to ramp.com. Let's give it up for ramp.com I wanted to cover this yesterday. We didn't quite have enough time, but Mary Meeker dropped the Internet Trends Report. Code two, Phil LaFleche, fired back. Fired back. Did not fire back. Very different take.

They're going deck for deck, John.

They're going deck for deck. The code two deck is a little tighter. They use Sam Altman's face instead of referring to OpenAI as just leading LLM company. They actually come out and say what OpenAI is doing. But it's a great read through. I gave it a full read through this morning and I thought it would be fun to go through on the show. And so this is from Masters Invest.com, I guess, is the account. It says, Tiger Cub, Phila Fontes, Co2s, public market updates. It's too early to tell, but it's surprising how bad sentiment is compared to economic hard data. And so it's a good read. There's Google.

It's crazy you can be Philip LaFont. Yeah. Still be a tiger cub.

Yeah, I know.

Still a cub.

Still a cub. I mean, Tiger Global is itself a tiger cub because they spun out of Tiger Management, Julian Robertson's fund, which is very fun. Still using the Tyler name.

Imagine if Tyler Cosgrove, our chief intern, spins out and just launches TB Global.

TB. No, TB Global.

Wow, he fully sent it. Don't get any ideas, Tyler.

Yeah, yeah. Anyway, this is an interesting fact that they kick it off with. 2025 was the worst start to the year through April 21st.

Bo.

Very depressing news. You hate to see it.

You hate to see it, John.

But it's really remarkable how quickly the year just turned from the Trump pump to the Trump dump, and it got real rough there for a second. So in the first, what, three and a half months, four months of the year, there was an 18% drawdown And that's compared to 2022 when the market drew down 16%, 2001 started the dotcom crisis, 12% in 2000, 10%. But then they go on to, and man, 1998 was a banger year, up 21%. That's the start of the dotcom boom. It's interesting because it feels like we're in the midst of like the AI boom and yet we had like one of the roughest starts to the year in terms of the market.

Probably healthy. in some way.

Yeah, yeah, yeah.

Maybe so that maybe Trump just, you know, the 4D chess was, hey, things are getting a little overheated. Let's let retail get another opportunity to buy the dip before we go on, you know.

I mean, I hate to, I hate to somewhat agree with you.

I mean, there was, but there was, I do think there was this, there was this sentiment that people were so bearish a couple months ago.

Yeah.

That it felt like retail was going to look really silly for buying the dip, because retail over the past five years has just been trained, always by the dip.

Always by the dip.

And it's played out generally well.

Yeah.

And so far, yeah. So good.

And so that's the, I mean, that is the interesting narrative here is like, is this a crisis or a correction? And is this really like the end of the AI boom? Is this the end of all the fun? Is the party over or is this just some slight turmoil? So CO2 does a great job of kind of giving examples of crises and corrections and how they kind of, they bucket together. And so for crises, they highlight in recent memory the dot-com crash, the global financial crisis, and COVID they put kind of in half crisis, half correction. And this was the question that we were talking about with the tariff. the tariff drawdown was, is this something that's out of the government's control? Can they pull it? And if they pull back from the brink, it winds up looking more like a quick down and up versus something.

That'S, you know, it's almost more of a reaction.

Exactly.

Exactly. A... Yeah, yeah, yeah, yeah.

And so it's a little bit... So because it's kind of manufactured by the administration, they can They can, by definition, pull back from the edge. And that leads to this falling more in the correction category, potentially. But that's the question that CO2 centers in on during this presentation. So for crisis, the first crisis that they highlight is the 2000.com crisis, which was incredibly long. And so the drawdown for the.com crash was 133 weeks. So over two years of the market sliding downwards. And then the recovery to get back to where they were took 242 weeks. Insane. The global financial crisis was similarly long, 74-week drawdown. So a little bit faster on the way down.

And Gen Z, or even young millennials, have not experienced a drawdown like that.

Like the global financial crisis?

Yeah, really, really ever, right? We have the new generation of investors has only experienced sort of kangaroo market, right?

Yeah, exactly.

Up and down and up and down and up and down and up and down.

Yeah, and so COVID was similarly a crisis, very, very significant, but the drawdown was only five weeks. But it was the craziest five weeks where there were circuit breakers breaking on the exchanges, and there was a constant, like, oh, the market's down another 5%, and then the market's down another 5% or 7%. And so those five weeks were miserable, but then the recovery happened very, very quickly, only 21 weeks to recover and get back to the previous high, essentially. And then they also highlight the peak to trough for the S&P 500. You can think of the dot-com crisis as a 49-50% drawdown. The global financial crisis was even worse at 57%. You're thinking about like, you just bought, all you own is S&P 500 index, and you're down almost 60% of your wealth, all the money that you've had invested. Very, very rough. And there were also commensurate earnings reductions in the dot-com and the 2008 crisis. companies actually started making less money. So it wasn't just speculative. And this happened in COVID too.

29% in '08.

Yes, exactly. And so significant impact. And that's the difference between a crisis and a correction. Yeah. Obviously corrections are shorter, but they also don't hit earnings as much because companies don't have enough time to really react and drastically change their strategies. Actually lay people off, actually cut back on spending, actually retool the business.

In the case of tariffs, they were sort of temporarily unsustainable, and then they got pulled back to the point of still meaningful, but potentially less of a broad market impact.

And so then the fifth indicator that Kowalchuk identifies here is the weeks of elevated volatility in the VIX. And so the dot-com crisis had 41 weeks of higher volatility. The global financial crisis had 49 weeks of elevated volatility. COVID had 26 weeks where the market was in turmoil, essentially, with elevated VIX. But then you go to the other corrections that they highlight. Black Monday, Gulf War, long-term capital management failure in Asia crisis, sovereign debt crisis in 2011, the Powell pivot in 2018, and the inflation interest rate hike in 2022. Now, they call it inflation, we call it the end of the Deserter era in Silicon Valley. We call it the SVB crisis. But that really was what was happening. There was too much money printed during the COVID relief era that led to some inflation and that resulted in higher interest rates, which of course blew up Silicon Valley and literally blew up Silicon Valley Bank. And the inflation correction, they don't put it in the crisis bucket, but it does have similar- yeah, it does have.

Similar- obviously the the SVB crisis wasn't directly because of overinvestment in venture. But venture got so ahead of its skis in 2021, in early 2022, that in some ways, the failure of the bank was this, that symbol.

Yeah, it was very symbolic, even though.

A lot of their issues was that they had a bunch of really long-term.

Yeah.

Loans that were priced well below what.

The open market was. Yeah, a mismanaged balance sheet. So yeah, that inflation crisis just to compare it to COVID.

It is funny that Silicon Valley Bank could not imagine effectively their balance sheet said, we cannot imagine a world with rates higher than 3%. I mean, it's just rocketed.

Yeah, I mean, I talked to people in the public markets who had the similar sentiment that interest rates would never go up because if they did, the government debt would be so unsustainable. And yet, interest rates had to go up at a certain point because inflation was a tougher pill to swallow. And so the Fed had to take their medicine and raise interest rates, which caused that correction. Now, that correction was pretty long, 40-week drawdown, 66-week recovery. 2022 was a rough year for sure. And the market drew down 25%. But CO2 still puts us in the correction category, not this crisis category.

Yeah, the crisis, just to give you some insight into '08, specifically Apple, went from $194 a share to $85 a share, 56% decline in Apple. Google fell 55%, 680. Alphabet was 685 to 300.

And those companies weren't even particularly indexed on housing. It was just that the housing market collapsed and when people lose their homes or their mortgages spike in payments, they stop buying iPhones. And so it really was just this broad economic sell-off and economic crisis, whereas the inflation issue. Yes, we saw a 25% drawdown in the S&P peaked trough, but the earning reduction was only 1% throughout that crisis, which is pretty remarkable because firms were still able-- and maybe that's a function of inflation, but firms were still able to-- this was the other term that was coined by Kyla Scanlon, who should have on the show, was the vibe session. This idea that the market was in turmoil, everyone was talking about how bad things were, but day to day, unemployment wasn't spiking. It didn't have all of the trappings of the global financial crisis, even though it felt very doom and gloom at the time. And there was a question about, can we get to a soft landing? Can we pull back from inflation? We wound up doing that, but it was still a very tumultuous time. And so the big question is, where do tariffs sit? Are we in correction or crisis territory? How long will this last? What is the drawdown? Already we're at 19% off of highs. we were in the beginning of the tariff correction.

Just to give you some insight back into venture in 2008, in 2008, venture firms raised 25 billion. 2009, people were saying basically, whoa, they only raised 15 billion in 2009, which then obviously last year was over 76 billion. very, very low.

Yeah. And so, Kottu goes on to kind of try and define the difference between crisis and correction a little bit more. So crisis, the drawdown, the duration of the drawdown is long. The magnitude is going to be more than 35%, whereas a correction, it might be less than 25%. Earnings cuts are going to be more than 20% with an average of 35% earnings cuts and corrections are less than 10%. Capitulation, peak VIX is more than a year for the crisis, but just a few months for a correction. And the sentiment bearish in both cases, boo. So the question is, great sound effect, is will we look back on 2025 as a crisis or a correction? And we've talked to a lot of people when the tariffs were happening who were telling us that this is an absolute disaster, this is a crisis, this is a permanent change in the dollar, permanent change in the global economic order. And things looked really, really bad as they were selling off. We saw whole days where the market traded down 5%, but we have obviously built back up. And so KOTU mentions here that 2025 may qualify as a correction. So they're tracking the S&P 500 since liberation day. The tariffs came in on April 2nd, just the day after April Fool's Day. It's rough timing.

Everybody says, was this a social media post?

Yeah, seriously.

Meant to go out a day earlier.

And so they highlight two key moments here. Is there a market put? Government announces a 90-day pause on tariffs. That was the actual market bottom on April 8th. And then on April 11th, the regional Fed president says there may be scenarios in which the Federal Reserve would need to add temporary liquidity, saying that if the tariff situation continues to worsen, if companies are really suffering, We might need to stimulate the economy through very temporary. Temporary, temporary.

Always temporary.

And so the big question is on earnings per share on corporate earnings, because that is one of the biggest differences between a correction and a crisis. In a crisis, the economy truly stagnates. And we've seen that over the first 15 months on average of a true crisis in CO2's definition. earnings per share dropped by 35%. In corrections, it's more like 3%. And so we are now about halfway. We're about eight months into this crisis as they're tracking it. And 2025 is at just 5% off of earnings. So not really tracking with the average. Doesn't mean that we couldn't see things fall, but that's where they're seeing that. And then also the VIX spike was brief. the volatility peaked and has since fallen from 52 to 23. And if you're trying to track any of this or you're trying to get in on the action, public.com investing for those who take it seriously, they got multi-asset investing, industry leading yields, and they're trusted by millions of folks.

Serious investing to check it out.

And public. Let's see. So the question is, where are we today? 48 days in, there's a severe year-to-date drawdown, which we saw 19% peaked trough in the last 24 days.

Rapid stabilization. We love that.

We love rapid stabilization. So we're up 14% since year-to-date low in the S&P. So the question is, are we on the verge of a breakout or is this a false start and we'll retest the lows? And so they introduced the Co2 Hard Truth Index, a great coinage.

Love to see a coinage.

And this is economic hard data divided by sentiment.

I would actually name this the LaFont Truth Index.

Truth Index. Just the truth.

Just the truth. Purely truth.

And I mean, this is a good, this is a good kind of more quantitative way to look at what Kyle Scan was talking about with the vibe session, because you can open up the COVID of the Financial Times and see a bunch of doom and gloom and the sentiment can be very bad even among consumers. But when Christmas comes around-- If they're.

Good for the news business. It really is, unfortunately.

Yeah. And so the question is, what is the strength of the consumer's balance sheet? How is the strength and the true health of the American consumer? And what's the state of the job market? So they identify, CO2 identifies a few different metrics that they're tracking. Lagging data like job market inflation, leading indicators like consumer spending, orders, bookings and permits, and then bleeding edge indicators like proprietary CO2 data available on a daily basis because of course they're tracking all this stuff very closely. Then on the sentiment side, they're looking at business sentiment. Are businesses investing in growth or delaying capital and hiring plans? Is CapEx increasing at important companies? Investors, are investors optimistic about future returns? So there are polls that go out to to investors just to ask them about their sentiments. And then if you aggregate all of those, you might get a picture of how the sentiment in the stock market is doing. And then you can decide if that is, if the investor sentiment is out of step with the economic hard data. And then of course, consumer sentiment, you can look at consumer spending or and saving rates. And so in terms of sentiment, it's not looking good. 25 year sentiment indicators are at peak.

Negativity across every sector. So investors, Businesses and consumers, everybody is bearish.

Everyone's blackpilled.

And I'm a contrarian, John. No, it's funny. I was having a conversation with a buddy of mine. I won't name it, but he started a widely used unicorn. And we were grabbing lunch last week, and the restaurant was just absolutely packed, middle of the day. It felt like it was and it was a restaurant that both of us go to routinely. We're like, We haven't seen this place this packed in a very long time. And it felt like the Roaring 20s.

I was about to drop in there.

And granted, that's in Los Angeles on the coast.

People are wealthy.

Yeah, but it still felt, you know, in that environment, if you asked everybody individually, how do you feel about the economy right now, whether they're a business owner, investor, consumer, everybody would it would have been like, yeah, it feels shaky, yet their actions are saying something else, which is that the consumer is actually decently strong, right?

Yeah, yeah, yeah, yeah, yeah. It is very interesting. And I mean, it comes back, I mean, we're gonna get into this, but it comes back to that idea of like the picking up pennies in front of the steamroller that is AI. So there can be like all these crazy things that are happening that make so much sense when you talk about.

When the entire stock market initially happening. It was like, Hey, is this a distraction from just winning in AI?

Yes.

And Jensen is obviously conflicted, right? Like every CEO, he's heavily, he's basically an index of AI progress. But in many ways, like if we are entering a new type of industrial revolution in intelligence.

Yeah.

And probably that's the only thing that matters is to really win.

I mean, we should probably cover this when we go through NVIDIA earnings, but as soon as the tariffs hit and NVIDIA was hit very hard, I remember there was this massive drawdown in NVIDIA specifically. And there was huge questions about NVIDIA's China business and just overall GPU demand, Deep Seek and Jevon's Paradox stuff. He was very much saying, no, things are going to continue to grow. We're going to be fine. And the big question is like, was that ever cope or was he just actually right the whole time? Or did he, more importantly, execute very strategically in a tumultuous time to outmaneuver something that could have actually been pretty bad for his business? And so I'm sure we'll go into that during the NVIDIA segment. But if you're speaking of widely used unicorns, you gotta get on Figma, figma.com, think bigger, build faster. Figma helps. Faster. Design and development teams build great products together. It is a tool we use all day long.

It's amazing. We know we love Figma.

Anyway, there are some more consumer data points that co2 calls out here. One from Visa. All spend bans remain resilient and consistent with past quarters. Overall discretionary and non-discretionary spend remain strong. We have not seen any signs of overall consumer spending weakening, says Visa on April 29th. What was that? Pump it up?

That's Ben. That's Ben.

Okay. Ben.

Piped down.

Capital One also says spend trends were largely stable through the end of Q1. Bank of America says consumers are still spending money. American Express says we're seeing our customers act as if they have acted in the past. What's happened with the stock market or consumer confidence hasn't been associated with our card member spend. They continue to spend and they're not spending off what's in the market. And so consumers continue to spend. And so there's this, like the vibe session narrative is still going strong, which is very interesting.

Yeah, I mean, it's interesting because obviously we're seeing exceptional growth in AI, but we're also seeing pretty exceptional growth in the core business infrastructure. So Ramp, other companies in the payroll space, like Rippling, things like that are seeing tremendous growth as well. And I just, you know, and AI is also a factor in that.

But so co to actually co to actually charted this hard truth index and they and they track econ they track sentiment as the blue line here and then they track green and green hard data. And so the Evercore company sales report. And so these should track very closely. like the sentiment should match the hard data. But from 2021, 2022, 2023, sentiment was lagging the hard data, which was very, very positive. And then more recently, sentiment has taken a massive spike down at -1.4, where the hard data is basically neutral at -0.3. And so, again, we're seeing more quantitative analysis of the Vibe session, and it just doesn't seem to be going anywhere. And I do wonder why we're seeing, you know, this data goes back 25 years, and the dislocations are pretty, pretty rare. You can see there's, it feels like there's a little bit more, there's sometimes a little bit of more oscillation where the hard data is a little bit more aggressive than the sentiment. Actually, during the.com boom, sentiment was more neutral, even though the hard data was worse. And same thing in same thing in the in in in the global financial crisis.

Everybody's everybody's like, you know, spraying champagne in each other's faces.

Meanwhile, the economists are like, this is terrible. Like we're cooked. You were cooked. You were valuing companies with $5 million of revenue at 10 billion. What were you thinking? Everyone's like, it's fine. It's fine. Well, now now we're seeing the opposite. And so to recap, KOTU sums it up, says this does not seem like a crisis yet, more like a sharp correction. It's too early to tell, but it's surprising how bad sentiment is compared to hard economic data and someone it must be wrong between Wall Street and Main Street. And so this is this is the question that we're trying to trying to dig into. So sentiment sentiment alone, they say, is not enough of a buy signal when business sentiment is below the 10th percentile. The S&P 500 has top 20% one year return 35% of the time, 8% of the time when business sentiment is below the 10th percentile. The S&P 500 has a bottom 20% one year return. And you don't know? 57% of the time.

More than half the time, you just don't know.

You just don't know. So based on history, they're looking at next 12 months price to earnings of the S&P 500, and the market still looks expensive. But of course, a lot of that is driven by tech companies that are growing very quickly. We talked about this interesting data point that something like the Mag 7 grew revenue in Q1 at something like 30% annualized, where the rest of the Fortune 500, the 293 that aren't magnificent apparently, they grew at like 8% annually.

Less magnificent.

Less magnificent. And so, I mean, the interesting thing here is that tech has seen a notable multiple reduction. And so, if you can track the price to earnings multiple of all the tech companies, which here are in, which one is which? So the tech companies are in blue and the non-tech companies are in purple. And the tech companies, you go back to the dot-com boom, they were trading at incredible price to earnings multiples, up at 45x on average. And then post COVID, there was a dislocation again where tech was pulling away, but got up into the 20s and 30s, but is still trading at 24X earnings where the non-tech companies are at 18X. So there's still this spread. So during COVID there was an 8X spread. AI had a 9X spread, and the current spread is 6X. And so AI is not the subcategory, but the trend that kicked off in 2024 when the AI trade got really, really heated up. And so some tech franchise names look more reasonable today.

We keep coming back to this, but Google is actually down 3% over the last year. So again, despite being one of the greatest AI innovators, just not getting any traction.

It should be a massive beneficiary even this year. But of course, the stock has traded up a bunch. And so some tech franchise names look more reasonable today. Reddit is down 49%, Tesla is down 40% off of the 52-week high. Arm is down 34%. Shopify is down 23%, Nvidia is down 23% off the peak, Amazon 22%, Meta off 19%. And so there are still some tech companies that are lagging their all-time highs. And it's always interesting to go to the public filter for what are the power law, super dominant tech companies that are still in founder mode, still led by founders. and investigate those businesses and see if they do have, if they are set up for the next wave, they could be good, they could be opportunities, especially if they're getting beaten up for kind of some macro trend that they will, that you would expect a tech founder to be able to dig their way out of the hole, like what happened in COVID and like what happened during the Powell pivot and the interest rates. like when interest rates went up, Shopify, Palantir, who else? I mean, a lot of these tech companies got absolutely beaten up. And the ones that really emerged, a lot of them were founder-led, a lot of them were willing to not really change course, or if they needed to change course, they did it quicker than the other companies. And so they included a quote from Charlie Munger here all the way back from 1998. Charlie says, forget what you know about buying fair businesses at wonderful prices. Instead, buy wonderful businesses at fair prices. And so the question is, is the glass half full or empty? Half full. You know, the, the bull case here, AI is exploring clearing order inbound. Macro is here. Macro will heal it. Will hear it will heal itself. The half empty, the value investor, they say there's a recession on the horizon. They say there's tariffs. Berkshire Hathaway has a 350 billion cash pile, lots of things that look not great. And so what is CO2's view? They believe that tokens are more important than tariffs. And this is the AI being more important than the token.

So CO2 is launching a crypto coin.

No, no, no. Don't even joke about that.

Yeah, don't we don't joke about that.

That's the one thing we will never joke about. It'll be taken out of context.

They'll be hump fun about it with you.

Yeah, in fact, just to clear the view, let's tell you about Vanta. Automate compliance, manage risk, prove trust continuously. The opposite of a fly-by-night meme coin, one of the most reliable businesses in the game. Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation, whether you're pursuing your first framework or managing a complex program.

Some of the best businesses in the world use Vanta. That's true. Ramp. Yeah. Duolingo. Intercom.

Very cool.

Mistral.

Really?

The French, prominent French AI company.

I love it. We got to get some from a straw on.

Iceland Air.

We haven't talked to them yet.

People never think about Iceland Air being on the cutting edge of AI.

Have you been to Iceland?

I have.

You have? Was it nice?

I had to go there. I once was in Africa.

Okay.

And I was in college and I had $400 in my account.

Okay.

And I had to get back to America.

So you went through Iceland?

And I charted the most heinous route from like From Morocco to London, through Iceland back to San Francisco, and I made it back.

What's the bull case for Iceland?

And the flight that was going to and from Iceland was, the chairs.

Were.

Like bench seats and almost made of plastic. It was a really, really long flight.

Like a bus.

But I made it home.

I feel like the bull case for Iceland Air is probably Greenland becoming a state or protectorate or something. Many people go there. And then Iceland Air probably is going to be the one to get you there, at least in the short term.

Totally.

They're going to have a direct flight from... Where will people be going to Greenland from? Probably Park City. Anyway, straight from Park City. Co2 goes into OpenAI and they're digging into this thesis that tokens are more important than tariffs. And so they asked, what just happened at OpenAI? The number of monthly active users, MAUs, post-launch. And so they're tracking the number of months post-launch for all of these companies. You can see Twitter now acts over 96 months. They grew to 288 million active users, monthly active users. Instagram's over a billion, Facebook's over a billion, TikTok's over a billion over that 96-month launch period. and you can see that these social networks, they grow exponentially, but it looks like a complete grind compared to what OpenAI is putting up. And of course, that's because OpenAI can travel so much faster because they can leverage the virality of the social internet and everyone already has phones and internet, whereas for something like Instagram, many people didn't even have iPhones yet. And of course, Instagram launched exclusively on the iPhone and took them a while to launch an Android phone. And I imagine that 36 month bump is probably them launching on Android, or maybe desktop, who knows. But OpenAI launched deep research in Operator just 24-ish months after they launched the original product, and they jumped from 400 million users to 800 million users. And this, of course, is based on a quote from Sam Altman, who says, the users doubled in just weeks. And so it's a little bit of like they're creating this graph somewhat speculatively. They're not exactly sure all the numbers, they're trying to work backwards. And there are, of course, questions about who are those users? Where are they? How monetizable are they? But all of that stuff is kind of something that investors will be digging into down the road. Right now, it's very clear that OpenAI is breaking through in a massive way all over the globe, honestly. And so, Kotou also highlights the comments of Satya Nadella for Microsoft talking about a major AI inflection. Microsoft says we processed over 100 trillion tokens this quarter, up 5x year over year, including a record 50 trillion tokens last month alone. So that quote just really says exponential growth in inference. And this is going to be very, very important for our discussion on NVIDIA earnings because Jensen previously had revealed the ratio between inference and training loads, and he hasn't given as much guidance on that now. And so there's a big question about how fast is the shift actually happening, how much demand is there, if you can compress these models and deliver a great product with less GPUs. Basically, you could see exponential growth in token generation, but if you see exponential savings in the amount of GPU cycles it takes to create a token, you see flat GPU adoption and roll out. No one's really expecting that. Everyone's building tons of these massive AI factories. And Jensen did actually go into unpacking the AI factory idea a little bit more, which I'm excited to talk about. Jensen continues, this is from the Microsoft earnings call on April 30th. While we continue to bring data center capacity online as planned, demand is growing a bit faster. Therefore, we now expect to have some AI capacity constraints beyond June, which is interesting because he was saying, oh, like maybe I'm going to be a leaser, maybe I'm not going to be a leaser.

Market clearing order inbound. The market is excited to be a leaser. Yeah, he said earlier this year he.

Was canceling some projects, but he said he's excited to be a leaser. And we saw actual capacity constraints at Google in their last earnings, remember? And so it'll be interesting to see. We see capacity constraints every day with you and Veo.

I know, I really-- Throwing your phone across the room, hitting getting rate limits.

I need to set a recurring reminder on my calendar for every night at 3. Yeah, because if I miss a day, it's like, oh, I can't get those back because I only get three a day and I'm paying the max. The actual best practice is to set myself a 15-minute calendar hold where I use VO to the back, basically. Although there's probably a better way to interface with it over the API or something like that. But anyway, Satchu goes on to write with deep reasoning and agent flows like Copilot Studio, customers can build agents that perform more complex tasks and also handle deterministic scenarios. Now, there's this interesting debate going on. I don't think we got to it in the timeline, The founder of Zapier was Wade Foster, not Mike, who's been on the show. Wade was saying, we are doing a lot of stuff with agents at Zapier, and he mapped out one of his agentic workflows, and somebody was kind of chirping at him being like, this is just a workflow. This isn't actually fully, this isn't AI, because you defined For example, you would have one agent that processes an inbound form and then another agent that... So let's say that we're doing inbound sales. You have a website and a landing page and then if somebody's interested, they fill out a form and say where they're from, what size company they work for, that type of stuff. You have one form that processes that and then that hands it off to an agent and hydrates and says, hey, go scrape their LinkedIn and see who else they're associated with. Are we affiliated with them? Are we already selling to them? Kind of like fact check that. then you might have a different agent that sits there and triggers, okay, based on what we know about them, now that we've hydrated the account, let's send them an email, or let's send them a phone call, or let's route it to Steve, who's already been talking to somebody else at that company. Yeah. Right? And then on and on and on. And so these are workflows, but they're powered by LLMs at every phase. But there's this question of, are they agents or are they not? It almost does you a half.

All I care about is business value. People want to chirp at Zapier because they're an established large company that's been operating forever, but you would have to apply the same rigor and framework to every company that says they're an AI agent.

Yeah. Yeah, I think people are maybe overly obsessed with the idea of, and I've certainly fallen into this trap before, this idea of like end-to-end learning, one-shotting, this idea that like it is, creating an ensemble approach of different models, different technologies, some business logic, some deterministic logic, some non-deterministic AI generating tokens and reasoning tokens to get you the result. It's all semantics. I mean, it does have an impact on the super long-term future and kind of the way we design these systems. But at the end of the day, most businesses just care about resolution.

Results.

Right. Results. We're seeing that at Microsoft, the Azure revenue, their cloud is accelerating. Microsoft Azure's revenue growth on a percent year over year basis through up in March of 2021, they were growing at 55%. That went down to 25% in March of 2023, and then it grew to 35% in March of 2024. It was down at 31% in September, and now it's on an upswing again.

We're incredibly back, is what Satish is saying. He also reported in that last earnings call that already 70% of the Fortune 500 is using Microsoft Copilot for a bunch of AI tasks.

70%? Yeah. I mean, we know what they have to do.

Which is, again, it's kind of like Google saying we already have have a billion people using AI overviews or whatever the metric is because they're just sort of slotting it in. Yep. Because, you know, I'm sure that 70% of the Fortune 500 already runs on Microsoft teams or whatever infrastructure they have set up. Or they run on linear.

Linear is a purpose-built tool for planning and building products. Meet the system for modern software development, streamline issues, projects, and product roadmaps. Linear for agents.

Check it out.

Speaking of- if you're not on agents.

Linear yet, what are you doing? Yeah, they got cursor, runway, complexity, retool, retool, retool, retool. Mercury ramps. Open AI scale for cell.

Arc.

Absolute dogs.

They're absolute dogs over there. It's fantastic.

They probably wouldn't describe themselves that, but we do. I don't think so. Yeah, we do. So speaking of agents, the reasoning unlock, this is back in the CO2 deck, agents, thinking on their own solving problems creatively. Quote from Sam Altman, We view this as the beginning of the next phase of AI, where you can use these models to do increasingly complex tasks that require a lot of reasoning. And so chain of thought reasoning is multi-step trajectory where the model thinks and iterates, it writes code, it searches the web, it creates the reports, build tables, and all of these. The model keeps asking, Does this make sense? I should verify the source. Let me edit my work. And so All this reasoning revolution, these agentic workflows, it feels very much like this narrative of like, oh, we kind of capped out on the previous thing. We need a new buzzword. But it's actually showing up in the data, and we're seeing it in upward revisions to cloud CapEx when you add up Amazon, Microsoft, Google, Meta, Oracle, and Tesla. Oh, we got the big six, and it's a different six.

It's a different six.

CapEx has gone from $213 billion to $310 billion, a 70% jump. And today, the CapEx forecast is up at $365 billion. Which is pretty crazy.

Again, Tesla's up here. Good work getting in. Not a huge contributor to this. They're expected CapEx is going to be around $10 billion.

Yeah, it would be interesting to see how Tesla and XAI kind of partner on the CapEx side, build these hyperscale data centers, because you're already kind of seeing that take shape with Stargate, where it's a separate entity. And within Stargate, there are also separate entities where who owns the land, who owns the power production, it doesn't necessarily make sense to always bundle all of those in the same company.

And so Oracle's projected CapEx for 2025 is $16 billion. So between Oracle and Tesla, you still have over $300 billion, well over $300 billion just coming from the core for Amazon, Microsoft, Google, and Meta.

Yeah, I mean, this is why we refer to the core four as the hyperscalers truly. And when you look at Amazon, Microsoft, Google, and Meta, you're seeing high double digits of CapEx in the billions, 60, 70, $80 billion going into CapEx. And that's driving the majority of that 365 billion of overall CapEx. But if AI wasn't real, if there wasn't real demand, it would be extremely hard to justify this level of CapEx investment. And so this is all driven by actual demand for Google Cloud, actual demand for Azure, actual demand for reasoning tokens.

I mean, Meta was in the news this week. Talking about how building they want to ultimately create the ads that they run. Yeah.

Well, speaking of ads, let's run an ad for numeral numeralhq.com, sales tax on autopilot, spend less than 5 minutes per.

Month on sales tax compliance. There is a fantastic company that runs on numeral Lucy. Lucy.

Yeah, my company runs on numeral. Anyway, going back to co2, they say tokens are greater, greater, greater than tariffs, exclamation point, exclamation point. I love it. This deck is, you know, obviously built by a crossover fund, hedge fund, like very successful, like very serious on Wall Street. I see them as, yes, they are a major player in venture, but they're also a major player on Wall Street. I see them as like a respected institution, and yet they're having some fun with the deck, and most importantly, they're clear. It's very clear, digestible, not too much jargon. Yes, there are some very small bullet points or footnotes that haven't been reading, but in general, done a great job of summarizing these things. They have three takeaways. From a market perspective, we remain vigilant but have confidence in the AI trend. Completely agree. Remain remaining flexible, tracking bleeding edge data carefully. AI could be such a game changer. We need to capture this trend. I think it's spot on. Spot on. And so that's their conclusion.

My only I know they can't give, you know, investment advice here, but I wish they took a slightly firmer stance and said, AI will be a game changer if we need to capture this trend. They sort of leave some ambiguity there. But yeah, I think that that's drawing it back to earlier stage venture that feels like people are doing rounds that they feel are overpriced, they're a little bit uncomfortable with valuations or the amount of investment that's being made at oftentimes with just a deck and a team, but at the same time, nobody can afford to miss AI. If you come out of the next five years and you don't have multiple bangers, don't expect to be a serious fund over the long run.

Yeah. So the polymarket on the US recession in 2025 is down at 30%, still a little bit higher. Wouldn't be that crazy if it did wind up happening. In general, that's looking pretty positive on Liberation Day. It was at 42% on April 1st, the day before tariffs. It was already climbing. I think people were expecting something to happen. It was very low at the start of the year, February of 2025. This was down at 22%, spiked all the way up to 66% post-trade war, was kind of hovering around 50-60%. more likely than not to happen, recession. And it's since been dropping and dropping in basically every day for the past few weeks, it's dropped and is now around 30%. But we got to get it to zero, folks. We got to get it to zero. It's a team effort over here.

It really is.

In the American economy, let's make it happen.

Don't just, you know, next time you're frustrated with the economy, don't be thinking, what's the economy gonna do for me? Ask yourself what you can do.

Exactly. Don't take it, don't take your frustration or your consumer sentiment out on the economy. It's okay if somebody calls you up, Hey, I'm from Pew Research. How you feeling? You can tell them the truth. You can tell them you're not feeling good. But when it comes time to do some Christmas shopping, just go all out. Just get on Bezel. Just buy the most expensive watch you can possibly afford. Do it. You're contributing to GDP. Yeah, I mean, your Bezel Concierge is available now to source you any watch on the planet. And what could help build the American economy more than fantastic luxury watches available on demand at getbezel.com anyway, the other side of this narrative going even yes, going even broader is there's now questions about the the US debt. And so Wall Street has been sounding the alarm on the US debt. This time it's worth listening, says the Wall Street Journal. Jamie Diamond says you are going to see a crack in the bond market. and he has plenty of company here. So I wanted to give a little bit of a review of this because even Elon Musk has been talking about the struggles of the government and how some of his optimism around writing the budget and creating a budget surplus did not come to fruition like he was hoping for. So the big question is, is the US going broke? It's not possible. Joe Weisenthal, we print dollars, we're good.

We have a dollar machine.

We have a dollar machine.

Yeah, the other element, this went out right as we were about to go live, but Elon posted at 10:31 about an hour ago. I'm sorry, but I just can't stand it anymore. This massive, outrageous, pork-filled congressional spending bill is a disgusting abomination. Shame on those who voted for it. You know you did wrong. You know it.

Wow.

So not mincing words.

It's rough.

And breaking rank from, well, he's not an employee anymore.

He's not an employee anymore. He's a citizen. He can voice his opinion however he wants. I think the bigger question is like, yes, this might be depressing to Elon. This might not be the outcome he wanted, but in many ways this is business as usual. and so how can the US economy steer its way through and how can markets react and how can companies build through what continues to be a pork-filled spending package as it has been for decades, right? And I think the answer we're gonna come back to again is just continue to dominate on the frontier of technology, continue to win in AI and space and all of the companies that Elon is already working on. So Elon is kind of refocusing on the thing that actually helps.

He had already shared something like a week ago that he said, I'm not convinced we can solve our problems politically through the government. We need to just grow GDP.

Exactly. And so if GDP does start growing at 5, 10% because of all the technology that folks are working on building right now, well, then that probably actually does start solving the budget deficit, unless we just find ways to spend more pork. who knows?

Cork, hopefully, cork filled spending bills.

Hold it in. So featuring an illustrated Uncle Sam with pockets turned inside out. That was the COVID story of America's most influential news magazine in March of 1972. Sounding the alarm about a debt crisis has been great for companies shilling gold coins and flashy financial products, but has made smart, sincere people look silly when nothing happened. Financial markets equivalent of Y2K. So why are several people suddenly worried? because the math is getting daunting with interest on the debt blowing past $1 trillion annually and Washington acting recklessly. Even people who have issued past warnings deserve a second or third or fourth hearing and they link to something which I'm sure is someone who has warned many, many times. So hedge fund manager Ray Dalio does have something to sell. His book, How Countries Go Broke Out Tuesday.

Yeah, so he launched this this morning.

Cool.

It's available now.

Let's invite him on the show. It'd be great to have him. Yeah. But the world's 172nd richest person is hardly staking his reputation on royalties and his arguments are compelling. Again, Ray Dalio, I actually did a, I made a YouTube video about like, when is a, like how would you quantitatively define kind of a, like a fake guru, someone who talks about business but doesn't actually, What is the difference between an entrepreneur or an investor that just talks or writes a book versus media is their only thing? And Dalio, yes, he has incentives around his business, but he does not make the majority of his money from his book. And so the optimistic case is that his book is is less about just selling copies. Yeah, it's more about actually like sharing his worldview or trying to influence like his broader business. Even maybe like hiring at Bridgewater could be more of a more of a needle mover for him than just selling copies. So I think it's worth investigating. Dalio told Bloomberg he gives America America three years, give or take a year, to avert an economic heart attack. Peter Orszag, chief executive of Investment Bank Lazard and former budget director, wrote last week, those who bemoaned the unsustainability of deficit spending and debt levels back during his time in government seemed to cry wolf a lot. Now he's worried, too, because the wolf is lurking much closer to our door.

The package. So this is what Elon was alluding to. The package of tax and spending measures sent to the Senate now officially called the one big the one big beautiful bill act could act like budgetary wolf bait. It would add around 3 trillion to debt levels over the next decade compared with existing estimates and 5 trillion if certain temporary features were made permanent according to the nonpartisan committee for responsible federal budget. For perspective, the federal interest this fiscal year already will be more than the defense budget and more than Medicaid, disability insurance and food stamps combined. Moreover, the Congressional Budget Office estimates assume that the bond market will not only tolerate a surge in spending but become more relaxed about it with lower yields. Consider if the yield on the 10-year Treasury note stayed at today's level around 4.4% for the coming decade, then the CRFB estimates it would add another $1.4 trillion in interest costs over that period. And what if yields surge instead in a vicious cycle? JP Morgan Chief and Absolute Chad Jamie Dimon warned on Friday of the consequences you're going to see a crack in the bond market, okay?

Yeah. So, yeah, it's interesting. I feel like as well, I'd like.

To, we should try to pull up what he was saying exactly before and after that because I'm assuming that it adds some important context.

Please, yeah, go pull it up. I'm just thinking about, like, throughout my life there's been whole news cycles around this idea that the defense budget is so massive and it's this massive part of the federal budget overall, or Medicare and Medicaid and disability insurance, these are huge, huge costs and they drive the majority of the federal budget. And now we're at a point where people would always talk about the defense budget or Medicaid as being maybe wasteful or maybe unoptimized, but at least you're getting something for that. You know, like with those budgets, even if it's 60% or 50% or 20% less efficient than it could be, at least you're getting something. With interest, you're literally getting nothing. You're just paying the debt. And that's it. It doesn't even have the opportunity to create any value or security or health care for someone. pure or just just paying back for something that after a long time ago.

After Jamie Dimon said, you're going to see a crack in the bond market. Okay. He says, and this is why the journal didn't want to include it. He says after stating this quote, Dimon emphasized that this bond market fracture is inevitable, though he is uncertain whether it will happen in six months or six years. He warned regulators directly that you are going to panic, but he himself would not panic and expects JP Morgan Chase to manage through it. He also highlighted that current bank regulations restrict banks' ability to hold bonds, limiting their capacity to provide liquidity during market disruptions, which could exacerbate the crack in the bond market. Anyways, he's seemingly frustrated with the new spending bill and spending broadly.

The Wall Street Journal continues, Yet the bond market isn't exactly collapsing, even if 30-year yields recently hit a post-crisis high. So who are you going to believe? Millions of fairly relaxed investors or some wealthy pundits? Another hedge fund manager, Paul Tudor Jones, calls the paradox an economic kayfabe, a.

Term from professional wrestling.

Let's give it up for professional wrestling. We love professional wrestling on this show. Round of applause. WWE, if you're listening, I'm ready to get in the ring. Let's do it. Let's do it. We could be the absolute, it's heel. that everyone hates. They'd be like, these slick tech bros, like the tech bro, and then we just get destroyed by like the-- Logan Paul. Yeah, Logan Paul. Great. So he says, those who know that the numbers aren't sustainable are happy to suspend belief while the show continues. Treasury Secretary Scott Bessent reiterated this past weekend that the US will never default on its debt. But it doesn't have to. Rapid inflation would accomplish the same thing if the Fed had to ride to the rescue through a measure called fiscal dominance. What is the tipping point for the.

Bond market to go from mild anxiety? Bad things, something so cool. Sounds great.

We do have fiscal dominance. Former IMF chief economist Kenneth Rogoff, an authority on debt crises, explained in April that they are never a matter of simple arithmetic. Almost every country defaults either through outright default or high inflation, occurs long before debt calculus forces it to. Despite Dahlia's guesstimate, knowing when doomsayers will be proven right is impossible. Consider Stein's law: if something cannot go on forever, it will stop. Contrary to urban legend, that line wasn't uttered by Ben Stein, who played the boring economics teacher in Ferris Bueller's Day Off, but by his father, Herb, an actual economist. Economist. Life comes at you fast. Very rough. Well, if you're losing sleep over the tumult in the public markets and the global economy, you got to get a good sleep. They have a Pod 5 Ultra, five-year warranty, 30-night risk-free trial, free returns, fast shipping.

Are you about to tease what I'm going to tease? I mean, we can riff it out. I don't know how to riff it out. John's had a good bit lately that eight sleeps are just so Well, comfortable that they actually, they should test them on animals.

Well, recently I've been doing some animal testing at home, actually, with my eight sleep. With your own dog. Yes. So my dog, obviously grumpy because there's now kids around the house and I think he's getting less attention and also he's just a grumpy old 10 year old Newfoundland. But I come home and he's always sleeping on the bed, on my bed, on my eight sleep. And he's doing it more and I don't know if it's correlation or causation, but I think he loves the sleep. Yeah, but he sleeps right in the middle. So he's getting like warm on one.

Side, cold on the other side.

He wants a bed. That's good. You can just roll to one side.

Yeah, yeah, yeah. So, yeah, proudly animal testing.

You can change the temperature on either side. Yeah. Perfectly adjust it.

Once he learns to use the app.

It'S gonna be a game changer.

You should get him an iPad by the bed. Yeah, we just with his nose. How do you like that?

I got an 87. I'm not proud of it. But I was up a little bit late, but I kind of slept in a little bit. I bet I got crushed on the.

On the regulation, the consistency, consistency, 82, 82. Okay. I got an 87. I just need to put up a little bit seven straight nights of. Okay, let's see.

Of sleep dominance. Sleep dominance. We'll see you tomorrow. Nvidia earnings. So from the Wall Street Journal, Nvidia's business is booming despite being shut out of China. We've been covering this for a while. Jensen's been running around the globe building AI factories everywhere on an absolute tear. And most recently, the Nvidia shares surged 5% after hours as quarterly revenue surged to a record 44 billion. Let's hear it for Jensen. I think you should hit the gong for Jensen. I mean, 44 billion.

Why don't you hit it, John? I want to see you on the.

Can we get the gong cam pulled up? Let's get the gong cam pulled up. Market clearing order inbound. There we go.

There we go. It just keeps ringing too.

Yeah, yeah, it just keeps ringing.

When we hit the soundboard, it's pretty.

Soundboards in and out in like five seconds, but the gong, the real gong remains. You can't beat it, it's the best. And so Nvidia's business is still booming, even with the company effectively shut out of one of the world's largest markets for advanced AI chips. The chip Titan has been on a roller coaster over the past few months after the Trump administration moved to limit sales on chips to the Chinese market. We're having Aaron Ginn come on later in the show, and he'll break down kind of what's happening most recently with Nvidia and chips. He's the GPU whisperer. You, of course, know him if you've been watching the show. Nvidia, which has emerged as one of the most valuable companies because its chips provide the computing power needed in a global AI arms race, posted another quarter of record-breaking sales on Wednesday, Let's hear it for them. Revenue reached 44 billion for its first fiscal quarter, a 69% increase that was curtailed by Washington's new limits on Chinese chip sales. The company was unable to ship 2.5 billion of its H20 processors and projected 8 billion in lost revenue for the current quarter due to the policy.

So you lose the flesh wound, John.

You lose 8 billion in revenue, which is high margin revenue too. We're not talking Amazon here and you still put up 44 billion and your stock goes up 5%. It's fantastic. Nvidia's longer term outlook in China remains a major question mark. The company admitted on Wednesday that has limited options.

The significant thing here, Nvidia surpassed Meta in quarterly revenue. Yeah.

Wow. That's insane.

Insane.

Yeah.

Yeah.

And I mean, Yeah, absolute dominance of the NVIDIA CUDA ecosystem. The number of developers on NVIDIA has been surging, as we saw in Mary Meeker's trends report. And the company seems very, very well run. At the same time, Dylan Patel, George Hotz, putting the screws to ARM. ARM seems to be, or AMD seems to be catching up, seems to be taking that feedback and trying to solve some of the problems that have been keeping developers off of that platform, even though the cost per flop has been quite good. on AMD. The ecosystem hasn't been quite there, but obviously everyone there is well aware of the opportunity of taking a little bit of Nvidia's market share. And so this is an interesting takeaway. Nvidia credited its new Blackwell gaming processors, part of the same family as its current lineup of AI chips that are seeing red-hot demand. Blackwell for gaming is a home run. Nvidia CFO Colette Kress said in an interview for the current quarter, Nvidia projected revenue of 45 billion plus or minus 2%, compared with analyst view of 45.84 billion. The company expects meaningful decrease in data center revenue from China as a result of the H20 ban. And so this is the interesting takeaway from Ben Thompson's post earnings analysis. So he said he is really focused on the difference between the transition from training build-outs to inference. Because training, we've seen the scaling law there. We kind of know the race to get a build, a 20K cluster to 100K cluster. And that's pretty, it's pretty easy to forecast because the CEO of the foundation model company comes out and says, we're gonna build 100K cluster and you know they're gonna buy that many GPUs, right? the inference is much more consumer demand based. And it's always unclear, these products seem great, but how are they actually going to get installed? How many tokens are they actually going to be generating? And then what is the trade-off within those tokens of reasoning and heavy test time inference workloads versus more lower volume workloads? And so Ben Thompson says, Last year I expressed concern that inference workloads we're not increasing as a proportion of NVIDIA sales. So he says, it's worth noting, though, that while the company reported that 40% of its chips were used for inference, that is the same percentage as the last quarter. So keep that in mind. 40% of NVIDIA chips are used for inference. It's 60% training. I think that's the breakdown when when he's talking about AI specifically. Granted, that is a large increase in absolute terms, by and large matching NVIDIA's and revenue increase. And of course, training large models requires ever more GPUs, but the arrival of meaningful AI workloads will be marked by an increasing percentage of chips being used for inference, assuming of course that inference workloads will be run on NVIDIA chips. And there was always a question of like, maybe you train the model on a big NVIDIA cluster, and then you bake it down so small that it runs on the iPhone locally, or it runs on a CPU or it runs, I mean, unlikely that that would ever happen. But people have been able to distill models down so far that you can inference them on other chips that might be more affordable. But that really hasn't happened because of the CUDA ecosystem. And it feels like in general, most of the inference has stayed with NVIDIA.

And we see that in their earnings. But yeah, now you have the TPU.

And then Meta's Artemis. Yep. And so Amazon has Trainium and Inferentia. So they have a training chip and an inference chip. There's different-- Inferentia. I think it's Inferentia. But both of those are in-house chips, but for most companies, they're training and inferencing on NVIDIA. And so on the latest NVIDIA earnings call, Jensen Huang and Colette Kress, the CEO and CFO of NVIDIA, were adamant that workloads for inference were skyrocketing. And here are a couple of quotes that Ben Thompson highlights. AI workloads have transitioned strongly to inference, and AI factory buildouts are driving significant revenue. This is from Colette Kress. And so I think we were kind of asking the question of AI factories. Is that just some buzzword? Is there a distinction there? And what I'm hearing there is the start of AI factory means more data center that's more-- Producing intelligence. Exactly. It's more inference focused data center design, which I think there are specific tradeoffs that you would need, because you don't necessarily need to run the same code over the entire data center. You can shard it up a little bit more because one inference job can be done on the single rack. And so maybe linking all the racks together is not as important potentially. I don't know enough about how training- you- were breaking down.

By the way, Amazon's in-house AI chips are called AWS Trainium and AWS Inferentia.

Let's go. Let's hear it from my memory. Yeah.

Still got it. Golden retriever mode.

Yeah. Colette Kress went on to say, We are witnessing a sharp jump in inference demand. OpenAI, Microsoft and Google are seeing step function leap in token generation, which we saw from that Satya Nadella quote. Microsoft processed over 100 trillion tokens in Q1, a five-fold increase on a year-over-year basis. And so she's now quoting Satya talking about token increases. Wong says, Reasoning, is compute intensive, requires hundreds to thousands of times more tokens per task than previous one-shot inference. Obviously, the difference between 4-0 one-shotting something and deep research, which sits there generating tokens for 20 minutes, wildly different. Reasoning models are driving step function surge in inference demand. AI scaling laws remain firmly intact, not only for training, but now inference too requires massive scale compute. Wang goes on and says, The first positive surprise is the step function demand in increase in reasoning AI. I think it's fairly clear now that AI is going through an exponential growth phase and reasoning AI really busted through concerns about hallucination or its ability to really solve problems. And I think that a lot of people are crossing that barrier and realizing how incredibly effective agentic AI is and reasoning AI is. So the number one, so number one is inference reasoning. and the exponential growth there, demand growth. And so Ben goes on to say, this is all great news for NVIDIA specifically and AI generally. But what was missing was an update on that percentage of chips used number. It certainly sounds like it's more than 40%, but it would have been nice to receive concrete information that that that number has in fact increased. And so I don't know how we get that information. We were talking to Dylan Patel on Friday. We can kind of see if he has a way or a methodology to try and figure that out. But that does seem like a very, very interesting question to ask. It might just be too soon to really be shifting, but maybe the plan for the next data center and the order hasn't hit Nvidia yet will be more inference focused and we could see a boom in-- and we could see a shift in that number. And maybe it just hasn't shifted yet. in Q1 revenue, but it's totally possible that Jensen has visibility into Q2, Q3, Q4 orders and sees that demand is going to go crazy for inference specifically. We're going to be looking at 10% training, 90% inference, but he can't say that yet because the orders haven't actually been placed and he's only looking backwards. So he's trying to kind of like, you know, message that inference is picking up. You can see a bunch of public data points from Satya talking about this. but nothing to report on the actual order side just yet. And so he doesn't want to send the message that it's not moving because then people will be like, well, maybe it'll never move. So I don't know. That's my reading on that. Yeah. Anyway. Let's go into Nvidia's China argument. So the upside surprise for investors is that Nvidia is growing just fine without selling to China. Although, again, although Ben Thompson says, I'm sure, I'm not sure this is a surprise.

Nvidia's sales have not been gated by.

Demand, but by supply for a long time.

Are you making this exact statement last year.

I got it from Alex Wang.

Yeah. Why do we need to sell to China at all when there is effectively.

Larry Ellison and Elon Musk are getting dinner with Jensen saying we will buy as many GPUs as you can produce. Like they are good for the deal. Like you can just sell to them. But of course, if you're a global company, like there isn't there is a shareholder responsibility to have some diversity of revenue. And then there's also that interesting argument of is there value to- It has to be dependent on American AI. Yeah, but yeah, I mean, I mean, the hypocritical thing here is that we, you know, I'm sure we have said, hey, you know, it's not fair that TikTok operates here while Facebook operate, while Facebook can't operate there. Well, now what would we be talking about right now if Facebook was actually in China operating? Would we be like they shouldn't be? Like, it's a powerful tool, it's beneficial. Google is there, that's a beneficial tool. Why are we helping them? So like, we're in this weird dynamic where you can't say both, right? You can't say it's wrong that companies are banned in China, American companies are banned in China, but also the ones that aren't banned should like self-ban, when in fact most American companies for a long time have. But this is all part of like the shifting narrative and the shifting geopolitics. equation. So it is unclear.

Ben goes into the timeline. Somebody's quoting a post from Financial Times. Jensen saying four years ago, Nvidia had 95% market share in China. Today it is only 50%. The CEO said criticizing the curbs for spurring rivals such as Huawei to develop their own AI products. Balding's world says this has been the story of every CEO that has gone to China. How do these I'm not going to say that. How do these guys not get this story yet? I won't even say. I won't call a big tech CEO an idiot. Yeah, it's just respectful.

I can't do it.

Anyway, so Ben says, what I believe Balding is referring to is a pattern in which a multinational moves manufacturing to China at the cost of a joint venture or technology transfer, only to see their market share competed away by Chinese competitors benefiting from their technology. This is, of course, the exact exact opposite of what is happening to Nvidia. The company's market share in China is falling because of the US government's export controls, not because of Chinese government policy. Is that- what? Is that fully true?

I think so.

You don't think competition is factoring in there at all?

No, no, no. Because if you look at the cost per flop, Huawei Ascend is more expensive on a per-cycle basis. if it was a truly free market, you wouldn't see nearly as much demand for that. And also in the same way that demand for GPUs, but in America, it's uncapped in China. They want to build a 100,000 cluster right now. They would if they could. They have the money.

In a supply-constrained market, that would naturally help other upstarts get market share, even if they're selling an inferior product, but if customers can't access But you'd rather have a Huawei GPU than no GPUs at all, right? Yeah, yeah, yeah, totally. Moreover, unlike the vast majority of other product categories, one would in fact expect NVIDIA to maintain its market share if it were able to compete. The difference, of course, is software. NVIDIA chips are not just about GPUs, but about the CUDA software layer, which has network effects. NVIDIA not only has the best performing chips on the market, matched at times in pure performance by AMD, they also by far have the largest ecosystem of software and developers, which makes them the first choice for everyone if you can buy them. Huang's argument is that not only does NVIDIA want to sell to China, but the US should want him to sell to China, too, from the call.

This is a tough argument to make, but, you know, let's hear him out. He says China's AI moves on with or without US chips. It has to compete to train and deploy advanced models. The question is whether China will have AI. It already does. The question is not whether China or whether or not China will have AI. It already does. The question is whether one of the world's largest AI markets will run on American platforms. Shielding Chinese chip makers from U.S. competition only strengthens them abroad and weakens America's position. Export restrictions have spurred China's innovation and scale. The AI race is not just about chips. It's about which stack the world runs on. As that stack grows to include 6G and quantum U.S. global infrastructure leadership is at stake. And yeah, it's interesting like Is like the control over Facebook was so strong that because there's network effect, but also all of the data is run on Facebook servers, that Facebook really has an insane amount of control. When you hand over a GPU, obviously there are some value to the ecosystem and the lock-in, but you can kind of do whatever you want with it because it's hardware. and you're not maintaining the software and gating access. So there is something different there. But I do take his point. And so Ben Thompson goes on to say, it seems clearly in America's interest for China to have a dependency on a US software platform. Then again, it also seems to be in America's interest to have China dependent on Taiwan for chips. But that argument is not winning the day either. this article goes on and it's great and you should subscribe to TechCrunch to get the full thing. Anyway, we should go to this Bloomberg piece that Ben highlights a little bit. NVIDIA's open ecosystem to rival chip makers is in global push.

Opening AI ecosystem to these chip makers.

Yeah.

So NVIDIA Corp. Chief Executive Officer Jensen Huang outlined plans to let customers deploy rival chips and data centers built around its technology, a move that acknowledges the growth of in-house semiconductor development by making major clients such as Microsoft and Amazon. On Monday, Wong kicked off Computex in Taiwan, Asia's biggest electronics forum, dedicating much of his nearly two-hour presentation to celebrating.

The work of local supply chain partners. He's gone for a two-hour presentation. Oh, it sounds crazy. The fact that he's going and opening up this conference and giving a two-hour presentation.

It's pretty awesome.

But then I remember we come up.

Here and do three hours a day.

Every day.

Well, he should come do the show with us sometime. Do the whole show.

Let's do the whole show. Imagine ripping some timeline with Jensen. We'll make it happen.

It'll be electric. But his key announcement was the new NV Link Fusion system that allows the building of more customized artificial intelligence infrastructure, combining NVIDIA's high-speed links with semiconductors from other providers for the first time. And so this is a little bit of what we're seeing where There's now chip startups that are building chips for specific use cases. So an example would be like etch, they're baking the transformer architecture down onto silicon, but you could even look at Apple silicon as something that might benefit from the NV Link fusion system potentially. I don't know if it's actually- people.

Are saying, hey, we know you hyperscalers are building your own chips, Yep. We still want you to be dependent.

On our software, even if you're going to build your own. Ecosystem. Yep. And so to date, NVIDIA has only offered complete computer systems built with its own components. So you buy all the chips and all the racks and everything wires together very nicely, and that is very important for data interchange, especially while you're training. This change gives data center customers more flexibility and allows a measure of competition while still keeping NVIDIA technology at the center. NVLink Fusion products will provide customers the option to use their own central processing units with NVIDIA's AI chips. Their CPUs now pair with NVIDIA's GPUs or pair NVIDIA silicon with another company's AI accelerator. So you can kind of do the inverse and bring up a specific accelerator for a specific use case. Santa Clara, California-based NVIDIA, they love to throw in the location of these companies. It's very funny. Oh, yeah.

Everybody associates Nvidia with Santa Clara.

I guess. They're trying to shore up its place at the heart of the AI boom and keep the surge going in the face of concern that spending on data centers isn't sustainable. The company relies heavily on a select group of giant cloud providers known as hyperscalers for much of the revenue. And all of those hyperscalers are, of course, thinking about building their own chips. Because if they're going to be ordering so many, the margins are so high, why not just go to TSMC directly? And so this bit of news, this is back to Ben Thompson, this bit of news from Computex two weeks ago, is a reminder that CUDA isn't Nvidia's only moat. The company is also dominant in networking, particularly the ability to link multiple chips together so that they function as a single large chip. The key to this is so-called scale-up capabilities, NV Link. And so, Nvidia will license its chip-to-chip C2C technology to enable custom configurations. For example, Grace Hopper was an Nvidia ARM CPU and an Nvidia Hopper GPU. Oh, I didn't realize that. Grace Blackwell is an Nvidia ARM CPU and two Nvidia Blackwell GPUs. If you don't want that configuration, you're out of luck, or at least you were, because now you can license Nvidia's C2C technology. License seems important there. to make any number of combinations. And it doesn't have to be Nvidia chips that are being combined, although Nvidia is obviously confident that you'll choose their GPUs because, yes, they are the best in the business. Fascinating. Anyway, we should move on to this Nike story in the information because there's a very, very interesting chart again from Ben Thompson that I thought everyone should be thinking about in any of their business. So the big news is this is from the information in May 2020, May 21st. So it's a little bit of an older article, but Nike announced that they are going to sell on Amazon for the first time since 2019. So it's been six years off.

So six years ago-- They went all in on D2C. They really got the D2C bug.

They did.

They said they read the thought pieces.

They read the newsletters, they cut out the middleman.

Yeah, they wanted-- they were pissed at the middleman. They tried to cut him out. And now they're making a deal with the middleman.

And so six years after Nike stopped selling through Amazon in favor of going direct to customers through its own stores and websites, the two companies are getting back together. I'm sure it was super rough because tons of people probably went to Amazon every day, typed in Nike and just got Nike knockoffs, right? Because you were just used to buying on Amazon. Amazon plans to start selling Nike products with supplies coming directly from the company, an Amazon One spokesperson confirmed Wednesday, winning back one of the world's best known brands as a victory for Amazon, which has long tried to expand its fashion business by wooing skeptical brands. Nike's deal with Amazon, meanwhile, comes as the sneaker maker is trying to revive slumping sales. So Amazon is going to buy Nike for the first time since 2019. And this is all an effort to revive slumping sales. Nikes revenue is down a ton in the most recent quarter. It fell to 11.3 billion down 9% from the same period a year earlier while wholesale revenue fell 7% to 6.2 billion. And so Ben Thompson calls this Nike's disastrous pivot. He says there's a lot that has been written by Nike struggles over the last few years, including a LinkedIn post by former Nike executive And he outlines three key mistakes here. The elimination of categories like running, basketball, et cetera. A massive de-prioritization of wholesale in favor of direct to consumer sales. And a shift away from brand marketing to direct response marketing.

It's funny, trying to avoid middlemen in business is rough. It's like, oh, we're gonna have our own retail stores. And it's like, okay, well, you have a landlord. So he's your middle man now.

And so, yeah, the middle man is selling online.

Okay, welcome, meet Mark Zuckerberg. He's gonna be your new middle man. It's like trying to avoid the middle man is very difficult, right?

It's interesting because the middle man analogy has actually worked just fine. And Ben Thompson has this cool chart of the CPG value chain and shows that Procter and Gamble literally does not cut out the middleman. In fact, they leverage the middleman and they're doing fine. So what does Procter and Gamble do? They do research and development and they do manufacturing, but they don't handle the middleman. So they don't handle logistics or retail, but they do dominate shelf space with trade marketing and then marketing, like brand marketing. And so CPG companies like P&G harvested most of the value by integrating research and development, manufacturing, marketing and shelf space. Raw materials, retail and logistics were modularized and commoditized. D2C companies meanwhile saw research and development as increasingly unnecessary in overserved markets. As I noted in the context of Dollar Shave Club, razors are a particularly salient example of overserving. Shelf space on the internet was effectively infinite. Their goal was to integrate marketing, retail and manufacturing. And so, The theoretical, if you scroll down, the theoretical D2C value chain is no R&D, just manufacturing, no logistics, but yes, retail. You don't have shelf space, but you do the marketing. And so Ben Thompson says, the problem though, is that marketing on the internet was entirely different than the analog marketing that previously dominated the CPG industry. They're being good at advertising, whether it be coupons in the Sunday paper or television ads during the evening news, was mostly a matter of the ability to spend which was itself a matter of scale. Digital marketing, though, didn't really work at scale, at least relative to TV. In fact, it only made sense if you could target consumers with advertising and track how it performed. On the one hand, this was a critical factor for making D2C companies viable. The advantage of targeted advertising is that it takes a lot less money relative to TV to reach customers who are actually interested in your product. The problem, though, is that getting Good at targeted advertising requires massive amounts of both research and development to build the capability and inventory across a sufficiently large customer base to make the whole effort worthwhile. In the end, no D2C company was actually good at marketing. They outsourced it to Google and Facebook, which both had the inventory and the capability to spend billions necessary to develop sophisticated targeted advertising. The problem is that in the process of depending on Google and Facebook for marketing, the D2C companies gave up their planned integration in the value chain and the associated profits to Facebook. and Google. And so the actual D2C value chain is they're not a manufacturer, they don't do retail, they don't do logistics, they don't do marketing. Instead, they just do R&D and ad inventory. And so the actual customer is just completely owned by Facebook. And that's why you see a lot of D2C companies that have really, really struggled. So the actual integrated players, Google and Facebook, integrate customers and research and development to dominate marketing. D2C may have online retail operations, but that is a modularized and thus commoditized part of the value chain. And, oh yeah, he calls that brandless, the D2C kind of like roll up that had a whole bunch of products. And he said, here's the problem for D2C companies. Facebook is really better at finding them customers than anyone else. That means the best return on investment for acquiring customers is on Facebook, where D2C companies are competing against all the other D2C companies and the mobile game developers and the incumbent CPG companies and everyone else for user attention. that means the real winner is Facebook, while D2C companies are slowly choked by ever-increasing customer acquisition costs. Facebook is the company that makes the space work, and so it's only natural that Facebook is harvesting most of the profits.

I think there's a dynamic where Meta is incentivized to make customer acquisition, like basically extract as much of the profit from the ecosystem as possible without putting the companies out of business.

Right.

And so it's this infinite dance.

Yeah. I mean, that's actually a very simple equation. It's take all the net profits, but not any further. So that the company can continue to go, but the equity value is effectively zero.

And this is why when revenue from Ridge was on the show one or two times ago, we were talking about how historically I've seen D2C brands benefit from net new platforms. like a lot of brands exploded from cheap advertising on TikTok, right? If, you know, it seems like LLMs are an interesting new acquisition channel, still relatively small channel for a lot of brands, but it's possible when ads hit LLMs that they're cheaper and that that could drive, you know, a period of where brands say, Hey, let's like shift a bunch of spend over here because it's, you know, really efficient, but who knows? Facebook is still the best marketing channel in the history of the world. It's the second best, in fact, the.

Number one, obviously ad quick, ad quick, out of home advertising made easy and measurable. So the headaches of out of home advertising, only ad quick combines technology, out of home expertise and data to enable efficient seamless ad buying across the globe. But seriously, Nike is the king of billboards. Yeah, they really are. I mean, they used to be, they used to be, they used to be, and the billboards were fantastic. And everyone remembers a Nike billboard. No one remembers a Nike direct response ad on Instagram. Yeah. Like, I'm sure people clicked on them.

And I'm sure they drove some sales, but the economic equation just didn't quite work. Yeah, well, we walked by some billboards today that are funny. There is a huge billboard campaign across LA called Just Blood.

Oh, yeah.

Talking about, basically it's a campaign to prove that Elizabeth Holmes is innocent. So there are effectively, I think there's got to be 20 billboards up around LA, so I'm interested to follow along with this campaign. If you go to justblood.com, it's quite a lot of AI generated assets. And it's kind of an angry website, but I'm interested to dive deeper there. Going back to some of Nike's errors, They deprioritized wholesale, and then the big one was a shift away from brand marketing to direct response marketing. So when people think about why they love Nike, maybe sometimes it's about the product, but a big part of it is how the Nike brand has made them feel over time. And it's really hard to make people emotional through direct response advertising, because direct response advertising is like problem, solution. It's not about the emotion. And when I think about the moments in my life where I've paid attention to Nike and made to feel anything by Nike, it was these sort of bigger brand campaigns. And brand marketing does provide a halo effect that can reduce overall customer acquisition costs because you see some Nike campaign that inspires you or makes you emotional. and then you see a direct response ad later, you're more inclined to actually click through. And so big miss by John Donahoe to just sort of not realize that so much of Nike's dominance was around the brand. The other thing is like, you know, you look at these upstart players in apparel, things like Vory, Alo, Lululemon, they're all, you know, we talked about this on the show before, Nike, they missed wellness, right? They missed this trend that a huge amount of the population, people, you know, consumers with a lot of discretionary income stopped seeing themselves as athletes and started seeing themselves as, you know, yogis, people that meditate, people that eat healthy but don't necessarily see themselves as I'm an athlete. So big miss that I believe they're still trying to actually get tapped into.

Yeah, well it seems like they're pivoting back to the more traditional strategy and probably we'll see some refocusing on brand marketing because they'll just be everywhere and they won't try and cut out the middleman anymore. They'll just be everywhere and hopefully we'll see some amazing Nike ads coming up soon. Anyway, we have our first guest of the show, Doug Philippone from Snowboard Ventures joining us.

We'll let you take the intro.

Perfect.

Be right back.

Doug, how are you doing today? Let's bring him in here. I'll give a little bit of background on Doug in the meantime. He has a degree from West Point, spent 17 years in the military, and then another 17 years at Palantir building the defense business from nothing into a global defense lead. So we're excited to talk to him about that. And it'll be very interesting to cover the Ukraine drone attack, which we talked a little bit about yesterday. But there's a lot more that we can dig into there. So, Doug, how are you doing today?

Great.

Thanks for having me.

Thanks so much for joining. I'd love to kick it off with just a little bit of your background first, and then we can go into the latest news in Ukraine.

Yeah, I know that's great.

Yeah.

So, well, I've been on a life journey. I've done a lot. Everything has been hard. I think about everything that I've done in my life. People have told me I couldn't do it, and I sort of thrive in that world.

That's awesome.

That's always the case, right?

Yeah.

Well, look, nothing. Nothing worth doing is, you know, is easy.

Yeah.

No, I think when I was a young man, I was just. I went to college for just a minute, and then I was like, this is stupid. I'm in the hamster wheel. And I said I wanted to join the Army. I went into the Rangers as a list a guy, and then, you know, got into West Point a little bit later. while I was there, I studied math. All my friends made fun of me. They're like, what are you gonna do with math?

And.

I definitely got the last laugh there.

That's amazing. Were they just saying like computer nerd type jokes or?

No, it was just like, I mean, basically you gotta remember I'm a little bit older, but in the 90s it was like, it was before the modern internet era and so forth and so on. you know, computer science was like a hobby more than a real thing. You know, the internet, you know, was just coming out while I was in school. And so it was, it was just like the idea of, like, you're going to be a high school math teacher, and that's like the, that's the end of what math is useful for. And so, look at my focus was, I'm gonna go be an army officer. I'm gonna go back in the Rangers, and that's going to be my career. But, you know, I did six tours in the war and got hurt.

Wow.

And so. At that point, I went back to grad school and then found a job in the very early days of Palantir. And literally, the only reason I got hired was they were trying to start their defense business and they sort of thought that a terrorist hunter mathematician was a really cool thing. I sort of found the perfect first job.

Walk me through how you actually met the Palantir folks. What was the first day on the job, like who was around the table? Was Sean Sankar there? Yeah, that was fascinating.

So back in the day, it used to be carp, you know, it's like maybe 50 people there, you know, they really had a product. But the interesting thing, so one of my very best childhood friends, Dan Serravalle, he was one of the best engineers at Palantir, period. He was like a alternative student, I say alternative, he's like older getting his bachelors when he was like 32 or something.

Yeah. Yeah.

And he was, he was interviewing at Facebook and Palantir. He's like, Doug, you got to see this, you gotta come over. And, and then he had just joined. And so when I saw, like, the first demo, even though it was all Intel based, it was very clear. I just went back to my time as a commander in the Rangers. It was just, I was just, I was convinced that if I had this, I could have done everything better. And I, and I really was. you know, a lot of my friends were still going back to the war and getting wounded or killed. And so it was just, for me, it was about purpose. You know, I think you probably, you probably know a lot of palantir people, but we're all, like, in this, you know, strange family that is palantir. Everyone is uniquely driven for a sense of purpose. And for me, it was like, I wanted to do something where I could give back. And if I, you know, if I'm honest, if I would have stayed in versus, obviously, I was hurt, so it didn't matter, but. I have easily contributed a thousand times what I would have if I just.

Stayed in.

Towards the mission.

So I feel quite proud of that.

So while you were in the service and you see what Palantir can do, could you give us a more concrete example of what you're thinking? I mean, I often give, there's a little bit of a meme online, like what does Palantir actually do? And the example that I often give is like, Think about if there's IEDs all over the place, some of those IEDs are going to have C4, some of them are going to have nails, and if you put those all on a map, you can start to see clusters of trends just on a map, and then you can decide, okay, there's a bomb maker in this town, we should go and investigate that. That's the most concrete example I've been able to come up with that I've read just from the literature, but as.

Someone who was there, what was it like? I mean, that's definitely like that would be the 2010 version. and just sort of like, at the core, it's like they make a software architecture that helps combine a myriad of data systems. So like all these legacy systems, if you think of the defense world, all these different areas, whether it's commercial businesses or the military, the larger they are, the more that they all look the same. The military might be the most complex, but it's all similar. in the sense that you have a myriad of legacy technology that's built and modeled around information capabilities at a certain time. And then over the course of history, it's hard to get rid of those old systems. So if you look at the IT infrastructure of the Department of Defense, it's like you have old mainframes on one end that's modeled certain things on the edge, and then you have the most exquisite AI sensor that's doing real-time data and computer vision on the other end. and as a commander, what you're trying to do is assimilate the real world in real time. So it's like you're looking at a map, you're looking at all of your feeds from either videos or satellite footage to understand what's going on, what's changing, and then if you're ever going to go, like to understand what's going on in the real world. So you're understanding things and you're making decisions based off that data that you might have to go without Palantir, you have to go a hundred different places. So that was like the existing paradigm that we were replacing is, How do you, if you, an old school analyst or an operator would have to literally go individually search 100 different systems and use an individual tool for each one of those systems that may or may not be good. And then they would put all that stuff in their mind and they write it back on PowerPoint. And then that's how, and you'd have no legacy search and discovery. You would be violating all of the handling of the data. You lose all the security of it. And it's just like on a flat file and you're emailing around. I can't tell you how long that was the paradigm of the military and most corporations.

Can you talk about the practical challenges of that? There's this concept of the fog of war, and I imagine you experienced this on your different tours where you have a bunch of data, but parsing through it and fully understanding, and I can imagine in many ways Palantir, if you can more quickly and accurately process information from more sources, you can reduce that sense of sort of unknowing that seems to be historically common across every different type of battlefield.

Yeah, I know. I mean, the short answer is like, it will save your life and you can have a successful mission or not. I mean, the historic vignettes that are like pretty powerful for regular people to understand is like, as the military would plan a mission to go in and out. Forget about the real time stuff that is obviously very relevant or just knowing what's going on and how strategic things are moving around and if someone's attacking you, you have to have that side of it, which Palantir does a lot of. But from a historic perspective, once you're planning a mission, to be able to understand the full history, if you're staring at a piece of dirt or you have like, Hey, we think the bad guys are at this compound or there's a military, a legitimate military target that you're going to strike to understand the whole corpus of data that you may know about that for the last 10 years. So to give an example of like, you know, the failures of the Afghan war or Iraq for that matter, just using the American experience, you can do this just about any war is that, you know, as units would go in and out of a region, you would have these like learning experiences that would last three to six to 12 months. And then they would, at best, they would put all the information that they learned, whether it's like, say, say I went on a mission, I interviewed somebody and I took fingerprints, I took pictures, I would write a little note that say, here's the human terrain of this area. This is what the people care about. Here are the projects. Here's where we did a mission. You know, they usually do an after action mission report about it. and you put on a hard drive and then the new commander would come in and you handed to them and then they would probably look at it for two seconds and then, I mean, that's literally how it was done for 20 years. And you're like, how can this, with all of our might and all of our superiority in every way and the money that we're throwing at this, why does it suck so bad? Anyway, so that was, you know, Palantir's mission was to like, how do we change that paradigm? So that, you know, the interesting thing, if I give a, one of the compelling things, we did this with the British you know, with, with, I don't even know if we want to say this, but think of their most sexy unit, the Brits.

Yeah.

And they were, they were going into a specific area. And what they didn't realize over the past five years, they had, every time they flew a helicopter into a region, you know, they have this hosted analysts that are like, they work with helicopter pilots and they do menstruated Graphics. So they understand the exact perfect helicopter landing zone for the mission. And what they didn't realize was that each new team would come in and they would always do their expert analysis and then always pick the same exact place. And so then all of a sudden, when we integrated all this data for them, they saw for the first time, holy cow, every single time for the last five years that we've gone into this valley, we always landed in the same place. So it ends up being like a very easy low tech thing for the enemy to put a bomb at the middle of the HLZ. And they're like, anytime Americans come, just blow up the bomb. But that's like, they call that knowledge management, which is like, that's a little bit old school, but it's like, that was a very unsolved problem forever. And then now, if you get to like real time what we're doing now, it's like, okay, well, now you're taking real time signals intelligence, you're taking real time video feeds where you're overlaying what the video of the surveillance aircraft will be looking at. And then with the entire corpus of information, you would know about the area. So as soon as looking at a piece of ground, you're like 10, 10 reports pop up of everything you know about it. So you immediately have situational awareness that'll help stay, help you stay alive. That's really important.

Can you tell us the story of D-Sigs and the distributed common ground system, like how that played out and like the loss, like at any, at any level of detail that you think is appropriate? Yeah. No, I don't know.

I mean, like, that was my personal Vietnam, except for the fact that I won. So it's like, it's okay.

It's not like I'm using Congratulations.

Yeah, think of a different long war that somebody won and that'll be it. But it was like 10 years of my life. Wow.

10 years.

Yeah, I know it was wild. And I will tell you that there was a lot of times when I just didn't think or win. I literally flew up to New York one time and rode in the car with Dr. Karp and I was like, Sir, do you still want me to do this? You know, because it's like, I don't know if there's any amount of hard work or expertise that is going to.

Allow us to win.

And he had this great moment of leadership where it was like, If any soldiers are still asking, this is maybe 2012, I forget exactly. But he was like, if any soldiers still want this, it's worth fighting for.

It's too important.

And it was just that moment of leadership that I needed. So I got out of the car and then just a few, maybe the next week, I had some massive win on Capitol Hill. But it was, I mean, look, I credit that. I mean, it's the reason I started Snowpoint. like the modern defense tech landscape exists because Palantir was able to break that paradigm. And if you think of the new executive orders, really focusing on acquisition reform. So it was like an old 90s thing. The lawsuit was about this, but there's an old 90s, it was called the Federal Acquisition Streamlining Act, I think it's 94. But it's during the Clinton days, I don't know how old you guys are, but it was basically, there were these scandals about the $400 toilet seat and the $200 hammers.

That sounds cheap by modern standards. I feel like every day I see.

A- We have a soap dispenser.

There was a soap dispenser that was like $50,000.

No, but it's the same BS. So the interesting thing is, so before 1994, and I might be screwing this year up by a year or two, but before then, there was this old term called mil-spec.

Yeah.

And the Mil-Spec idea was that the preference of the US government was to buy custom things as the priority. And so all of those scandals turned it on its head and created this idea where you have a commercial item preference. And so, you know, this is sort of like the, when I talk about the invention or say that, you know, the start of the modern defense tech era, it was the winning of the lawsuit, which I think was 2016 or so. It was the first time since a lot of them passed and it actually been enforced. So our lawsuit was like the precedent setting thing. And all we were finding out, they have like, there's protests all the time, like almost every major contract gets protested. So that part of it wasn't unique. What was part of that was unique about this was we weren't, it was called a pre-award protest. And so normally a protest is like somebody else wins and then Lockheed gets pissed at Raytheon and then they just argue about who's better. and then the court decides. Our protest was that we thought we were getting screwed by the Army, which we were, where they specifically wrote the contract where I had put so much pressure on the Army to expose all the flaws of D6. D6 was simultaneously failing while they spent billions of dollars on it, while over half of the Army's brigade commanders were requesting urgently Palantir. So there's this huge battle where, you know, it was like units would request Palantir, the army would block it, I would go to Congress, expose it, and then there was just this back and forth. And I was like enemy number one for years. And this is where all my war fighting skills came into play. It was like, in many ways, fighting bureaucrats was just as hard as fighting terrorists. And so I had that, like, it.

Was good that I had- I'm gonna.

Call you a boardroom general, right?

Yeah.

Yeah, I was on the front lines, though. But anyways, yeah, so in this particular case, so the lawsuit, it was interesting because this is another experience of like, you know, between Peter Thiel and Carpe, it was like they trusted me. I couldn't have done it. Like, it's very unique in the Palantir world that like, they would that everyone said we were going to lose, period. And I think like statistically, they weren't wrong. I think the chances of winning these cases are like 3%. So it was like clear. And then if we want it, so basically this is, so everybody was like, you guys are crazy. And they were like, Doug, just do this. And, but it was like we were uniquely driven in the sense that we had the moral high ground, we had all the evidence. It was the perfect setup. But anyway, long story short is we did end up winning and then we won the appeal three to zero. And so it sort of set up this thing where, you know, at the end of the day, when we won, we didn't get any money. It was like basically we just burned 3 million, 3 to 5 million bucks in the hallway on legal bills. But then we had to go compete for the original D6 contract. So it wasn't sure, like I was halfway terrified we went through all this trouble and then the army would blackball us, which could have happened. But long story short is that was sort of the real legitimate beginning where it was no longer a guerrilla campaign for Palantir, where we were just winning enterprise contracts Yeah. On the run and then on the merits. And it's just, you know, the software is 10 times better. I mean, it's just, it's so impressive to watch. I mean, obviously I'm officially retired, but it's like, I'm in the mafia for life.

Yeah. Yeah. So can you talk about like the long term build out of the defense business at Palantir, where it is now, and then where you're seeing opportunities in defense tech, whether on top of Palantir with their AIP program or just generally outside of it just seems like the army is modernizing and the DoD is modernizing generally. So companies won't need to go through the same 17-year journey necessarily.

No, I mean, look, that's the whole reason I started Snowpoint. Was the idea was like, you know, during my journey, say I did 100 things right and 1,000 things wrong. It's like, how can I help other companies accelerate through that cycle and, you know, ultimately help our whole society? It's like having everyone succeed, including for the services. it's a win for everybody, long story short. But I think it's interesting, because during early years, I got the seed corn in across like SOCOM and JSOC, and then a few initial Army units, and then that was kind of growing. But we were still searching for the elusive program record. And then we went across, there's like 17 different countries that we kind of got started in the military business. I was flying on airplanes everywhere. and we got the C cord done. And then there's like, NATO is like, it was kind of the inside joke of like, that's never going anywhere. And then now, you know, of course, this is 18 years later. It's like we finally, you know, won this huge contract with NATO. I'm like, you know, it's taken forever, but it's, it's, it's been fascinating to watch. And I think that, like, what I determined, you know, again, this is rewinding a little bit, was like, we have to win in the US. And then once we went in the US, it was sort of like this, the dominoes started falling where, and this is a deep history with Palantir, and I tell a lot of my portfolio companies the same thing is this idea of like exit in crises is that, you know, many different companies with a lot of different technology, well, they all look similar on paper. And so it's like, how do you prove it? And one of the like defining things that I believe about Palantir in particular, but a lot of my, I hold my portfolio companies to the same standard is this idea that when there is a moment, an exigent moment where it really matters. And so in the military context, that means that like a commander is doing something, people are dying, you have to win. Then at that point, all the bullshit goes away and you have to have something that actually works that's real. And that is something unique in the sense that like Palantir has never screwed that up. and it was just sort of like knowing that it's like you gotta, obviously it took us forever, but it's like we're past that, we're way past that, you know, say by a couple of years, but past a momentum point where, you know, you were always fighting, like the biggest barrier to entry was like some loser CIO or CTO inside a corporation or the government saying, Don't worry about this, I have this PowerPoint slide. and I can build this cheaper, faster, better. And then all the incentives are wrong. And then they would slow roll you and you wouldn't make anywhere. And so the challenge where we would always say is that the open door was around who cares? And are you actually talking to the profit loss owner, like where it actually matters, where the bottom line matters. And in the military, that was like the operational commander. Like if you got buried in Pentagon where they're talking about programs, you can't walk four feet without running into, like, a 300 million dollar disaster. They're all in denial about it. And it's like, because it, like, either tests out on paper or they're like, it's coming. It's gonna be here. It's gonna be so awesome. Just don't buy Palantir. It's too expensive. We're gonna do something. You know, I I mean, dude, don't even get me started.

We're getting you started. Let's flip to something much more concrete and timely. what's going on in Ukraine? What's been your reaction? And what is the solution or impact to the defense tech ecosystem?

Yeah, you know, it's interesting. I traveled to Ukraine in the very early, when Palantir was just getting started there in the very early times of the war. I went to Kyiv just to see what was true and what wasn't true is we got kind of started. And, you know, it's fascinating to think there's a deeper political thing and there's obviously a military and technology thing about this. But it also applies to NATO and the defense tech market in Europe versus the United States. And what the hell is Europe going to do now that they have to do something real and stand up their own capabilities? I think this is a fascinating time of accountability. And I think it's good for the world. I do believe that the world is better when the US is the hegemon. I believe that deeply. It's like we can argue about whether or not we can afford it. But I'm like a firm believer in terms of like what I'm doing in the defense tech market. But it's also good for Palantir, all the companies, it's like with less budgets, I mean, I suppose it's always good with more budgets, but it's like with less budgets or trimmed budgets, it's like if you need something real, and you have like a short timeline, that's the perfect environment. There's no free chicken. Like the entrepreneurs still have to actually deliver on what they said they're going to do. They have to deliver. But that's the time when you can really succeed. Because when it's like you're kind of doing nothing and you're just like talking about these like five year programs, the tendency is just to waste money. So going back to the Ukraine example, you know, in many ways, like Zelensky single-handedly saved NATO, you know, because it's like Biden's plan was to get him on a helicopter and get out of there. Like it was just on the heels of the failures in Afghanistan. And then it was like, it was clear, like enemies of the United States, this is your time, screw with America as much. Do whatever you want. Biden's not going to do anything. So it was a very, so it was a very tricky time. The fact that he decided to stay and fight, you know, was super impressive and definitely saved NATO. and you have to think also initially, and this kind of goes to like the deeper problem of like what's the resolution of this and what can we learn from it? And I'll try to go faster, but think of this initially is Ukraine stopped Russia by themselves in the beginning.

Yeah.

That's something to like really pay attention to. And there was a few scenes early on because I used to model this of like China versus Russia's like the Russia was the real military. and China is like this paper tiger. And I just kind of, like, flip-flopped all that in my head now. But it's like Russia could not have screwed it up more than they did. And, you know, there was those scenes of, like, I was like, God, if we only had a squadron of a-10s on board, we could have, like, you know, I don't know. You had this, like, Trail of Tears of, like, it was literally just, like, the first Iraq War where it was like a, you know, a 10 or 20 mile Convoy of, like, every military vehicle in Russia was stuck on one road, and I like, please, baby Jesus, where is an A-10? That's all we need right now. But look, there's a very deep asymmetry with Russia versus Ukraine. I think the truth is, you know, I don't have a good filter here, but it's like the truth is that both armies suck, both military suck. and they've devolved into some sort of World War One BS thing. If you look at the numbers, it's like, you know, in terms of pushing back, pushing back.

Yeah, this drone attack seemed extremely sophisticated. This did not seem like it's evolved.

You're not wrong. You're not wrong. But that is, it's a brilliant move. Will it have any impact on the course of the outcome of the war? un, untold. It, does it also mess up the, the peace? Oh, peace talks.

Yeah. Yeah. It's sort of like, deep in, in Russian territory. It could be seen as aggressive.

Look, you could, you can talk about this like there, it is a, it is a massive and incredibly loudable tactical Victory. It's impressive in every possible way. I, I love it, you know, in a lot of ways, you know, because you sort of think of Again, this is a very asymmetric fight. If you look at people, it's like 1.3 million on the Russian side for active duty, 200,000 on the Ukraine side. But they've just devolved into this front line thing. And so is the deep attack going to actually change the front lines or the course of the war? I'm not convinced of that at all. The challenge here is how do you degrade the ability of of Russia, because they have a huge production problem. They outnumber, if you look at tanks and armor, it's like 17,000. But it's like one of the interesting statistics here is that Russia has like 17,000 tanks, but they've lost, during the war, they've lost 8,000 tanks. The US only has 4,600 tanks. Just to put it in perspective. But it hasn't made Ukraine win. They're still stuck with this sort of like, hey, how do we do... So there's this incredible, like...

But tactically, is this return to World War I style almost trench warfare that we're seeing, is that a byproduct of new technology coming onto the battlefield that is asymmetric and, you know, generally, both armies don't fully understand how to deal with yet.

Yeah.

The interesting thing is, like, you know, if I say to my, you know, military colleagues, it's like there's a lot to learn from what's going on, but don't learn too much because this is a very unique thing in the sense that Russia in particular, but Ukraine as well, have completely failed to be able to execute a joint combined arms fight. And so this is the thing that the United States is uniquely the best in the world at, in a sense where like, what does that mean? It means that you integrate space, air, land, sea, artillery, all into one, like you have a decisive point of the battle, you have a bunch of shaping efforts to the battle, and then you can succeed and you can pour troops into and accomplish something. The US, this is the most complicated thing you can do, and it's super expensive. That's why we have a $800 plus billion dollar budget. And that's what the US is like a thousand times better at than anyone else in the world. And what I can tell you is the Russians, the reason that they lost against the Ukrainians by themselves, it's because they don't know how to do that. They were treating Air Force as like artillery. None of it was combined. They totally failed, like in terms of communications. the ability to communicate and command control, like all of that stuff, total disaster. And here we are four years into it or whatever. It's like, what's the outcome?

I'm curious how, I'm curious what Russia's early stage defense tech market even looks like. I imagine it's a lot of it is government directed, but how are they adapting? Are there-.

Yeah, like the thing about Russia is like, Again, they've been very good at advertising specifically like very low production or one production type things where they're like.

Exquisite systems.

Exquisite systems. And then they have two of them. And then the majority of their stockpiles are Soviet era bullshit that's horrible. and if you think about it like this, think of the Soviet Union back in the day, it's like, and they're still not good at this, is that, or even you go back all the way back to World War II, it's like they've never been good at manufacturing. Like, think of how many cars are made in the Soviet Union, North Russia, right? How many real planes do they make?

Right.

And so, like, all that stuff is like, they can build one or two off, but they're fully reliant on the West for all this stuff. They've never been good at it. What they've been good at is, like, Can we stamp out thousands of really cheap, attritable systems?

And- and do tanks fit into that category?

Are they- yeah, but they're not fighting with them because then you could argue that like because of the tactics, how relevant are they on the modern battlefield? You know, and I think Ukraine with like a ridiculous underdog approach has been able to use cheap drone warfare. To be clear here, I think that Ukraine is probably the best in the world at drones. They have really, at a tactical level, again, it's not changing the outcome, but they're at least holding the status quo.

And there are millions of them. They figure it out.

And they're legitimately good.

Yeah.

I mean, that part is fascinating to watch. I just don't know that the way that we fight, if you were to put the US, if you take away the nuclear component of this, which always makes it tricky, the reason we have nukes is make it tricky, right? So you don't do these things. But I'm just not convinced that we would do it this way at all. And how quickly would we win? That part's fascinating.

How do you think about counter-UAS technology right now? You have to imagine that Russia's thinking about every and any and every response to make sure this never happens again.

What are you seeing? Yeah, and look, I apologize for being too bearish about how much does this matter. The attack is like stunning. It's almost, It's not quite as good as the Hezbollah beepers, but it's pretty close.

It's up there. Okay.

Hezbollah beepers. I did my master's thesis on Hezbollah, so that gave me like, you know, specific joy. But I, you know, the idea that like the uncertainty and kind of a step back that it would take within Russia. So basically now they're searching every single truck, you know, they're like super heightened security. This has caused mass chaos across Russia. Now they can't mass those jets together anymore, so you have to space them out. So this has serious consequences to their strategic capability, let alone their ability to just go bomb Ukraine.

Do you think that going forward, various militaries, Russia in particular, will seek to place strategic bombers and aircraft like that in hangars and provide some type of defensive mechanism? Because the fact that they were just sitting out in the open, We had.

To start from scratch. No, I actually think that there's, I don't know specifically this off to look it up, but there's like, it's part of the, there's a part of a nuclear treaty.

It mostly got rolled back. Russia rolled it back at the beginning of the war, from what I've seen. But you can imagine they would say, hey, we're not going to participate anymore, but they wouldn't necessarily build all the structures necessary to actually adapt.

Yeah.

No, I mean, look, all this stuff takes time, too. I mean, that's the, the biggest thing is, like, you know, for the, a lot of those playing, I mean, first of all, they're gonna have to do that, whether there's a tree that says they have to be in plain sight or not, they're gonna have to do that. I mean, so this is like this, this attack has caused a strategic impact, whether that gets Russia to the, to the table faster or better or it gets land concessions for Ukraine or it gets some, I'm not convinced of that, but it definitely is like a very successful thing. And we also don't know the ramifications like, you know, does this up the ante? Does this get us towards World War III? I don't know. That part scares me. I do want the war to end and it definitely gets in into the Trump, like Trump administration is trying to end the war and neither Putin or Zelensky are like playing ball. It's like that's kind of the disappointing part of this.

Well, I, I mean, we have to have you back on. I feel like we could go all over the world and talk about geopolitics for 4 hours. So expect to get another calendar invite because this is a fantastic conversation. Yeah.

Thank you for jumping on.

I really appreciate it. Thank you both. We'll talk to you soon, Doug. Thanks so much. Bye. And we have our next guest already in the studio, so we'll bring him in. We're going to talk. We're going to recap some of the Middle East AI deals, talk about Geopolitics in AI, some of the chip stuff. I still have yet to figure out what the horse means.

Why I just love horses, John.

I just love horses.

I grew up, I wasn't a horse guy. I have an extreme respect, admiration.

My mom texted me after the show last night. She said she was watching the show and because you were playing the horse sound, her dog was running around looking for a horse. So give the dog some horse. For all the dogs that are listening, you're now on the hunt.

For all the dogs.

Anyway, welcome to the stream. How are you doing? Doing great.

How are you?

Great.

Welcome.

Thank you so much for joining. Would you mind kicking us off with a little bit of introduction on yourself.

For everyone who's listening?

Right.

Yeah.

So, Divyansh Kaushik, I am at Beacon Global Strategies for a National Security Advisor firm. So basically what I do is, you know, work with US industry on how to be better national security partners to the United States.

Great. And I wanted to have you on to kind of look at a retrospective on what happened in the Middle East. We saw Jensen Huang go over there. I think Elon and Sam and Alex Wang and all the big tech CEOs went over. What is your read now that we have a little bit more distance from how those relationships are changing, what we should expect, what the actual trade deal came, how everything netted out basically?

Yeah, I think it's a very interesting thing that happened over there because you look at what the Biden administration did in January. They tried to approach AI diffusion with three key strategic objectives. They're like, we want to prevent offshoring to the Gulf. We want to prevent these chips being smuggled into China. through, which is mostly a Southeast Asia problem. And the third one, like we want to prevent remote access where Chinese entities are just using these jobs through the cloud. Like there are hyperscalers trying to build big data centers for Chinese companies and Malaysia and all. Using this technology as a tool of economic statecraft, I think, was a secondary objective, not primary, whereas I feel like what has happened, what we've seen with the Gulf Trip, is more often upside down. where we're using this technology as a tool of economic statecraft. We're trying to get concessions, like we're going to give you these chips, but we want reciprocal investment in the United States. We want one-to-one investment. And we want certain security conditions. Now, a lot of that is yet to be cleared out, right? We haven't finalized, it seems, per media reporting on what the security conditions are, whether they apply to specific entities or whether they apply to the UAE government or the Saudi government as a whole? Do they get to host Huawei chips or Huawei infrastructure in the same data centers where they're hosting US chips? There are reasonable concerns that some people have raised and those are things obviously I think the security details will probably be figuring out because if you remember Microsoft and G42 did a similar deal. at a company level last year in April. That was accompanied by an intergovernmental assurance agreement, which required G42 to do certain things like not have Chinese hardware in a certain number of years, just restrict access for military or intelligence purposes. That deal was really well drafted. There were a lot of things that people pointed out that could be done on top So I think we'll see how this shapes out in the next coming few weeks, how we see the security guarantees coming together. But overall, I think this is in line with what the president has said, that we'll give you our technology, but what will you give us in return? So that's the theme that I expect to continue.

Part of the deal.

Part of the deal. How important is Llama in this emerging geopolitical narrative around AI? We're hearing that these, I've been calling them like kind of jump ball countries. They're countries that aren't full allies and they kind of could do business with China and that's kind of the geopolitical backdrop for all this. If they don't do a deal with Nvidia, they're probably gonna run Huawei Ascend and Deepsea and Manus on top of that. Is an open source stack important to some of these countries?

Yeah. Now, I think there are two questions in there. First, whether China can even export, which there are actually the data suggests, you know, China has enough dies that TSMC shipped to China, TSMC shipped to Huawei in violation of US export controls, like about 2.9 million dies enough to produce a million 910Cs, about 900 910Bs, Now, you know, that's a lot of GPUs, but they also have a lot of domestic demand. The question becomes, will they be able to meet their domestic demand, let alone export? I don't think that they have the capacity to produce it domestically at scale to export. There was this announcement from Malaysia recently, which the minister walked back afterwards, which made for a splashy headline that Ascend Data Center is coming online in Malaysia. but if you look at the details, it was 3, 000% by 2026.

Yeah, it's nothing.

So they can't meet that.

That's, that's one little asterisk. Yeah.

The second thing to your point.

Yeah. Right.

Because some of these countries have worked with China a lot.

Yeah, of course.

But one of the things of bringing countries into our fold is that we have to work with countries that are in China's world right now.

Yeah.

That is a precursor of bringing them into our fold.

Yeah.

But you have to be realistic. What the risks are and how do we mitigate them? Some of these countries have been doing joint military exercises in Xinjiang of all places or hosting military bases. And Saudi Arabia, I believe like a month or two months ago, their national telecom champion did a strategic announcement with Huawei. And so you have to be realistic and put in place those mitigation agreements. But separately, to your point on open source, I think we should be flooding the zone with American open source stack at the expense of Chinese models. I think we are not paying enough attention to what can happen when a state-controlled company to some extent. There are a lot of PLA ties that Deepsea has been scrubbed from the internet. Now, if the state wanted to, there was a great anthropic paper last year, not sure if people remember, but on sleeper agents that can be converted into models. The Chinese state really wanted to, you know, cripple US infrastructure or get into your phones. Why wouldn't they just put in sleeper agents that would activate in a year or two years.

If it runs into critical infrastructure, it changes the way it reasons.

Exactly. And so we should be pairing ICTS action, right? We're doing a lot of export controls on American tech not going to China. But where's the import actions on preventing this Chinese tech from coming into the United States or helping other countries deploy American tech? Come on, let's be real. Nobody in Kenya is going to pay $200 a month to access advanced AI models. We should be deploying American open source stack there. We should be incentivizing companies with loan programs from the development finance corporation, Exim Bank, whatnot.

Yeah, kind of a new AI, Belt and Road strategy, something like that. Exactly. Makes a ton of sense.

Do you think we'll see something to the to the effect of a Stuxnet within AI in the next 10 years, 15 years?

I'd be surprised if there are not zero-day attacks that have already been exploited.

Yeah.

That we don't know yet.

Yeah, that makes sense.

There's been a lot of talk around NVIDIA's earnings, around the shift from training to inference. And we saw with Deep Seek, some really, really creative engineering around distilling models and switching to, I think, FP8 and floating point and just a whole bunch of optimizations to get a really great result out of less significant hardware. And so I wonder if there's, is there any risk that as pre-training plateaus, we wind up with a scenario where the distillation process continues and inference shifts onto lagging edge semiconductors. And all of a sudden, the gap between TSMC's leading edge three nanometer and China's ability to produce in mass at seven nanometer, even beyond 10 nanometer beyond, that actually changed the dynamic.

Yeah, and I think right now we have the lead there. It's possible that that happens. And the answer I would tell you that many people in the US government would give you is we should not be complacent about it. We should not be providing components to Chinese companies to build semiconductor manufacturing equipment there to produce these semiconductors.

Yeah, so what is your take on the chip ban and this idea that you kind of either need to be all in or all out? It's either don't just worry about the chips, but also worry about the entire supply chain, or is there a creative argument that you could potentially make where there is value to the American semiconductor and I guess Western semiconductor stack ASML, TSMC actually having a benefit in China. It feels very counterintuitive, but I was just remembering a decade ago, everyone in technology was making the argument that Facebook and Google should be allowed to operate in China. And now we're making the argument that the other tech company should not be allowed to operate in China. And I kind of agreed with both of those, but they do seem at odds. Walk me through that argument.

Yeah, I think, look, there are choke point technologies. where we do not want the PRC to gain access. There are multiple arguments for why you should not give them, but the effectiveness really depends on whether it's a choke point technology or not. ASML's machines, when we export controlled extreme ultraviolet lithography, then, well, China has not been able to produce anything. domestically. They are still stuck on seven nanometer dual node patterning, multi patterning. And I think that set them back by seven to ten years at least. So export controls are not about the point that they will never develop that technology. It's about buying time and space so that we can extend our lead. And that is what matters. if we say tomorrow decide further tools should not go to China or other semiconductors should not go or as a tap and EDA software should not be sold. We also need some promote agenda to really extend our lead.

So Huawei Ascend versus Nvidia, is this a quality or quantity battle? What is the shape of differentiation?

I think it's both quality versus yields. They're at about 20% or so when producing the dies for Ascent.

So just cost?

It's extremely inefficient. And so without state subsidies, the latest Huawei server that came out, it is using technology that NVIDIA was using seven years ago.

Wow. Right.

Putting more GPUs on the same server to come up at the same amount of memory and interconnect bandwidth and whatnot. Now, yeah, can they have a state-controlled economy? They can do a lot of this. It's not sustainable at the end of it. So that's one thing. I think there are a couple other things with regards to should we be okay, for instance, if it turns out that many of these technologies, like look, to be real, China is an economy that has an, it is a system of government with an explicit civil military fusion strategy. Every research and development project that happens for civilian purposes is and will be used for military purposes. The vice president of research for ShanghaI said that explicitly. So are we okay with them developing technology that may be used to kill American soldiers or threaten America's allies in the South China Sea or take over the Taiwan Strait? So those are questions that are, you know, they're very nuanced questions and hard to answer, but normally, but in this scenario, I would say perhaps not. Perhaps it's not okay for us to benefit PLA's modernization. So while they can do what it is doing right now, I don't think it can scale. I think the US export controls to much extent are working. The question is how do we extend our lead further?

So reviewing this year so far, do you think we're in a better position than we were last year or are we trending in the correct direction, I guess?

Yeah, I think overall we will, you know, depends on how do you define the strategy here, right?

And this is the strategy to win. I mean, yeah, the goal is to win.

The strategy is to stop their development of the technology altogether. That's not going to happen, right? Okay. Technology races never stay one. The Soviets learned that.

Yep.

So the Chinese have learned that too. But if the idea is to maintain a lead, then they come second, we come first. Yeah, I think we're headed in the right direction so far.

Now, that's great to hear.

China has tried to weaponize, like, to try to mimic what BIS does with export controls through their own export controls. But yeah, their export control system is just like five years old. Their law passed in 2020. The nuts and bolts rules for dual use items, they did not arrive until December of last year. So they're very much in an amateur state over there with regards to this. So the US does have a lead on that. as well.

Very cool. Well, thank you so much for joining. This was fantastic.

Super insightful.

We'll have to have you back and talk more soon.

Thank you very much.

Have a great rest of your day. We'll talk to you soon. And next up, we have Aaron Ginn, the GPU whisperer himself, to take us further into the discussion around AI chip controls, geopolitics, and all of the fun things that go along with that. Jordi, I'll let you take the intro. and I will be right back.

Perfect.

Aaron, welcome to the studio. Play some soundboard.

Let's go. Let's go. The GPU Whisperer is back. It's great to have you. What's happening?

When I come visit you, I'm gonna bring you this.

Wow. Is that a new-- is that for the gong?

No, it's not a gong. No, it's also not a sexual toy. So it's called a talking stick. so in Canadian culture, I got this in Canada, the, the basically a person that would represent the tribe would carry this around. And whenever you'd show about a tribe, you would, you would show them this and they're like, so like, hey, like, I'm, I'm coming to the, the, the TVBN tribe. Who should I talk to?

Right.

And then you find the guy with the talking stick. Got it. Yeah, you find the guy, right. And, and so, yeah, we can, if you represent your sort of head of your tribe, head of your family, you carry that around. and then, and the great thing is like you can, when you're in tribal meetings, if like, you know, someone's arguing, you hold the talking stick, right? And you're like, hey, yeah, yeah, it's like, you know, so we can span two different continents. You have Gong represent my Chinese heritage, and now you have, you know, something from the African continent. So you find something from Europe.

I like that. Maybe we could make it the, Sam Lesson from Slow was on the show talking about the potential of the e-cane. sort of a smart cane device, maybe that could be the American equivalent of a talking stick. It's a smart cane that the leader of an organization carries. Definitely something there. What's on your mind this week? I know you've been busy publishing, also scaling HydraHost. What's going on in your world?

So, video released its numbers. I frankly was surprised they still beat expectations. despite having a lot of regulatory headwinds for at least a quarter. And of course, as you all have covered, those rules changed. And now we're back to October 2023 rules, which still had restrictions, but it was a little bit more friendly to some of the countries that were trying to form deeper relationships with, AKA Europe. So previously, half of Europe was on, you're on the crap list, you're a bad boy. basically anything.

Switzerland.

Yeah, yeah, Switzerland was bad. Austria was bad. Greenland was bad. We can buy the country if we can sell them GPUs. And Mexico as well, which assembles 90% of all of our GPUs. So that's all gone. So a new era has, or to say, returned to the old era, which is 2023. And we have this kind of new thing, which I've talked about previously on the show, which is the sovereign AI with America being deployed into regions. The biggest one was the G42 slash Saudi Arabia to build out GPU capacity within regions. And this will, I think in the next coming years, Americans will see gradually over the course of time how important both this infrastructure is to the world, but also how we ourselves will most likely become a default exporter like Boeing, like a default exporter of this technology, not necessarily like we are always using it, but the rest of the world is using it. and they deeply rely on our infrastructure. That's the trajectory this is heading. And what was surprising from the Nvidia numbers, which I know that John mentioned, was that Jensen has jumped more head first into politics. Under Biden, he was very hands off. He's just like, I don't support this stuff. But now he was meeting Trump in MaR-A-Lago. He's basically proactively advocating both at conferences and in earnings calls, a actual posture about how the world should work and how the world should operate according to, you know, the video preferences. But Americans and Americans don't accept that, like Nvidia is our champion. They are a national champion. If they lose, we lose. Like we're tied together at the hip. And if they go down, we go down with them.

There's not like AMD- It makes sense that he's getting involved with politics when, you know, or geopolitics when Nvidia is valued at the GDP of the United Kingdom?

Yeah, I was thinking about that. It is basically in many ways just a function of the size and scale of that company. Because you don't see the same, yeah, you're not seeing the same behavior from like the other pieces of the stack, like the DRAM supplier or the transformer supplier or the energy to some degree, but certainly not getting as much FaceTime because it's just a less critical piece of the economy.

And some of the important things he said in the earnings call was that NVIDIA lost 50% of its market share in China over the last four years. And he mentioned the acceleration of Huawei from seven nanometer to five nanometer processes. So when we as Americans think about GPUs and export controls, we're kind of battling two different metaphysical complexes that we don't fully accept or understand. We have one, which is generally was part of the Trump train back in 2016. China hawk, they're a threat, they're taking our stuff, you know, they're breaking into Geordie's house, stealing his watches, you know, they're following John, trying to take a couple inches off his height right now. Like, Chinese are everywhere, they're in our cities, they're taking a that was like Trump, Trump 1.0, Trump 45, we gotta be aggressive. And he changed America's view on China. He and him alone, like he was the lone Republican on stage. And in fact, Rubio and Rand were saying that he's saber rattling and the whole stage was like, no, we can engage them. But then there's this new version which has co-opted itself into, which is part of the accelerated ultraist people mixed with anti-trade mixed with protectionism. who took that laundered it and are using it as justification to be China Hawk. And this goes to my most recent piece, which was which we must return all, which is that America has to be an AI exporter under. I'm like OG China Hawk. I'm like, we gotta win with Commerce, the dollar, free trade, sell to many people as possible, like Boeing, as many Boeing airplanes in the world, because it's a point of Leverage, it's a point of. us being able to spread human rights, freedom, Western values, and not have to have bombs attached to it on the airplane, right? It could be passengers or other bombs. And this other wing, which is small but very loud, they've co-opted themselves into the altruistic, you know, basically deceleration people, kind of like protectionism, anti-trade unions, and China hawk people. They're the like abandoned Taiwan, don't sell to anybody. We, if giving it to the Gulf region means giving it to China, which is bizarre, it's a 12 hour flight away. I don't understand that. And it was also insulting to the Saudis. They actually will use it. It's them saying the Jerusalem arm, it's like, it's all this complex kind of vagaries around patriotism. After that earnings call, people called Jensen unpatriotic. Well, you got to put profits you know, above patriotism.

Right.

And it's this really bizarre, like, world that these people live in where they think that, like, well, if they just don't sell them it, there's no harm, right? They can just wait. They can just wait for us when Huawei is like, please do. And, and the, the customers in China that they were part of the information, they did this on the, the publication, the, the customers of gpus in China, they said that, like, by the export controls, they created the market for a while. So we only have two options as a country. We either win or we lose. Like that's it. There's no way in between. There's none of this, like we'll just choke you off, right? It'll just hold you on the inside. You can see this is using an non-technology example is when we've done this with food aid, when we've done this with other semiconductors, like when the first semiconductors came out, we switched it to India and China. All of them produced their own industry. And in fact, there's a famous episode between, I think it was Johnson and Gandhi whenever she was prime minister, when Johnson was like, Hey India, you've got to back us on Vietnam. Otherwise, we're going to gradually send this food aid. There's massive famine going on. And Gandhi, she said publicly, I will never depend on America ever again. And after that, she spent all of her money invested in agricultural improvements that became self-sovereign. So this is not a weird abstraction of, well, well, this is only unique to the situation. When they have the engineering horsepower, they cover 90% of all rare earth mineral production, not ownership, but just production. They build, they have their own boundary program. It's already active. We don't have one. And it's the focal point of all the accessory products of a server, like fans, cables, things like that. And we're like, well, we'll just choke you off and just wait whenever we want to give it to you. It's like, no, like, why would they ever do that? They just build their own. and they show the can. So in that world, the Gulf Stargates, we should have a South America Stargate, we have a European Stargate, we have one in Africa. We have to own footprint, otherwise Huawei will just go out, sell its cheap gear, own the footprint, own this rare, like, you know, data center space is not prolific anymore, own that, and then all the government data in those regions will be running on Huawei stuff, which.

Will hurt our natural security interests. Yeah, what about actually selling to the Chinese directly? Like should we be building a Chinese Stargate? There's this like weird like thread we've been pulling on.

China Gate doesn't have a great relation to it.

But I mean, I know it is different, but I want to kind of unpack all the different ways it's different. Because for years, tech was beating the drum of Google should be able to operate in China. Facebook should be able to operate in China. This would be a net win for the United States if we could operate there. GPUs are different because it's more than the supply chain and there's different things you can do with it. We might be powering adversarial dual-use technology. But is there a world where these two situations are more similar than they are different?

Yeah, the dual-use argument is an argument in favor of ignorance. We don't have an example of what this dual-use could be. and there was a response to Jensen in some article I read somewhere where they could use AI for missile targeting. They don't need AI to do missile targeting. They have perfectly fine missile technology to target.

What.

About just the idea that large language models, chat models, they increase productivity. And so if you give a soldier on a base, ChatGPT, they will be able to inventory the rations faster and that will increase military productivity.

Oh, happens, right? So like one is that by us not entering in the Chinese market, we created a TikTok problem.

Yeah.

We created an Alibaba Rock. We created like, maybe it's because I am.

Unpack that because I feel like we definitely tried to enter the social media market in China. We were blocked and I don't understand how we created that problem.

Well, it's because we wanted to enter on our own terms.

Oh, sure.

Yeah, I'm not saying we have to do business with them always. That's not what I'm saying. I don't want to sell F35s, I don't want to sell F22s. Logical things. When we sell F777, it should logically have our GPS technology or mood, anything that's aligned with our connectivity. Sure, we were doing that today. That's fine. But when it comes to actual core platforms, GPUs is not just the card. because just because, you know, you own a G63 Mercedes doesn't mean I can build one. There's just like misunderstanding of like Jensen is pursuing a three nanometer, two nanometer sort of direction of this company. They don't have the ability to do that. And just because I own a Jeep Wagon doesn't mean I know how to build one. Doesn't mean I can build one. And there's a disconnect between that. So one is that we should restrict anything that is on the supply chain level, I'm totally cool with that. If we want to target foundry equipment, if we want to target data center equipment, totally cool with that.

Problem with that, we don't build any of that stuff. Other countries do. Yeah, but we have American IP, so we actually do have a point of.

Leverage over ASML equipment, if I'm right. I remember that correctly. Yeah, foreign direct rule, right?

Yeah.

So, yeah, so it's a stretch because Japan can still do what it wants. Like, it doesn't have America has limit limit, like the, the regulatory framework. Yeah, the regulatory framework, the Commerce is limited to its ability to, to basically tax authority. So if you have money in America, then they could do something. But if you don't have money in America and you say foreign direct rule in Japan, it does like, like, they'll be like, okay, that's nice. Thank you for your memo. Like it. So I, I, but again, it goes to, like, how we treat GPUs that if we want to create a point of leverage between the Dutch and Japanese over, over founder equipment, we should probably not cut half of the Dutch economy off from GPUs, which is what we did. So if we want to be allied with them, we should realize what's important, what's not important. To me, supply chain production is important, not consumption. Consumption is not very important, mainly because there's another risk factor, which is significantly more important, which is like, is it worse for the world if Huawei is everywhere or is it better for the world that Nvidia is everywhere? and, and when you, we're talking about the dual use question, it's like they don't even think about that. Well, if we, if we don't actually engage them on our own platforms, we finance their current platforms and we encourage them to do that. They will still do it. Like, I, I think we have to understand that it's a both and framework in China. If we, if we sell them a fighter jet, they'll buy the fighter jet and make their own fighter jet. But, but the point is that they is to starve and to use like, doesn't starve the local economy as much as possible and to create dependency on us. The more dependency on us, like it's the less like the war. Also, we can maybe gain leverage on, like, other trade issues, like, you know, cheap goods, fentanyl, Taiwan. There's a whole other layer of additional things. And just saying dual use, which, which, which, again, we don't know what that means. We'll just say it, and they did. We don't. I don't know what I mean. And as well, you can still do it with Huawei. So, like, like, what does it matter? Like, it's. this kind of vague vagaries around it without understanding the clear differentiation between production, consumption, title ownership, cloud consumption, and building an LLM. All those are commingled around Nvidia when each of those are distinct different things that shouldn't be commingled.

Can you give us an update on the Biden administration in January was focused on remote access. Is that a wide scale problem today. You can imagine it's a lot easier to access GPUs remotely in another country. If you want to train a model than it is to smuggle them in across different borders and things like that.

Yeah, the January controls assume two things ended up being 100% wrong. One, how close was Huawei? They assumed like five years, totally wrong. It was like basically 12 months. The other one is that LLMs are remote and they're not. so like what does it matter? Deep sea is the number four, right? Because mainly because it's open source, that's the main reason. And it became a brand name so people know it. But there's lots of alternatives to Deep Sea. There's lots of different reasoning models, there's lots of foundational models and multimodal models. So it's not a moat, so what does it matter? They make their model, okay, we make another model. It's not an innate, just use a different model. We're not going to prevent them from making models. They're going to do what they're going to do. The question is, we're not going to be better. So it's like, if you're racing a car, I'm going to have one race in Canada next week. If the F1 driver is staring at the guardrail, is he going to win? And that's what we're doing. We're looking at it. We're like, oh my God. Are we looking at the finish line? Are we looking to win? Because if we win, it doesn't matter because we choke off the local industry. We become the global dominant player. and then they become Comac, which is the Chinese version of the airline, the alternative Boeing and Airbus. It's gonna be decades before they get anywhere with that airline. And that's because Airbus One, Boeing One, in market penetration is a moat.

And.

Not everything that happens in China is just because they copied us or they stole something. They're legitimately good at engineering. And this is what the point of genius has been making. is that when half of all AI researchers are Chinese, we can't just be like, you know, can't see it, can't see it. Like, we gotta engage with it because at least it's better that it's on our platforms. And if the federal government wants to spy on that, like, why not? Like, it's like we can't even appreciate the asymmetric style of that of like, imagine if we sold them the video stuff and we just spied on it. Like, let's do what they do to us. But it's they were like, oh, right, like dual use, like, oh, military, right? It's bizarre to me.

What's been your reaction to the Russia-Ukraine development? I've always been struck by the fact that Russia has produced so many incredible mathematicians and yet seems so irrelevant in the AI conversation. Any chance that changes? You talk about dual use, seems like they can barely, you know, Deal with any technology over there. Yeah, they make a few drones, but like they're they're years behind the AI race. We're certainly not testing the limits of artificial intelligence on the battlefield over there.

The let me do this that so with our with our company being international, we come across all sorts of characters so in Riyadh, I met a Russian who Russian actually broader caucuses so he speaks Russian, but he's not really Russian who runs data centers in the region. And he was telling me that about like what Russia was doing. And he said that like, don't underestimate them being quiet doesn't mean they're not doing something. And this goes to the Stargate thing and proliferation is like when somebody does something bad with this, which is gonna happen, without a doubt it's gonna happen. The Ukrainian attack on Russian planes wasn't AI, it's just fancy remotes. And there's no evidence that I had a model that was, I don't even.

Know why you would need a model.

You can look at the plane. Just like, yeah, yeah, yeah. It was a very bizarre way of kind of reading that situation. It was very clearly like remote controlled.

But.

When someone does something bad with this technology, which is inevitable, the world will be very woken up to how one, the power of this technology is, but two, how lazy they've been on some of these things that we're talking about. where it matters if we're everywhere. Because if we're everywhere, and let's say something, someone does something bad with us, what does that mean? That means that we have the ability to project influence now across the hundred countries that have our infrastructure. That could be everything from utilizing the infrastructure in a way to bring along allies to counter something, but as well as creates a new type of NATO, a new type of framework, which is the defense of the next century. there is China will pursue. It's almost like a known quantity. China has been trying. There's very little things we can actually can do to impact the directory of 1.3 million people outside of the natural entropy of China itself. China has imploded seven times in its whole, basically, 6, 000 years of History. So, yeah, exactly. Sure. I've got. Please.

Right.

So.

So communism is just the next iteration. of whatever revolution, the Qing dynasty, the Shang dynasty, the Shang dynasty, the Mongols, right? All those were different iterations of the same millions of people die, build a new regime, billions of people die, build new dreams. The ability of us externally being able to influence it is very limited unless we do what we did to Japan and South Korea, which is we totally obliterate any other type of society. We built spaces there, we gave them democracy, and we gave them markets. then you can influence and they can be part of the order, but they don't want to do that. So China's going to China. What we have to do is we do best, which is win, trade, commerce, be the best, make, give us, invest as much money as possible into NVIDIA products. What their recent announcement, I don't know if it's all in Computex, but they're at this basically NVLink as a platform. It's the best way to understand it.

Yeah, break that down for us.

So yeah, so because all of this new GPUs are coming out, like the Groq, Positrons, Huawei, the one of the most valuable parts of Nvidia's infrastructure is its interconnect. And interconnect is both software defined and hardware defined. Nvidia is the king. Like on the GPU level, Huawei is getting closer and closer as AMD, but from an interconnect level, Jensen's in the stratosphere, the key is so far beyond. So it's an effort to keep them in the CUDA system, which is this most significant software mode. he's releasing in the link as a platform to where you can plug in other GPUs and you can work through NVIDIA Fabric. As John just said, the Commerce is going to foreign direct rule the asset, that thing. So it's probably not going to make it into China. And that actually has legitimate claim. Like, you know, we made this software reproduce it and it goes versus Japan is a little much looser claim. And so I don't think it's going to make it in China, but that's the premise, is that he's trying to protect the CUDA The CUDA mode is the mode of NVIDIA. And if Huawei continues to accelerate, open source, framework accelerates, NVIDIA's mode declines, which means more why web product basically gets released into the wild. More models are defined by Huawei.

Is that NVLink licensing more relevant to international companies that want to combine different CPUs and GPUs together? Or is is it more relevant to just maintaining dominance in the American market? Yeah.

With hyperscalers that are developing their own chips, things like that.

Yeah. Yeah.

The pitch is to hyperscalers, but the actual Target is, is China. I'd say that that is the, that's.

The, but, I mean, but you, I mean, you're predicting that it won't make it to China. So why, why even, like, pitch that if it's, it has very low shot of actually getting through?

I mean, they, they 75% margins. So they could do lots of things that don't go anywhere. And also they view that there's two risks to Jensen just as putting on Jensen hat. I'm not an investor. I'm not talking my own book. I'm not much from off. I'm not talking my own book. So they is public cloud with their GPUs, which goes to the same platform in Huawei. Those are the two risks. So he's targeting to both. But the issue with the, what public cloud is doing is that they are trying to de-leverage from them as much as possible, which is why most of the future deployments I suspect in public cloud are going to CoreWeave. CoreWeave is like a third party just reseller into cloud. But his real angle is to Huawei gear where they do not have the ability to build this interconnect via the Huawei products. CloudMatrix is closest and it's still pretty far behind. As well as you look at the networking expense in terms of the margins in the public reporting, They're still growing in revenue, like significantly on networking side, but they're not making much money. It's because they view networking as the moat. So they're really just trying to push something into the Chinese market so they can retain that CUDA advantage. Because they do think that China is accelerating on the most important models, which are the open source models, that if NVIDIA sees the more and more open source accelerates, the farther and farther it gets ahead. the more that it's going to basically cannibalize NVIDIA's footprint. Because if NVIDIA is only selling to closed source models, that again limits their ability to sell widely and to keep our platforms. America has to realize that I'm not just talking about NVIDIA, that's our national champion. So it's like Lockheed, it's like a Boeing. It is our expression into the world of what the future should be and the ability to have our values built into technology, which is privacy, security, Western values, freedom of freedom of thought, freedom of information on the internet. And so I'm fine with the product idea. I don't think it's a risk really much to anything. But most likely commerce will restrict it because if they're already restricting founder equipment, which I generally agree with because it's founder equipment cost $100 million on the side of the house. You can regulate that. We export control airplanes. Well, yes, because you have to put on a ship or you fly it over there. That's not, something you can go buy at Best Buy and put in your backpack.

I mean, Nathan Fielder was able to buy an airplane recently for his show. I want to switch gears for a second. There were some news this morning that Meta is partnering with nuclear energy company Constellation Energy to power some of its AI, I guess, AI factories, data centers for the next 20 years. post these new executive orders on nuclear energy, do you expect to see a lot more deals between hyperscalers or foundation model companies and various nuclear players? Unfortunately not. I'm a big fan of fusion.

I'm a realist when it comes to politics. I'm conservative through and through. Adam Smith, you know, Milton Friedman, Thomas Sowell, like they're my heroes. But I'm a realist when it comes to DC because DC is ideology is an excuse for decisions that are self-serving. It's the way DC works. Nuclear, I love the technology. I think it's immensely safe. I don't believe in this irrational phobia related to it. It has, the third party or the existential causes of the pollution or whatever, if the energy is put in a barrel, put it in the ground, it's like so amazing. And even that, the recycle, making recycle like 98% of it now. But our generation is very different than energy. America's really good at energy, is really bad at power. because power is a federalized decision. And I just don't see a significant amount of willingness to change about that. I hear lots of talk, but the talk is also a little bit like Doge, which is the operational mandate of Doge is discretionary. And even that it has to get the deference of the secretary. It doesn't have a lot of power. You have to have mandatory spending to have Congress do something because it's the way it works. Same with power. is like Department of Energy can't just say, do it, right? Like, like the state, because it's federalized, that's the whole point of freedom, is that if the state says no, they're gonna say no. It's not gonna happen. So maybe on federal lands, maybe on Indian lands, that stuff makes sense. But it's, I think, extremely unlikely. I think it's more likely the, like, Starkey is essentially a sovereign power infrastructure where they're buying natural gas turbines, which is we have significant amount of natural gas, they build it in deregulated energy environments and they attach to the data center. I think that that's more likely because that doesn't require 10 years of approval and, you know, people who have degrees who never done anything in the world showing up and telling you how to do business, like it doesn't require any of those things. It's natural gas. It's fine. And their pipes are everywhere. So what we hear on the data center level is much more of a stretch towards natural gas and that nuclear while there's entertaining. I want this stuff to happen. I'm just so realistic. I think it's, it's, you shouldn't bet on it. And many of those contracts you see, they're more like LOIs if you like dig into it, because the Amazon, the Microsoft, the Facebooks of the world, they know the regulatory hurdles of doing this stuff is years. And so they're basically using the announcement as a means of trying to press.

Through the regulatory cycle to get stuff done. Well, just to give you some more insight, this deal in particular is for a power plant called the Clinton Clean Energy Center. So it's from a different era, right? It's not a it's not a it's taking an asset that we've already had and then doing a net new deal. It's not as exciting as, hey, some new nuclear company just signed a 20 year deal with with Metta.

So yeah, approval of putting servers in the building. or putting it around the building. Like that's not something that's gonna be- Yes, exactly. Like the fear around nuclear is unfounded.

But it's still real.

It's real. And this is democracy. Like you don't get everything you want. And if people don't want in their backyard, they don't. And it is what it is. So like, but yeah, but I'm pro those things. I'm pro clean fusion, like low heat, like cold nuclear reactors. I think all those are super cool.

Technologies that we should Yeah, this thing is interesting because the reactor at this site was going through a decommissioning process, so they're effectively bringing it back online.

Oh great.

Similar one to the deal that Constellation did with Microsoft in Pennsylvania with the Three Mile Island plant last year. So yeah, at the very least it's great to see plants that are clean and functioning, staying online and serving of purpose.

Anyway, thanks so much for stopping by. This was fantastic.

We'll talk to you soon.

Always a pleasure. Yeah, see you man. Cheers. Let's close it out with some timeline. There's news about Perplexity. Apple is reportedly planning to adopt Perplexity as an alternative to Google search on the iPhone, as well as a replacement for ChatGPT integration with Siri. This is kind of rumor mill at this point, but apparently Perplexity is also in talks with Samsung. It could become a default installed search provider on Samsung Galaxy phones and Android phones. Arvin's cooking. Arvin's cooking. Yeah, it's interesting. It's like, obviously the product is not.

Like completely broken through, but-- you- think.

They'Re completely broken through. Do you think they're completely broken through? But-- sponsored an F1 team. Maybe. Maybe it's that, didn't they do a.

Super Bowl ad or no, they just.

Did a really cinematic ad.

But they've definitely been out there sponsoring stuff and-- Ferrari.

Ferrari, really? No way.

They just fully sent it.

That's amazing.

Tim Cook saw that. He goes, okay, let's talk.

Let's talk. It's time to talk. Yeah, I mean, it's certainly smart to break through. And I mean, a lot of these companies win on these default installs. It's always tough when there's another company that's going viral and just growing crazy like ChatGPT. But you got to get creative and figure out where the value is and who's hurting. And Apple and Samsung seem like they're at the top of the list or needing a partner to really keep up with ChatGPT in some ways. Anyway, we should tell you about wander.

Find your happy place.

Book a wander with inspiring views, hotel great amenities, dreamy beds, top tier cleaning, and 24/7 concierge service. It's a vacation home for better folks.

Figure out plans for summer.

Yep.

Book a home on wander.

Yeah.

And Report back.

We also have some cool, probably AI images from if Tesla was made in the 1980s. I don't know how I found this. It only has five likes and 220 views, but I really like this. I thought this was a cool design. Looks kind of like an Aston Martin Bulldog. Or there's a lot of Aston Martin in there. But yeah, I hope with Elon spending more time at Tesla, we get some crazier designs. I want to see a convertible. I want to see something competes with the Jeep. Full-size SUV, the Cyberurban, the Cybertruck converted into an SUV.

I mean, we've talked about this. A sedan with the aesthetics of the Cybertruck. Super cool.

Rip. The Cyberuricon, the Huracan version of the Cybertruck, like the new Roadster with the Cybertruck aesthetic. I mean, I still think the Cybertruck, the Cybertruck aesthetics are really cool, but.

It'S just like not everyone needs a full-size truck. You never see somebody with a Cybertruck using it as a truck.

I've never seen anything in the bed.

Yeah, never seen.

But also we're in LA, so most of the Cybertruck are influencer based. Semi-Analysis breaking down Tesla's optimists. Speaking of Tesla. Ever wondered how those humanoids are dancing in the videos? Check out Tesla optimists here hitting a smooth shuffle. Let's dive into one of the methods they may have used. In the paper ASAP, Aligning Simulation and Real-World Physics learning agile humanoid whole body skills. The researchers did train the humanoid to perform these actions in simulated environments, but there's more to this. So we've been talking a lot about simulation in the context of robotics. It seems to make a lot of sense that you could model all of those joints in a physics engine, run that in Unreal Engine or something. I mean, it does look remarkable. I don't know if you can see that, Jordi, but the, I mean, the dance is definitely not the Biden walk, right? So the problem, play it.

Let's see it.

Yeah, yeah, here, there you go.

Okay.

What? What are you thinking? Conspiracy time?

No, no, no.

It's pretty good, right? I mean, it's definitely a unitary level. Yeah. I don't see why they'd be any farther behind unitary, but obviously there's a.

Lot of gen Z would say the robot low-key has aura.

Aura, yeah. The problem, the robot policies, like guidebooks for performing an action trained in simulations often stumble when transferred into a physical robot. This is called the Sim2Real gap. We'll have to talk to more folks in robotics about that. Many complex interactions can occur in real life that are difficult to model. ASAP decides to train their policy in simulation by using retargeted human motion, like Ronaldo's SUI celebrations motions, but mapped onto a humanoid so they can produce the same So it's the same motions in its own body. However, to mitigate the issues upon transferring to real life, they then deploy the policy onto the Unitree Robotics G1 humanoid. This is from a researcher and train a Delta action model. The model watches and learns what the differences are between the desired simulated motion and the real world.

This Delta correct in particular, this jumping motion, it's wild. is wild.

Wild.

The jumping, because you, because that's using the same, effectively the same physics as a human to do that. And it just feels a lot more.

So they do a bunch of simulation, they transfer to the robot, then the robot does it and messes it up because there's all these fine details that emerge when you're crossing that sim to real gap. That delta correction is then fed back into the simulator, effectively aligning the virtual simulated world and robot with reality. The result is significantly improved agility and coordination and now fluid reproduction of famous celebrations. ASAP is a major leap forward for humanoid robots, but there still remains some challenges in simulations. And semi-analysis says, what do you think? And so it's interesting because it's basically creating a real life feedback loop for humanoid robotics based on simulated data. So you go and simulate the robot to get it 80% or 90% of the way there, then you run those simulations on the real robot and let it make a bunch of mistakes and then feed all of that data back in. So you're generating a lot of robotics data and you can actually have a learning loop to improve this. So still unclear how close we are to fully utilizable, fully useful, humanoid robotics, but it does seem like more and more algorithmic progress is being made, more and more like AI training strategies are being made to really make this stuff valuable. And then of course you care about.

Scaling it up and manufacturing stuff. And then you have to do the robot games where they're gonna have these robots do surfing at Jaws, cliff diving, things like that.

I can't wait. Let's tell you about ADIO. ADIO is customer relationship magic. Adio is the AI native CRM build that builds, scales, and grows your company to the next level. You can get started for free at Adio.io.

They have range in their customers. Everyone from Union Square Ventures uses them to Coca-Cola.

Wow, that's a big range.

Big range. And TBPN, of course.

Of course. When I first opened Adio, I instantly got the feeling that this was the next generation of CRM. It is trying to make money, trying to sell.

If you're doing deals, get on out of here.

We also have some news. There's a new company launching, Amy from Joe Perkins, the AI agent for private market investors. Stop wasting 30% of your week on busy work. Amy handles sourcing, diligence, ops, admin, and research. 24/7 remembers every interaction and sharpens over time.

Yeah, this is interesting. I have the website pulled up. They do deep diligence, so you can imagine some type of deep research product. watchlist monitoring, anti-portfolio analysis, which is fun. Basically, track your passes, memo drafting, exit planning, customer diligence, auto-sourcing, market mapping, team evals, meeting prepper. Pretty extensive. It's cool.

Well, the timeline was in turmoil as Megan Neifeld kind of sort of anonymously called out Alex Lieberman. for co-founding too many companies. Oh, wow.

I saw Megan's post.

You saw that post? Yeah.

I had no idea.

I had no idea who it was either. And Alex kind of self-doxxed himself here. And he says, Someone called me out publicly for co-founding five businesses. What? A man can't co-found five businesses? Come on. I think I've actually co-founded maybe four businesses at this time, although I kind of moved on. Well, I wasn't doing them all the time. You were doing them one at a time? It was one at a time, yeah. I love the reaction. It makes total sense. So I thought I'd explain, says Alex Lieberman, the business barista. He says, I very much don't recommend people do what I do. History has proven time and time again that relentless focus on one thing for decades is how you build anything of consequence. I didn't set out to build a startup studio and co-found five companies. It was the result of trying to optimize for my two highest order goals post-exit, be an A-plus entrepreneur and an A-plus family man. Read, have full control over my schedule and time, spend 90% of my my time in the negative 1 to 1 of business. Early stage businesses make me feel like a kid in a candy store. I do not work 120 hours a week. I work about 50 hours a week. Most mornings I start drinking coffee, walking the dog with my wife. Most days I finish eating dinner and walking the dog with my wife. Those moments are sacred for me. I have an amazing co-founder and CEO for each of my businesses. So he's not the CEO of these businesses. That would be very, very difficult. Who I trust deeply and gives space to focus on time. I view myself as a media company that creates surface area for an array of opportunities. So yeah, a little bit of an uncommon decision, but he's certainly enjoying it. And so, you know, hats off to him.

Yeah, I mean, I think the thing that matters is- Build your life however you want. Having a CEO that is a regular founder, as incentivized as a traditional founder, or over time you'll get cooked by a founder, truly founder-led company.

Totally, totally. So yeah, there's a big question about like, you can be a small angel that owns 0.1% of a business, you can be a VC who owns 20% of a business and uses the word we when describing the company, then you can be, you know, an incubator or a silent co-founder, or you can be the co-founder CEO, the founder CEO, and these are kind of a spectrum and everyone kind of plays in different roles, but, you know, Historically, it's served to work at kind of the fringes of those, but there's different models that work for different folks. This is an interesting trend that I want to dig into, kind of the future of Hollywood. We've been joking about this, but there is something interesting with what Tony Lashley says, Sequoia investing in Essence and Mubi and Thrive investing in A24 in the past four years.

Pronouncing Essence properly, John.

Yes. makes sense. And Thrive investing in A24 in the past four years is quite fascinating. And so apparently the essence investment maybe didn't do so well. Definitely they're clearly struggling, but the model works. And so Sequoia is getting into the independent movie business. Thrive is already in with A24. And so very, very interesting to discuss the impact of AI on Hollywood and independent movie making as these Movies, it feels harder and harder to get the Power Law Outcoming Movies, the Avengers moment, the Avatar moment, the Top Gun moment. But A24 has clearly made a different strategy work, and they're breaking through all the time, and so it's justifying increased investment. Another Perplexity news, they did the thing that we've been talking about. You can now travel, search, and book on Perplexity. No scrolling, 847 hotels and photos trying to spot a decent gym. I asked, find me a hotel in Amsterdam with a serious gym, walkable to conference venues, under €300 per night, result, perfect recommendation, the exact hotel I actually used.

Could you actually click book?

I think so. I think so. And so they're definitely working on it. And this is like an obvious agentic workflow. But Simon Taylor, let us know. Did you actually book? and did you go through with the booking? And was it easier than just going to the website or was it merely better search? Because I think people won't be fully satisfied until it's an entirely chat-based workflow and they don't need to do anything and they never get routed to another website. Yeah. David Sachs has a little review of Mountain Head. He says, Mountain Head costume designer Susan Lyle on the All in Pod. It was a very good source for research for me. One of the hosts, Chamath Palihapitiya, went to the inauguration of President Donald Trump and he showed pictures of all the things he wore. I couldn't believe it. He had a Loro Piana jacket here and a Brunello Cucinelli there. He obviously really knew his stuff.

I mean, that's off to. Very, very cool. That said, it shouldn't take too much research to understand that capital allocators enjoy Loro Piana and Brunello Cucinelli.

Yeah, it's good stuff. This is an interesting post by Ashay Sangvi. Anthropic just went from $1 billion in annualized revenue to $3 billion in annualized revenue in five months. Crazy growth at the scale. And Ashay says, I remember the days when Snowflake doubled at $500 million ARR in a year and people lost their minds. And yeah, the numbers have gotten so big in the AI race that going from $1 to $3 billion is like, what have you done for me lately, Claude? Step it up. But yeah, congrats to everyone at Anthropic. We're having some Anthropic folks on the show this week.

Fantastic. Extremely impressive, but I don't know why that is super surprising. It's impressive, but shouldn't be surprising if you're paying attention.

Given the actual technology. You mean that?

Just how useful these tools are?

Yeah, if you just look at- I.

Guess there's a little bit of like- Trailing, trailing OpenAI. They have less of a, they have less significant of a consumer business, but it's still very big in enterprise.

It's loved by developers. Yeah. Yeah. What else do we want to go through? Yashua Bengio has a new nonprofit research lab called Law Zero. Sharon Gaffery has the story in Bloomberg. She's coming on the show soon. And I thought it was interesting because I was wondering, If, like, has anyone in AI safety actually just tried to use Isaac Asimov's three laws of robotics to design a safer AI system? Like, it would be very ironic and hilarious if that was the solution all along. It was just hard code in, don't hurt humans, which is basically what the Isaac Asimov three laws of robotics are. But of course, the premise of the Asimov movies and books is that but even though they try to encode that, it still goes wrong. So there's still a little of an AI doom scenario, but he raised $30 million of it on it. So congrats to Yashua, and hopefully there's some good AI research that comes out of it. And as we've seen, even if you don't believe in the AI doom scenario, the paper clipping, there's still a ton of other ways that AI can go wrong and have negative effects, even if it's just creating overly addictive products or dual-use technology. There's a bunch of interesting, the Manchurian candidates and the sleeper agents that Anthropic discovered or hypothesized. AI stuxnet. Yeah, and so I think this is worth funding. I think this is worth researching, but you have to take it with a grain of salt and you have to realize the full picture and the full dynamics of geopolitics in the AI race before you actually go and execute against what one of these nonprofits is saying. I thought this post from Justine Moore was hilarious.

Have you seen the new Google AI feed? feature. It's on Gemini. It's literally on AI Studio. You can probably find it on Google Ultra one. Dude, it's on flow. It's a search Labs Labs original. You can find it in Google Docs.

It will appear as a pop-up when you least expect it. So ridiculous and so true.

Just put it. Just put it in the Google search bar if you care so much. They just want to test their product Market fit. They want to see how hard they.

Can make people work for it and still get adoption.

Yeah, you know. They're both different. We have some news separately. Tomo Bravo has announced this morning the completion of fundraising for new buyout funds totaling more than 34 billion in fund commitments. Yeah, hit the button. Can we get the Gong View, please? Can we get the Gong View? There we go. Nice hit, John.

Tomo Bravo.

Nice hit. And notably, they soft circled 1.8 billion out of the 34 billion for Europe.

Oh, interesting. There we go. Get ready if you're a software company over in Europe. You might have a bio offer on your desk soon.

That's not great.

Yeah. If you're in the need of transformation and on optimization, get ready. Good to see private equity finally raise get some real money.

Yeah, you love to see it. Getting some real size.

In other news, open evidence. The Sequoia backed AI medical assistant is reportedly raising $100 million of $3 billion valuation just months after its $75 million Series A at 1 billion. GV and Kleiner Perkins are in talks to lead the round, though details are still being finalized. As soon as it's finalized, gotta have the CEO on the show. Some of the investors would be interesting. I am very interested in this. the vertical, it's like vertical SaaS, right? The niche verticalization of the platform layer. Everyone's agreeing with like value accrual at the application layer. The question is like, how many applications will there be, right? Will there be five or ten or 50 or 100? There seem to be a lot of companies getting into the unicorn territory. How many will make it through and win? Yeah, but it's all exciting. Also, Chime has just launched their IPO road show. Dan Primack has the story in Axios. Market clearing order inbound. A $9.1 billion market cap if it prices in the middle versus the 25 billion market cap valuation. VC-backed unicorns are maybe are finally done clinging to their ZIRP-era valuations, seeing this more and more.

And we come to taking their medicine. Yeah, taking their medicine. Hari calls out, I think they bid it up to 12 billion or so during the roadshow and pops up to 15 billion on IPO day. fair value is in the high teens is his point of view.

Yeah, but we saw this in the information. A lot of private companies are done with their venture journeys. They're ready to go out into the market as soon as the IPO window is open just enough to get through. And so they might take their medicine. They might take a slight down round when they're going out. But then historically that hasn't been that bad in letting them actually perform once they get into the market and go public. So good luck to them.

We should have lesson on.

Yeah, just to talk about this next.

One to discuss this next post from boxate. Very exotic sounding. He says, clearly intern son or VC Scout daughter, which is.

I love that Roy's in the comments immediately. Let me get both.

You gotta have Roy. We gotta have Roy back on to get the update because he and the team seem Just be wild and out right now. It's actually insane. I see again, I'm starting to feel like a bit of a boomer because I see some of their exchanges. And I'm like, oh, this has to be. Yeah, I'm like, there's no way this is real.

Yeah. There's no way.

There's no, you know, it can't possibly be real. But I don't think they ended up hiring 50 interns. Yeah. But it seems like they have a lot. Gotta have them back on. Yeah. To break it all down.

We should have all the interns on. We should have him with all the interns standing behind him. And they can just shout answers over each other. Loudest intern wins.

He had an interesting-- so he had a post that went fairly viral yesterday. He was quoting somebody who said, what genuinely pisses me off about younger founders or just new people getting into startups is how obsessed they are with raising. Goes on and on.

Roy said that?

No, no, no. Roy quotes that and says, Hot take: founders need to focus more on hype-chasing. okay. Startups don't live and die by a product being five percent better. They die because of lack of money or delusion. Hype chasing helps with both. I don't know how hype chasing helps.

Yeah, you still gotta build a product, but distribution is really important.

And so, like, you can. He said, inflated revenue is still a user. Is still Revenue. Users who pay out of curiosity are still users. And good Engineers who are down to interview after a cool launch video are still good Engineers. Interesting. Viral moments are not so easy to come by. and the potential upside of making the most of them is higher than you think.

I like this because it's a debatable point. Roy, his story is clearly going to break one way or the other. It'll either be a breakout success or kind of a learning lesson.

Yeah, he had another post that I saw where he said he's paying founding engineers hundreds of thousands of dollars in cash because he said that he can just raise infinite amount of equity. Which is built different.

He's just built different.

He's built different. And yeah, I'm just interested to follow on. I'm rooting for him despite him being controversial. I want to see him succeed. I think he is, he's very good at causing controversy. Like that, even just that, the idea.

Of like, oh, we're not giving equity, we're giving all cash, like that's a controversial approach. Some equity.

Yeah. But he's just like, I value I value equity way more. I'm just going to pay.

Yeah, but he's selling the equity at high value. You could do that with the shares. It's very odd, but it's like good bait.

No, I mean, if he says my, if he believes that his equity is worth a multiple of what.

Yeah, but he's selling it to pay the cash.

He's not making money. He has cash in his bank account today. Okay, yeah, he said. I'm in a four x.

Sure, sure, sure, sure, sure. In a couple months. Yeah. Makes sense. Anyway, interesting post from Eperus. We haven't had them on the show yet, but we have to at some point. So quoting defense index unstoppable China can produce 500,000 FPV drones every month and scale up to 700 in wartime, which actually feels like roughly the same scale as Ukraine, maybe within an order of magnitude, which is shot. It's actually kind of lower than I expected, but it's obviously a lot. And Epirus says, this is not unstoppable.

Yeah, so Ukraine is doing just under 400,000 a month right now at their current rate. Okay, so they're doing that basically in a cave.

They have scraps. Yeah.

Like China has all of the, you know, so I imagine China could actually do well beyond this.

Yeah. It's also a weird narrative because in what scenario, are these FPV drones, can they get over the Taiwan Strait? Because I believe they're pretty short range and so it's more for, like, What is the trench war scenario that would happen over there where these would be super relevant?

Yeah, it's interesting to think about they could have basically a bunch of floating docks that the TV does.

I guess that's it. Yeah.

Just boats.

I mean, especially if you can just put them into a shipping container and then move the shipping container. So it's like instead of an aircraft.

Carrier, you just have a shipping, like a cargo ship. Yeah, so with Eperus, I'm interested to see how Eperus DARPA is actually doing against these FPV drones. Soren yesterday said he hadn't seen an American electronic warfare that could really reliably take down FPV drones. I'm sure Epirus can. It has a different-- conditions and it should be a part of a broader system.

Yeah, so they say that their mission is to overcome the asymmetric challenges inherent in the future of national security. And so good luck to them. We want everyone to win here. who's working for America.

And we should note earlier today Chamath was sharing an image of a, I forget what machine this is. It's not a scale. I think you hold these devices that, Like an in-body. Yeah, that sort of track your body fat. He says, Hot girl summer, bring it. She's putting up 11.5% body fat.

Pretty good.

At allegedly 6'2. And Chamath responded in saying, Are you trying to get your body fat percentage as low as the IRR of anyone that invested in your 2021 SPACs? Fortunately, I don't think it's possible to go negative on the body fat front. But then Chamath made a very lewd comment. Chamath fires back. They're in each other's faces. Said, Brothers, this conversation is unbecoming. of the high technology industry. I propose we shift the conversation to what really matters: capital. Would you both like to join TBPN today? And Chamath, unfortunately, turned it down, so we didn't get here.

Maybe in the future.

But he's got a podcast.

I like everyone working together to maximize returns at all times.

Yeah, I think we need to work on our pay-per-view format, get the boxing ring set up, Imagine smashing the soundboard as two allocators just duked it out. I mean, no better entertainment.

It'd be spicy, but that'll have to be for another day. Anyway, thanks for watching. This has been a fantastic show. We will see you tomorrow.

Looking forward to it. Have a great afternoon.

Bye.

Cheers.