AMD CEO Lisa Su Live on TBPN | Oliver Cameron, Mohammad Norouzi, Anjney Midha, Lisa Su

You're watching TBPN.

Today is Thursday, July 23rd, 2026. Jordi is wearing a bandana. What corporate logo is on that bandana? TBPN collab here with AMD Advanced Micro Devices. Uh, yeah, beautiful. Thank you to the team for hosting us. Thank you to the AMD team for having us here at AMD Advancing AI. Uh, we're on the road in SF today. Uh, we have a bunch of great guests for you. Lisa Su's joining. In about 90 minutes. But let's take you through a little bit of the news of what AMD's announcing, what's happening today. So Lisa, of course, will be joining us at 1:30 PM. We have Oliver Cameron from Odyssey, Mohamed from Ideogram, and Anjney from Amp coming back to the show for round 2. Round 2 or round 3? We've probably talked to him a few times. 3. Might be the third time. Advancing AI is AMD's biggest AI event of the year where it reveals new chips, software partnerships, and how it plans to compete with NVIDIA according to Jackson Fordyce from the TBM newsletter.

Fun fact.

We just found this out this morning. AMD is actually powering the new Mod Retro N64. Very cool.

It's very, yeah, very cool.

Crossover Palmer, Lisa.

Yeah.

So we're going to be spending all 30 minutes of our time with Lisa Su on gaming specifically.

Exactly.

How is, how are they going to bring—

and suits.

And suits.

Yes. Suits.

Wear fantastic suit collection.

Got a fantastic suit on today.

So the 3 key stories that people are focused on out of AMD. Advancing AI. Our AMD unveils Helios, the new rack-scale AI system powered by the MI450 accelerator. We'll talk to Lisa about that. Also, a new partnership with Frontier Lab. AMD and Anthropic are partnering to deploy up to 2 gigawatts of AMD Instinct MI450s.

Sizgang.

Sizgang, $5 billion deal, and a whole bunch of, a whole bunch of Collaboration across compute and the software stack. The big question around the software ecosystem around AMD is always a hot topic. I'm sure there's a ton of developments there.

Only makes sense. And on stage just a little bit ago, they were talking about how Anthropic's own models are effectively speeding up their ability to ramp up new hardware.

Yep.

Which is something that has been talked about for years now. Some questions around the CUDA moat, right? If you assume that making great software will just get easier and easier. That's been a big question. And they're starting to see real movement there.

Yeah, but also a lot of that speed up requires actual deep partnership like this because there are some closed source packages that you might need to work on. This is the old George Hotz interaction. Yep. Talking about little bugs that he was frustrated with, wound up working with AMD developer team who we've talked to on the team, on the show. Fantastic outcome there, but interesting to see that flywheel just get tighter and tighter and tighter, and I expect it will. Also, AMD is expanding its partnership with Cerebras, teaming up on AI inference. We love the team over at Cerebras. They introduced a powerful disaggregated inference solution, pairing the right engine to each phase of the inference pipeline. This is what agentic AI has been waiting for, the fastest production inference at massive scale. Excited to talk about that as well. But our first guest is here, Oliver Cameron from Odyssey. I have a whole bunch of fun backstory. You can come on down.

Here he is.

I first met Oliver almost 5 years ago. You were working at Cruise doing easy stuff, self-driving cars, and you took a ton of time to talk to me and walk me through exactly how you were thinking about self-driving. You explained a lot of different stuff to me about what's being done in machine learning, what's being done with C++. Very excited to have you back on the show. How are things going?

Things are going great.

Thank you so much for having me.

Yeah, great to have you. Congratulations on the massive progress. Last round was what, 300-something billion, a million, a million raised? It's gonna be billion next week.

A million billion.

Yeah.

310 million.

310 million. Yes. Fantastic. And where is the product right now? Like how are you describing, how much is in the research phase? How much are you ready to deploy this? With early adopters? How are you thinking about that?

We are in the GPT-2 phase of world models.

Yeah.

And what we're building is foundational world models. The idea being that this is a technology that can be useful and applicable to tons of different industries, whether that's, for example, robotics or science or gaming or driverless cars.

Yeah.

And so we've been on this journey now for 3 years.

Is there still a market in self-driving cars? Because it feels like the world models A lot of that research was done at Cruise and Waymo and Tesla, and that was where I was first introduced to them. I was like, oh, they built a video game version of San Francisco, and of course they're training it there. Is there a world where you wind up powering future self-driving initiatives and autonomy initiatives?

I think so. I think world models can power all sorts of virtual and physical systems. The virtual systems might be gaming, for example, might be education. And then the physical systems will be driverless cars, robotics, drones, all that sort of stuff.

Sure.

What I believe is that driverless cars today and lots of automated systems look very much like NLP systems in the 2010s. They are these very complex, very hand-tuned systems. They're intelligent, but they're intelligent in sort of isolated ways.

Yeah.

And that those systems hold back what these technologies can do. And so I think you can replace these very hand-tuned driverless car systems with a single world model that has this very deep understanding of physics, of cause and effect, of human behaviors, and just lightly tune that world model to the task of driving.

Yeah.

What we're seeing early evidence of is you can indeed do that. You can take these very big general world models, tune them with just a few hours of experience, and then the car drives itself.

Pretty cool.

So what are your interactions and relationships, partnerships look like with various types of robotics companies? So think of like if someone's building a humanoid or we were talking with Travis Kalanick yesterday doing a autonomy in mining. So like, what, what does an Odyssey relationship look like with those companies today?

Absolutely. So we very much believe that the robot companies are customers of ours. We provide that base intelligence. They then have their robot, their embodiment, their task, and they can take that base intelligence and tune their model to make that task amazing. And what we see earlier evidence of is that using a world model versus a vision language action model, the number of examples you need to show the world model is dramatically less than a vision language action model. Which means it's just much more adaptable to the world. Sure. Than a vision, uh, language action model.

Makes sense.

Talk about data right now. Um, a lot of these world model systems, it feels like they were bootstrapped on Unreal Engine systems or just video game systems generally. Uh, I imagine that the amount of data that's getting poured in the pot and mixed together is huge at this point. But what are some of like the power law data sources generally that are useful in building world models?

Yes. So I, I'll get to the point eventually, I promise.

Yeah, take your time.

So what I observed in driverless cars, yeah, and we've seen in robotics, is that you typically train those systems on very narrow data distributions, right? So you're a driverless car company, you collect lots of data from the road, dash cams, basically dash cams, but also cameras around the car.

Exactly.

We've all seen the Street View cars.

Exactly. You're really teaching the car just from the perspective of a car. Right? It's solely seeing what cars see.

It's got the blinders on.

And what those cars don't see in training is video games, or is inside a conference, or in an office, or in a home, or just the multitude of scenarios that exist in the world. And what we very much believe is that a world model shouldn't learn from a narrow distribution of the world, like a series of driving examples or robot examples. It should be that you learn from every possible thing that could exist in the world, and then you tune the model to the task of driving. So you have a base that's just very general, and then you provide driverless car data at the very last step. And so really, our belief is that that will produce a more robust, more intelligent system than these very narrowly trained models.

I love that. Makes sense. Completely tracks with the GPT-2, GPT-3 progression in language models. There was no point when people said, oh, well, we really want to solve IMO-level math, so let's get rid of all the fiction. No, it was always include everything, right?

Exactly.

You need Harry Potter to solve math problems.

Hilarious. Very odd scenario, but it is true.

Absolutely.

I'm interested in the more expensive pieces of data that I could imagine you'll be acquiring in 6 months, maybe a year, maybe 5 years. I don't know. But we've heard about data labelers making 7 figures creating RL environments for very specific tasks, very niche use cases.

Just 7?

I mean, it's going to be 8 next, next month, next quarter. But, but, but truly, like the idea that the data labeling that went into early driverless cars, but also early reinforcement learning, human feedback on text LLMs was just is this acceptable? Is this readable? Does this have typos? And now it's, does this actually track to case law, or does this look like a financial model that I could turn in at an investment bank? What is the world model equivalent of that? Can you imagine what that will look like?

I can. So I think if you look at AlphaStar, which was DeepMind's training of agents inside StarCraft to then beat StarCraft players, which was incredible. Yeah. Really what you had was agents learning inside this environment that doesn't update or improve. StarCraft is StarCraft. The agents then have a ceiling on their performance because it's always StarCraft that they're learning within.

Yeah.

And so I think what world models really promise is the ability to be a learning environment that's continuously improving and adapting to the agent's intelligence. So for example, an agent should be able to learn within a world model that is consistently improving over time. It's getting more robust, it's getting more diverse, there's more scenarios. And so the agent's intelligence being trained inside that world model should also get more robust.

Yeah, it's so fascinating because, yeah, training a model to win at a video game is one thing, but the real world is never fixed in place the way a video game is, right?

Exactly, yes. And a world model really can be thought of as this sort of infinite simulation, right? It's continuously generating new types of environment that the RL agent can explore and adapt to and fight within and all these, these cool things. And so I think actually one of the key or the killer applications of world models will be a learning environment for AIs, language models, other types of AIs can all learn within world models. Gets very much philosophical.

Exactly. Yes.

Staying on philosophy, if there are, there are many, many labs that have made the bet that text is the universal interface. Scale and maybe coding software-only singularity. There's a number of different buzzwords to sort of define this idea that like LLMs are on the path. Then you have the folks who are arguing that everything from LLMs are a dead end to deep learning is a dead end, blah, blah, blah. But what I'm interested in is your— if you're looking at— there's clearly progress going on in text-based, code-based LLMs that feels like a path to something called AGI. Maybe that happened 2 years ago, maybe it's happening 2 years from now, but it feels like there's a trajectory there. Are world models like the next step? Is it a separate curve? Will they come together and work together to produce whatever the next, the next paradigm is? How do these two technologies and like paths converge or diverge on the tech tree?

One slight tangent is one of those critics for language models, Gary Marcus.

Yes, I wasn't going to call him out by name, but you're welcome to.

Me and him and many others in driverless cars have been bickering for some time.

Oh yeah, because he's an Uber guy. I forgot.

Well, before language models were the debate, it was all about driverless cars and when they would come to market. Is it 2050? Is it 2040? It's like, no, it's just going to happen much sooner. Conveniently, he's now moved on from that being the debate. But anyways, so my belief is that language models are learning.

There was some new cope yesterday as well that I was seeing that maybe I shouldn't bring up.

Fresh cope? Okay, anyway, let's continue.

My belief is—

Cope is evolving almost as basically as fast, fast takeoff, as the models themselves.

Okay, yeah, yeah, that makes sense.

That's the frustrating thing, the goalpost is just moving, moving, moving.

We have this ghost goalpost that we move around the studio because there's always a new goal.

It's ridiculous.

But that's the, that's what makes it fun. It would be boring if we reached the goal and we were done. We want to keep building.

That is true.

And so what I very much believe is that language models are going to continue on their trajectory. They're exceptional technologies. They will lead to a form of superintelligence that's Amazing. Changes the world, of course. Um, what I believe they learn from though primarily is a sort of biased representation of the world, and it's our writing of the world. And it's also a remarkably, I think, inefficient representation of the world. If you were to describe what's going on here in text, imagine everything, every detail was somehow describable by text, how long would that representation be? It'd be huge, right?

Yeah.

As I pull out my phone, I capture 2-3 seconds, I've captured so much detail of humans intermingling and everything. So I very much believe that they are almost distinct, somewhat complementary technologies. Yeah, there'll be cases like coding, incredible for language models. I don't think world models will play in that market too much. Yeah. And creative writing and many other emails and things like that, amazing for language models. World models, I think, will play this role of operating within either virtual or physical worlds. And that those will be just distinct applications. A world model will prove to be a better driver of a car than a language model. A world model will prove to be a better operator of a robot than a language model, flyer of a drone, creator of a video game, educational experiences even. I think they'll prove to be distinct. I do think there is a convergence somewhat of the, the learning environment thing I mentioned, that world models will serve as this universe for language models to learn with them.

Yeah.

And that's maybe where it converges at some point.

Selfish question, please, on autonomous driving. How do you see the market evolving from here? I've been having a lot of conversations recently where people are buying, buying Teslas, not specifically because they want a Tesla, but because they want the autonomous functionality. And I feel like a lot of the the manufacturers are in a lot more trouble than maybe they even realize because there's plenty of drivers today where if you have a maybe 10, 20 minute commute daily, you're not really thinking about how do I solve this problem? But for people that are driving any longer than that, it's starting to be like top of mind. That's like the number one factor. How, how do you see this evolving? Is it gonna be like language models in that there's a bunch of different providers with like pretty good autonomous driving? We're not seeing it quite yet.

I think you can see, you know, the two wildly different companies that are both exceptional, right?

Right.

Yeah.

I mean, right now you have Waymo and Tesla.

Yeah.

And Waymo.

And then there's a bunch of, but like wildly different tertiary players.

Yeah.

So I see it going something like this, which is that firstly, consumers, word of mouth matters. And as many people are now trying Teslas, it's going to become almost insurmountable for many of the car companies to compete with just how good a Tesla is. Yeah. All things, not just driverless functionality. And so I think we'll see the continued sort of attrition in the smaller car brands converging into Tesla. That's one thing. Second thing is I think it's becoming almost, what's the right word here? Like, a national crisis to not enforce driverless technology. I have two kids and it's today still the leading cause of childhood death is car crashes. And the idea that we have this technology now where I can literally walk out here and be driven fully driverlessly back to my house like 40 miles away.

But can you, can you put on a sympathetic hat and try and have some empathy for the ambulance chasers?

Yeah, the trial lawyers.

The trial lawyers.

They are people too.

What are you— I mean, yeah, you have to think that. No.

I knew where that was going.

Coming back to diffusion and self-driving, I want to go back to that idea of like 2050 prediction. It sounds ridiculous because you can walk out on the street and see one driving around San Francisco. But if you rephrase the question on driverless cars to a billion driverless cars, full diffusion such that There are, you know, like they make themselves available everywhere over the world. It's the dominant vehicle pattern that actually does take time. It's more of an industrial process than a technology breakthrough. How are you processing diffusion of AI tools, LLMs, and will world models follow the same trajectory as LLM diffusion, which is really fast? Everyone uses them for everything, but they haven't taken all the jobs. And we're sort of in this interim thing where it's amazing,, but it's additive and it was sort of unexpected to everyone, both the people that were like, it's fake, and both the people that were like, it's God. You know, it's like we kind of got the middle case, right? It's just like, okay, things are going well, keep building.

Exactly.

I, so specific to world models, I think that there'll be a very similar sharp progression of adoption that we've seen with language models. There'll be some killer applications that become very clear very soon.

Yeah.

The idea that you could type a prompt and get out a AAA-level game.

Incredible.

Doesn't seem crazy to me. Multiplayer games too. The idea that you could have generally capable robots in your house and offices and everything else enabled by these models becomes clear. And then yes, driverless technologies and many others. And so yeah, I think as what becomes the thing I always lent on in driverless cars is even if you didn't believe in the technology, you just have to look at the volume of people and the smarts of the researchers behind it and how much dollars was going into investing in that category. To assume it's going to get figured out at some point, right?

Yeah.

It's not an impossibility that we'll have driverless cars. And so you put enough smart people behind, it's going to get done. And I think the same of humanoids, for example, like there's just so much capital, so much talent flowing into that space. It's going to get solved, right? It's just inevitability. And you just keep going. And I think this is why I have such faith that, yeah, we should. I mean, there's no reason now we shouldn't have a billion driverless cars in the world.

Yeah. But timelines, by 2030, will I be able to walk in and get a Honda and have it have autonomous driving capabilities that's as good as sleep Tesla's is today.

Okay.

Not even, not even, not even what Tesla will be in 2030, but like a Tesla, Ford, you know, these other manufacturers.

2030 is like 2 years away.

No, but, but, but I'm just saying like, it's like, I, I, I'm curious because I think every passing year all of these manufacturers are just in more trouble.

Maybe.

Yeah.

Yeah.

I think they get in their own way. Right. So I think a company will build a technology that they could sell. To a Honda or other companies that is driverless ready on that scale in 2030? Absolutely. I mean, we've got literal superintelligence coming that can write code of crazy quantities and such. So I think so. Then it's really the car company's decision about, yeah, do I want this? Will I adopt it or will I continue to burn money internally failing at building it myself?

Last question. How important is at least developing some of the application muscle, the go-to-market, like Suno just went viral in our office. We're all addicted to Suno now. And there's been these moments where like, okay, everyone's on Midjourney, it's really fun. And like, it's this platform and then it accrues a lot of value. It can sometimes be baggage because you're building, you're like, okay, now I'm a consumer company. How do you think about value capture and how much you want to be the front end to just prompt, get a game?

Exactly. So my answer would be that, uh, I'm actually not happy with my answer, but sometimes I get asked, what are we good at? Yeah, Odyssey, right? And so my answer to that is that we need to be great at everything. And the reason I say that is because if you're building a very foundational technology, it really does need to be great at all of these different tasks. Um, and we need to stay focused on that instead of getting really focused on a minute case. And, um, so Yeah, long story short, I very much believe that by building this foundational technology along the way, you'll discover by talking to companies, talking to researchers, really cool stuff.

Some of them might be partners, some of them might be in-house. Well, I'm very excited for it. I'm so ready. I can feel that the moment's coming. Congratulations on all the progress.

Thank you.

Thank you for all the work you're doing. Thanks for coming on the show.

Thank you.

Have a great—

Thanks so much.

Cheers.

Thank you for your time here. While we bring in our next guest, let me tell you about Ramp. Time is money. Save both. Easy to use corporate cards, bill pay, accounting, and a whole lot more all in one place. And we are— we have a soundboard here. We are joined by the founder and CEO of Ideogram, Mohammed. Welcome to the show. How are you doing?

Suited up.

Thanks for having me. Looking good.

Thanks for—

awesome. You guys are looking good too.

Yeah, it's the right place for us. Why don't you introduce the company a little bit? I know a fair amount, but for the viewers who don't, the shape of the company, and the product and sort of the target market at this particular juncture?

Yeah. So if you think about it, we already have content creation in front of us. If you think of the industry, you know, we had the early days of image generation. Yeah. And if you look through your social media feed, you see a lot of AI-generated video. Yeah. But we think there's an inflection point that design marketing can also take advantage of. Yeah, generative AI. And that's what we're building at Ideogram. Yeah. AI foundation for enterprise adoption of the next phase of content and design.

So do you think that something like image generation can be fully solved within a shorter period of time than, let's say, intellectual intelligence?

It's kind of interesting because when you think of image generation, you can have a very detailed document and diagram and really technical design as part of an image, you know. Oh yeah, it's not—

at some point, if you're saying generate a blackboard with a solution to an IMO-level math problem on it, the model needs to understand that level of math.

Yeah, like think of the circuit design or architecture design.

Yeah.

And, and oh yeah, That's the same level of complexity as a lot of the reasoning problems we're trying to solve now.

So one of the first times—

how you look at it.

Yeah, yeah. I was asking from the lens of, of— I think everyone on Earth now has now viewed an AI image, not known it was— not known that it was AI, and just assumed like, okay, looks like a normal image to me. And I think when you're talking about like if Pepsi is working on a campaign, they don't need like as soon as it looks photo real, they don't need something that's necessarily more real. And so I just wonder what that means for the shape of Ideogram and where at, at some point, like what the, what the core, uh, competency will need to be for Ideogram to continue to evolve as a business, if that makes sense.

Yeah. If you, if you think one is we still have issues with consistency.

Yeah.

It's not at a level that a brand can use for their marketing for their design. And we focus a lot on image generation and not as much on design generation, which is, okay, I have certain typography I want to translate across languages. I have my logo, it has to be 100% accurate, you know. With product photography, it has to be exactly correct. I can't buy a piece of clothing and then it doesn't, it doesn't match the picture. So when it comes to consistency, uh, you're still lacking. But then part of the dream is to help with the design of future products, you know. The creative part of, uh, design requires a lot of understanding of the context, uh, and then the brand DNA and putting them together to design the next generation of cars or design the next generation of shoes. And that's where we are focusing on. And we are working with brands to help supercharge their design and production.

How are you— to me, that is, I will just say, the most— my most addictive AI experiences is designing products that don't exist yet. Just anything I could possibly imagine in my head. And then the process of like prompting and, and, you know, getting, getting close and, you know, getting to maybe 95% and then trying to get it again. You know, the last 5% ends up taking, you know, uh, 10 times more time than, than getting to the 95%.

Yeah.

Um, so related to that, what are enterprises looking for specifically from you like this month? Because we just went through the token maxing, uh, up and down where enterprises were just, you know, spraying tokens all over. You can use whatever model for anything, check the weather with, you know, the best advanced model. But I imagine that your customers are concerned about cost, but also speed, quality, consistency. I know you're partnering with AMD obviously, but walk me through a little bit of like the problems that you're hearing from enterprises and what you're focused on solving in the short term.

Yeah, one thing is data sovereignty. And obviously when it comes to design, IP is extremely important. Lots of competitive industries that, okay, I can't let a competitor see the design of my future car. Oh, that's interesting. And imagine sending that over to the cloud. Yeah. And then that creates a lot of potential issues.

Does that mean they want to run your models on-prem or does that mean that they just want to know that you're hosting it in a certain secure way with a certain SLA around, we're not going to train on this?

I would say both. Both, okay. Both. And with the newest model that we released, it's actually a relatively compact model that's at the frontier and it's open main. Wow. And that creates a new set of opportunities for us in terms of licensing and partnership where companies can host these models on-prem. Yeah, that brings down their cost, but also data sovereignty is really important. So on the design side, that's one thing we're hearing a lot.

Yeah.

And then on the brand side, this is still, uh, consistency is an issue, but we think we're going to all that in the next few months.

Yeah, there's a big, there's a big debate over, uh, open weight models, geopolitical discussions, and one of the themes that's coming up is like American open weight companies might be at a disadvantage if, uh, other countries' open weight models have less respect for intellectual property. And so I would think that you would be fighting with one arm tied behind your back, but based on your business clients, they might not want a model that infringes on IP because that could wind up getting them in trouble. So is it less of an issue for you, or do you see like sort of the move fast and break things that may or may not be happening internationally actually being a like a headwind to your business?

On the image side, actually seems like American companies are doing really well. Yeah. And we are at the frontier, so it's not as big of an issue. And then the customer side, As long as we give them the protection they need, they are happy.

Yeah.

But one other set of customers that we are seeing is actually generating data to train really sophisticated models for manufacturing problems, for really unique defense problems where you don't have a ton of training data. And these models are getting to the quality that they can generate images of defects, images of really rare scenarios that you can't find in the real world.

Yeah.

And that's another example where, again, having it on-prem is important because they don't want to share that out, share that data with everyone else.

Yeah.

Interesting. Interesting.

What has it been like working with AMD? We've talked about the software ecosystem around AMD. You're obviously— give me some benchmarks on what you've actually seen from your performance, and then I wanna know about the process to actually optimize one of your models for an AMD stack.

Right, so it started with us testing some of the early models. It was MI300, and then we got to know about the roadmap. It's a really impressive roadmap with MI450 and the following ones, 455. So we started testing it, we got good results, And with AI agents, it's become so easy to, you know, give it the, the architecture and then get it to be optimized for, uh, for the chip. And the AMD team has been really accommodating in terms of helping us out whenever some issues arise. And then we went live a few months ago with our 4.0 model.

How much does the roadmap at AMD and other semiconductor companies affect how you're designing your product? Do you work backwards from this model? I know it's going to be able to do X, Y, and Z, hold this sort of memory, this flops, like, and then, okay, let's design the best model for that rack or that chip.

Yeah, exactly. So we often customize the details of an architecture based on price, performance, latency. And then work backwards and train the model when we get to the details of the architecture. There are certain changes that you can make that wouldn't have an impact on the quality of the model but would have an impact on the downstream performance and latency. So I think everybody does that in the frontier AI, that you kind of work backwards. But the problem is you also don't know too much about the future chips.

Maybe that's why you got to come here, meet the people, whisper, oh, really? That's what it's going to be? Okay, we'll work backwards from that. Or maybe you just have a proper deal and everything's about the same.

We didn't get to this, but before you leave, what were you doing before this?

Oh, I was at Google before this. Before that, I was a competitive programmer, then turned AI researcher, and then started Ideogram. At Google, I started image generation, this brand called Imagine. And I thought I can have a bigger impact.

How much of the team is researchers at Ideogram now?

It's about 50-50.

50-50?

Wow. Yeah, we have a relatively small team. It's primarily engineering.

That's really cool.

And half of it focusing on the product and half of it is focusing on the model.

Sounds like a dream. Well, congratulations on the progress. Thank you so much.

Thanks so much. Great to have you.

Great to meet you.

Thanks for coming on.

We'll talk to you soon.

Cheers.

Let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI, own the data platform that powers it. And I'll also tell you about Codex. Codex is a powerful workspace for getting work done with AI agents. Whether you're writing code, analyzing data, creating content, or automating business workflows, Codex helps you move projects forward from start to finish.

Welcome.

Welcome back. How you doing?

Great to see you.

How you doing?

I love it. Just saying hi.

Amazing.

Thank you, man.

What's happening?

What's up, boys? Good to see you again.

Good to see you again.

I think this third time.

Oh man, that jacket is so sharp.

The green, the TVPN green.

I—

that's the TVPN green.

Okay, well, we'll coordinate. We'll connect you to our tailor.

Do you have a— do you have a brand color yet for AMP?

We have the Cream and the Force Grid.

That's Cream, right?

There you go.

We actually have a Force Grid coming.

Okay.

There you go.

There's a new video model that just got announced this morning called Flux 3 that's training on the Amp Grid. And it's from Black Forest Labs.

Very cool.

And they're up in the Black Forest, like near Freiburg in Germany.

Wait, say Amp Grid again.

The Amp Grid.

That's a good sound effect for the Amp Grid.

I like it.

It's good branding.

We'll get you the sound effect. We'll have our tailor talk to your tailor.

Yeah.

Figure it all out.

Yeah.

Careful what you wish for, man. I would love the recommendation on that, but Black Forest Labs, their color is forest green.

Okay.

Yeah. Because, yeah, Black Forest, etc. And I find it's a very soothing color.

Yeah.

There's some like— I think there's some like neuropsychology around like green immediately calms you down.

Yeah. You don't want caution yellow or, you know, high-vis orange. That's a little bit more aggressive.

Well, there's some companies that make orange work.

Yeah, well, that's not a— Anthropic does not have a high-vis orange, although high-vis orange is maybe the one white space in branding. A16Z is orange, but that's not high-vis.

That's not high-vis.

High-vis is like reflective orange. You know what I'm talking about? You see it on the road.

That's hard, man.

That is an ingress. That's hard to pull off as a brand. How are you thinking about the brand? Do you have a whole brand book yet?

We're going with the vibes and the aura, man.

There you go.

I like it.

As I've learned, just The vibes, generally the vibes like are in the space of frontier AI, which you guys have been an extraordinary voice on. Stuff changes every day.

Yeah.

And so nobody's got any time to look at the first principles, the details, what's actually happening, what's like, which model's coming out, what's the difference between Fable 5 and like Codex, blah, blah, blah. And so it's all aura-based decision-making, right? Because the brain just needs some easy heuristic to latch on.

Yeah, I'm surprised we haven't seen a chief aura officer. Yeah, that's the next, like, chief brand officer. But when somebody does that, sell everything.

It's probably too much. Once there's a name, you should try to own it fully.

You make a good point, which is that if you have motion, It doesn't matter what your site looks like or what your logo looks like. People just attach their own feel, what they feel when they're interacting or to whatever you're putting out.

Motion is underrated. There should be an AI company research in motion. That would be good.

Wasn't that BlackBerry?

Yeah.

Anyway, I'm not that.

So, but I want to answer your question, which is what is the brand? The brand is what you stand for. And we stand for output maxing.

Okay.

Right.

I love it.

There's just too much— like, what are we living through right now? We're living through the AI scaling era.

Yeah.

Bitter lesson holds.

Yep.

That scale, scale, scale.

Yep.

But the brain is naturally anchored to large numbers.

Yeah.

But nobody— like, if you actually deconstruct what's going on in this space, there's so much wastage because nobody's doing the output maxing efficiency play, right? Which is what's the output divided by the unit of input?

Yeah.

And if you start measuring businesses or model capabilities or whatever through efficiency, the story is completely different than, than what just the headlines say.

Yeah.

Right.

A lot of people were token maxing, and I think we will look back on that era as really valuable exploration. It was actually totally worth it. Yes. But I like output maxing as the correct coinage as opposed to token maxing. I was trying to say token min-maxing, token optimization. It's way better output maxing. I love it.

It's like, what are we all here for? It's to try to grow GDP, not at all expenses. It'd be grow GDP in the compute-optimal way.

Yeah, right.

You cannot beat China at pure scaling. It's just not gonna happen. Okay, industrial, like industrial-scale coordination where they go, you know what, ban the H100s, H200s, all of you use Huawei, and like get together and figure out how to scale with— even if you don't, we don't have the leading-edge chips, put out Kimi K3. I don't care, I'm Xi Jinping, like I just want for us to be at the frontier.

Yeah, and make, you know, 100 new nuclear plants, whatever.

Build, build, build, baby. Like, you know, build, scale, whatever is required. U.S. permitting, who needs permitting? Just like, you know, they will just clear all the bottlenecks, and they have been at a scale we don't understand, right? They will bring at least 100 gigawatts of new energy capacity online in the next decade without even blinking.

Yeah, yeah. Meanwhile, they weren't even AI-pilled, and they were planning on doing it anyway. So they're like, yeah, yeah.

So what do we have to do? We have to be smarter.

Okay.

We have to do output maxing. We've got to be more efficient. Only 15% of every single tenant data center today in the United States is being utilized. NetFLOPS utilization is less than 20% in the United States. It is crazy.

Is that just downtime from chips not being used at the right time, like misconfigurations, like the water supply is out or something? What's going on?

Basically, for every dollar of long-term lease that an AI lab takes to a day, you lose about 30-40% flops in just bad scheduling. The nodes are just straight up not allocated.

Yep.

And then within—

Chip's there, but it's just not getting work at the right time. Correct.

Then within the chip, you only have like 15% of the chip being utilized. That's called MFU, model flop utilization. Because the chip is waiting around for some other process to complete, storage, memory, networking, or some other chip to hand it off. So as a result, if you compound those two things, of every dollar, only 15% of flops are actually being utilized. Yeah, that 85%, that's a national security crisis. Yeah, because that's what's keeping us from staying at the frontier. Yeah, that's why we all keep complaining about, oh, KVK 3 came out. Sure, you can talk about distillation. Distillation is definitely happening. However, everybody who feels compute constrained in the United States should be asking, am I doing both? Am I getting more capabilities? Am I building up more supply? And thank God Lisa is on it. Yeah, but also, are we utilizing the existing capacity to get the maximum output possible. And both those things are what's going to keep us at the frontier.

Yeah.

Does that make sense?

Yeah.

Yeah.

It's sort of, I mean, it's natural cycle. Any company that, that has grown really quickly or any industry that's grown really quickly, there just ends up being all this waste and misutilized resources and all this stuff. And then, but, but the key, and I think what you're pushing for is like, even if you have this exponential growth, you can still try to be more, it's still important to be more efficient. Now so that we don't give up the advantages that we do have?

I don't think we have a choice because the supply chain is backed up for like 2 years. So we're just not like we at AMP, the AMP grid, we're procuring energy and new capacity for 2030.

Wow.

And we— and there's not enough sites. Yeah, we're going to need nuclear. Where's—

are you spending time with nuclear companies?

Yes, actually, there's Seth right over there.

No, no way.

Hey, Seth, how you doing?

Seth is a founder of a company called Mavria that we've just invested in.

I don't think Good to meet you. He was—

he was—

pop on the show after.

Yeah, after he gets off. Well, yeah, we'll talk to you. Go over there. Talk to them.

Yeah, yeah, yeah, yeah. Talk to the team.

Seth was at the DOE.

Yeah.

And was one of the people responsible for a bunch of permitting that just finally got, like, approved for new nuclear projects in the United States.

Cool. And he's—

thank you for your service.

Okay, well, we'll give you a proper mic in a couple of minutes. We got one more question. We— I know.

Yeah.

So, so walk us through the shape of your business right now? You have the fund, you're making new investments, you got a big allocation in, I think, obviously every Anthropic round to date. But otherwise, what is the shape of the business you're building out? To me, it's like two different businesses. They work, they're very interlinked, but how are you spending your time?

Yeah, so we've got Amp Foundry, which is our venture capital firm. And capital business, and we raise venture capital funds.

You don't make a single chip there.

We do not make any chips.

Still not yet. Not yet. Who knows? I wouldn't put it past you.

Lisa's a great partner.

But our infrastructure arm, which we're actually spinning out pretty soon as an independent entity.

Oh, interesting.

And we're not ready to announce the name yet, but you guys are going to love it. There we go. So that's for a follow-up.

Okay, great.

Is building out about 2 gigawatts of capacity in the United States.

Okay.

And we're not making our own chips yet. Yeah, but we're building out new— there's new sites. Yeah, there's new—

Yeah, you're marshaling the capital, getting the right partners in place so everything can come together and the compute can come online.

Yes.

Speaking of 2 gigawatts, what can you tell us about this Anthropic-AMD deal? I think it's 2 gigawatts, $5 billion investment. Oh, 6 gigawatts. Is that what it is?

Up to 6.

But I'm interested in First, it just seems like another positive sign for the industry for Anthropic and AMD. Yep. And it feels like we're also, I don't know, I haven't checked all the reactions, but it feels like, are we post-Circular Deal FUD? This seems like a very logical deal. You know, everyone's seen the ARR charts, everything's growing, businesses are using the tools, like the full, it's a full ecosystem, the economy's growing. And so this part, this type of partnership makes sense. But what do you tell, how are you telling an earlier stage founder if they're trying to do a strategic deal?

Well, well past the circular part. I mean, these were all concerns maybe 2 years ago.

I agree.

And so if you look at the value creation, when you have like Anthropic go from zero, literally when we started the business like 5 years ago, yeah, well north of $40 billion in run rate.

Yeah, right.

And then, and most of that revenue is coming from everyday customers using Claude to code. Yeah, that's not circular, that's net just new value creation.

Yeah, right.

Now the question is, how do you keep a healthy independent ecosystem, all kinds of frontier capabilities churning? Sure. And the problem is that the, the core problem, one of many corporate bottlenecks, is that private credit markets do not see startups as investment-grade counterparties. So when they're trying to procure capacity, compute capacity, like 3, 4 years out, they just can't get— those contracts are not financing. And so what we need in the United States is a financing program that says, okay, we're going to be able to turn startups— I mean, Anthropic was a startup not too long ago that was not considered investment grade. At this stage, they actually likely are still not investment grade. Neither is OpenAI, but they're starting to become more underwritable by the financial markets, right?

Yeah.

And I think what we've got to do in the United States is do a little bit of innovation to go, how do you get the ecosystem to understand the quality of, of these startups? Yeah, they're the engine of innovation. Yeah, there's where, you know, Anthropic is 5,000-ish people today. Yeah, Google is, last time I checked, like 60,000 people.

Yeah.

Accenture is 75,000.

And neither Accenture nor Google are responsible for frontier model innovation today. Like, Google still hasn't put out an Anthropic-grade coding model.

Sure.

And they're trying their best, but it's a focus thing, right? It's a focus culture. Like, talent-dense teams that are focused get more done.

Speaking of that, are large founding teams underrated?

Are large founding teams underrated?

Because there's a world where Anthropic's like the exception to the rule. It's working for them, right? Crazy. I mean, Young company, but huge company. All the founders are still there.

I think Underworld is another great example.

Underworld is 5, and strong team. There was a while where the Silicon Valley wisdom was like 2, maybe 3.

Right.

Yeah, most startup advice, as you guys know, is designed for the media.

But for podcasts. Yeah, that's what I was going to say.

But the outlier businesses, they— They have large founding teams. They have family founding teams.

Yeah.

Like the Stripe brothers.

Yeah.

Conventional wisdom is don't start, you know, like, yeah, don't make personal business.

Yeah.

And Dario and Daniela are brothers and sisters. The Collisons are brother and sister.

Yeah.

Black Forest Labs is 12 co-founders. They're all researchers.

I had no idea.

Yeah.

So I think outliers do their own thing.

Yeah.

And that's the point is like, whatever it takes to compete at the frontier.

Yeah.

And as long as you're super aligned and tight, I think that's what matters.

Okay.

Uh, let's say you're, let's say you're talking, uh, you do have a panel in a second. I wish we could go longer. Uh, you're talking to somebody who is now willing to admit that this AI thing is real. They see the revenue ramps. Let's say they're using various models. They're like, this is very capable, but they're getting into it. And they're saying like, okay, a lot of, a lot of the revenue is tied to coding right now. These tokens are deflationary over time. What is the next leg up in consumption? Where is it gonna come from? Right? Because like finance is an interesting category, but it's, I, I don't see, you know, $200 billion of, you know, financial modeling spend on tokens, right? So where are you predicting or thinking about these sort of like next legs up after coding?

So the gift that keeps giving still is reinforcement learning. Like RL is just working.

Yeah.

Like I said, technology, it's so simple. It's like training a dog, right? Like, hey Fido, go fetch. You did it right.

Here's a treat.

Here's a treat. And reward modeling used to be like this bespoke craft thing, like 2 years ago. Now we're getting to the stage of industrial-grade, like repeatable reinforcement learning as a service. And so I think taking models and doing post-training where you have the weights and doing post-training on data that you care about on some small set of samples that are high quality, where the reward is very clear, and you're doing RL gets you your own jagged frontier very fast. Yeah, you know, because you guys know this, like capabilities, there's a jagged frontier. You know, coding is clearly one area because of formal verification with unit tests. The progress has been super fast. I think verification is— people just still haven't truly internalized like how generalizable this concept of like the context feedback loop with verification built in is. And the algorithm is so simple RL just works, that wherever you can do formal verification and RL, you're going to see crazy capabilities. So one example is Periodic Labs. It's one of our portfolio companies. They're trying to discover a room-temperature superconductor. The pace of progress there has been extraordinary in the last 6 months. They've got a 40,000-square-foot facility in Menlo Park where you've got AI models that are predicting new materials, new candidates, And then you have robots synthesize the material and then test with an X-ray diffraction machine. Does the material have the superconducting properties that the model said it would? And because that's verification from reality, like from nature, like physics, you can do X-ray diffraction to test. Does it have it or not? The feedback is the signal is very clear. And then you take that signal and pipe it back into RL like as many times as you need.

And it's exciting.

The capabilities ladder is extraordinary.

I want to go way deeper with you on science and what's next in AI. We got to let you get to your panel, but thank you so much for coming on and hanging out.

Good to see you, boys.

Pleasure.

I hear I'm the opening act for Lisa, so she's next, and have fun.

Yeah, and thank you for the introduction. We'll bring in your portfolio founder CEO first. I'm going to tell you about Public, public.com, investing for those that take it seriously. They got stocks, options, bond, crypto, treasuries, and more with great customer service. Welcome to the show, introduce yourself.

How much do you bench?

Uh, this Morning, 3:35.

There we go.

Amazing.

Can you lift this microphone up to your mouth?

Thank you.

And maybe move it over a little bit. There we go. The other way. Oh, yes.

Okay.

Loud room.

Loud room. Anyway, loud and clear. Give us an introduction yourself and the company.

Hey, guys. My name is Seth Cohen. I am one of the co-founders of Mavria Inc. Okay. We are an AI infrastructure company, and we're looking to tackle some of the biggest problems in the space. And we're so thrilled to be working with Ange.

Yeah.

Before this, I was chief counsel of nuclear policy at the Department of Energy.

Oh, no way.

The— depending on who you ask, either a senior advisor or the conservator of the Nuclear Regulatory Commission.

Yeah.

And an advisor at NASA.

Oh, cool.

So I was the DOJ operative, one of two responsible for implementing the president's nuclear executive orders across the government.

Very cool.

Yeah.

What were you doing before that? Because that doesn't seem like the first job.

Supreme Court and appellate litigation at a firm called Kirkland.

Oh, okay.

Yeah, I heard of that.

Yeah. So the hat says nuclear. You introduced the company as AI infrastructure. Like, is that because AI is the hot trend and nuclear is just like the byproduct, or are you pure nuclear company? Like, walk me through where the Venn diagrams overlap between just AI infrastructure broadly, what you'll do and what you don't do, and then nuclear, what you do and don't do.

Yeah. So one of the things that my, my Doge team leader, who is now my co-founder, Adam Blake, one of the two things that we are— three things that we first honed in on with nuclear were a set of bottlenecks that we thought were really going to be— we're going to prevent anything we had done— we did from actually taking effect. We needed to solve a couple of other things. Yeah. And one of them was narrative, right? Nuclear is this incredibly incredible technology that has so much potential for the world.. And yet for 60 years, it's faced this strange opposition, including around nuclear waste. And by the way, if we, if we can, I'd really like to talk about nuclear waste. It's my favorite thing in the world.

We love talking about nuclear waste. So you're in the right place.

It's good to be among friends.

But when—

so what we did with nuclear waste was actually we shifted the narrative and went from Yucca Mountain being the only place where you could put high-level waste, completely off limits, to where we are now. 27 governors applied for this program that Adam and I started called the Nuclear Lifecycle Innovation Campus. And there are— I can't tell you the exact number, but it fits on two hands. The number of governors currently fighting to host the full back-end fuel cycle.

So would that be— they're competing for one additional nuclear waste storage plant?

3. 3 to be one of 3. And these are going to be cities, by the way, that, that if everything goes according to plan, will be competitive with our major metropolitan areas because we're going to reindustrialize.

Wait.

Yeah, yeah. Cities. I feel like you're just talking about a hole in the ground and you'd like 2 guys to protect it. What else is going on around a nuclear waste protection site?

Well, so it's not, you know, to take a step back.

Yeah. Is it too late to offer up my backyard?

The ultimate Yimby.

No, I think what So first off, when we talk about nuclear waste, that's really— we have this strange idea and it's probably The Simpsons' fault.

Oh yeah. Glowing.

Yeah.

I have seen the blue glow of a spent fuel rod, which is actually very, very cool. But there's no green goo. Yeah, right. So almost all, all waste in the United States is medical stuff. It's like gloves because some person went behind a wall and pressed an X-ray button.

Oh, interesting. And yeah, by volume, that falls in there. You think of it as like the pellets. But it's really anything that touches that. So there needs to be a whole processing facility for that.

That makes sense.

Yeah, that's most. But the second is commercial spent fuel.

Yeah.

And just so we're clear, it's— it is fuel, but it's not spent by any means. About 97% of the potential energy remains in a fuel rod after it's done and has to be taken out of a reactor.

So why, why take it out if there's— if 90% of the rod's potential is still still there.

Mm-hmm.

So there are two main isotopes, one of which is getting split and used up. And so as that potential goes down, the ability to sustain a fission reaction also goes down. And then you accumulate byproduct materials. So it's everything from plutonium to some really, really interesting stuff. So, but like before we get there, right, it's that let's talk about like that remainder Right. So natural uranium is around 0.9% of this really good isotope. Most commercial fuel is 3.5 to 5. When you're done with a rod, it's around 0.7. We have 100,000 tons of this stuff. And this is from the lifetime of the commercial fleet.

Yeah.

Which, by the way, that fits in one football field.

Yeah.

Like 60 years of energy, but also should be repurposed in some way.

Yeah.

Yeah.

That there is 4 times more energy sitting on those pads outside of our reactors than Saudi Arabia has in its proven oil reserves.

That's actually crazy.

But that's not where the money is.

Yeah. So where is the money? Where are you spending your time? How much time is in D.C. lobbying, like leveraging the old stuff?

But close, close the loop here. So there's, there's all this sort of like latent energy potential. I imagine you're doing something with it.

Well, we got this program into motion and I just wanted to talk about it because I'm so excited about what's going on here. And I think that we're on the cusp of a real energy revolution in the United States. But I think to understand why we're shifting, right, it's, it's the weave.

There you go.

To understand why we're shifting here, you have to understand that what we really were successful at is getting people to see this thing that historically has been treated as a liability, as a real asset. An asset where governors are going out and staking their name on joining this. Like Utah and Tennessee both made their applications public and they are awesome. That is a successful narrative shift. So that's bottleneck one to AI deployment is energy. So now we can back away from nuclear. We actually think that's moving in the right direction. So what's next? Well, Right now, data centers are facing a tremendous amount of issues in the United States. A lot of it is public perception.

They're deeply popular with like 5% of the country.

Yeah.

Less.

On a good day. But these things are really, I mean, they're critical. These are critical national assets. And so we want to help shift the narrative in order to make AI deployment in the United States possible and make sure that we are able to, roll out and remain competitive with China. I am dancing around what it is we exactly do, and part of that is because we're— we are in stealth.

We'll come back on the show as soon as you're ready to come out of stealth.

Yeah, thank you guys so much for having me on.

Fantastic pleasure. Good to meet you.

Enjoy the—

enjoy the rest of the conference.

Great to meet you.

And I will tell everyone about Shopify. Shopify is the commerce platform that grows with your business and lets you sell in seconds online, in-store, on mobile, on social, on marketplaces, and now with AI agents. Uh, let's go through the rest of the news, get you up to speed before Lisa Su joins in about 25 minutes. Uh, Google has the muscle to overpower spending worries, says the Wall Street Journal. Uh, uh, Google has put up, uh, some very, uh, impressive cloud growth numbers. There, of course, was some back and forth on, um, on the nature of that revenue., with Semi Analysis sort of jokingly maybe posting Google Cloud generates revenue primarily from the sale of TPU systems. The actual line that did change in the SEC filing that's drawing attention is that, uh, there's a line in there that says Google Cloud generates product revenues primarily from the sale of TPU systems, which is a little confusing because you think of a company's product as their business, but of course they have services as well. And so the— it's not that Google Cloud is fully pivoting to TPU sales exclusively. It is that they have sold some.

But it's becoming meaningful.

Yes, yes, yes. They've sold them to a number of labs and they're seeing lots of process.

And that was— rewind a year, that was still a question.

Yes, yes. We can go through— there's a number of different posts here. We can take you through Eric Sufert talking back and forth about—

Soufanator?

The Soufanator's putting a little bit of truth zone on Google Ads growth, what's going on there. They saw a lot of growth in monetized LLM queries. First, I'm gonna tell you about the New York Stock Exchange. Wanna change the world? Raise capital at the New York Stock Exchange.

Just do it.

So it's a bit of a longer post, but it all started with Sundar Pichai's news. He said Q2 was an amazing quarter with our AI investments redefining what's possible across every part of our business. Alphabet revenue grew 24% year over year and Google Cloud accelerated to 82% growth, which is insane. That is the foghorn for sure. Uh, we saw a, an exciting momentum across the board from Search to YouTube to the Gemini app. Uh, our model APIs are processing 22 billion tokens a minute. Uh, and At this point, I'm like, the numbers have gotten so big on tokens per minute, tokens per month, that it's very hard to keep, uh, keep in sync with what's actually, uh, impressive until you actually see a full chart. Um, but that's driven by their Workhorse Flash model, which makes sense. They've been focused on very efficient token generation because they, uh, vend them into so many Google systems for effectively free because they're still, you know, monetizing via advertising. So Uh, Max Anderson said, as someone who has personally spent $500,000 a month plus on Google Ads for years, I can tell you with certainty this revenue growth in search is artificial and extremely unhealthy for Google's business long term. A little bit of fear, uncertainty, doubt. Search volumes are declining as legacy search is being increasingly cannibalized by non-monetized LLM queries. Google's response: manufacture revenue growth via short-sighted, highly extractive, customer-hostile tactics, i.e., charge advertisers more for lower quality clicks, including clicks they do not want and explicitly do not approve Google to charge them for. He gives a few examples. He's going back and forth saying that this is, you know, short-term optimization. But Vesuvinator comes in, he says, unfortunately, every claim made in this viral thread is either false or mischaracterizes Google's policy. One, search volumes are declining as legacy search is being increasingly cannibalized by non-monetized LLM queries. Uh, Google stated in its Q2 earnings that search usage hit an all-time high during the World Cup this year. It also stated in Q1 earnings that search queries reached an all-time high that quarter. Ads and AI overviews monetize in parity with those in legacy search, and Google's search revenue increased by 19%, uh, and, uh, in Q1 and 17% in Q2. So a little deceleration but still growing. Um, he also debunks the idea that Google silently deprecated the second price auction. Google still uses generalized second price auction for search. He also debunks the idea that Google previously had precise keyword targeting settings that allowed advertisers to pick individual search phrases to bid on. Uh, there's a different thing, but Souvenator goes back and forth and, uh, and adds some more context to the Google earnings. People are digging into, and we can dig into more later this week. Uh, let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases, and more while Railway automatically takes care of scaling, monitoring, and security. Over in the Wall Street Journal, there's a couple articles. Uh, Alphabet rides cloud business, revenue jumps 24%, but data center outlays push cash flow into the red. I think— the first quarter for negative cash flow. But this has been telegraphed for a long time. They're investing in—

not a surprise—

boom, not a surprise. And there's been talks from all the hyperscalers around debt issuances, equity issuances, all different things to fund the AI buildout. So Google parent company Alphabet posted 24% revenue growth year over year in the second quarter, fueled by its booming cloud business. But concern over the company's heavy spending on AI infrastructure damped investors' enthusiasm. Alphabet sales came in at $119.8 billion, exceeding analysts' expectations. Its cloud business brought in $24.8 billion and its search business generated $63 billion. Net income was $112 billion for the April-June period, also ahead of analyst expectations. So they beat, primarily driven by its gains in other companies stocks that Alphabet owns. They own, I think, almost $100 billion of SpaceX. I'm not sure about the current valuation, but, um, in other AI news in the Wall Street Journal, uh, two big, uh, AI infrastructure stories. You have Lisa Su herself right here. Uh, AMD and Anthropic sign AI gear deal. Uh, we talked to Anjani a little bit about that. And OpenAI boosts cloud spend to $750 billion. New projects and Georgia data center will be part of efforts to increase capacity. OpenAI is scaling up its data center ambitions and its budget for spending on them. The artificial intelligence company has raised its projected spending, uh, on computing power to around $750 billion through 2030, up from a projection of roughly $600 billion. It's a huge number, obviously. A lot of discussion about that. But given the most recent leg up based on coding agents, a lot of people are looking at that positively as a sign of continued faith in the progress and the scaling law, of course, because we are in the scaling era.

Serious momentum.

Yeah, the scaling— the increase reflects new agreements with cloud computing providers as OpenAI races to lock up the enormous amounts of computing capacity it needs to to develop and run its AI models. OpenAI spending on cloud computing has become a central focus of Chief Executive Sam Altman's leadership team and has been a source of tension between him. Um, the company said Wednesday would invest $20 billion in, uh, to kick off a data center called Project Camilla in Effingham County, Georgia. So that's new news. Also, NVIDIA has a supplier that's—

have you ever been to Georgia?

Yes, I've been to Atlanta. It's a great time.

Gotta visit.

You've never been?

Maybe on a layover.

It's like LA.

I don't think I've been outside.

It's like Los Angeles. You're in the car for half an hour driving in traffic to get everywhere.

Everywhere.

But it's a really nice city.

Uh, Liangwen Feng over at Deepseek had an investor call. I don't have context on how this, uh, actually came up because they're a hedge fund and they have Deepseek, still a private They're raising outside capital now. They're raising outside capital.

Yeah.

So this was probably part of that roadshow.

Yeah.

I think at a certain point he needs to raise enough money that you got to just get on Zoom and sell the dream.

But he was grinding for 4 hours, just monologuing for 4. I don't know if he's monologuing for all 4, but it is, it is sort of a long call.

But that's bullish though, because if you can go and basically do a Joe Rogan podcast episode, it's just monologuing. No, it's just like, can somebody actually carry a conversation for that long? Because after, you know, this has always been your take on Joe Rogan. It's like it gets interesting on Joe Rogan because by the— after an hour, people go off script.

Yeah.

And so if a CEO can carry a conversation for 4 hours straight.

Yeah.

And still be compelling, it's, it's bullish. But there are some details that Zephyr shared. He's over at Citrine. Their inference margins are around 85%. Strong. They only had around 20,000 hopper equivalents until May. Still sort of— I can see how they would say that. That doesn't mean that they don't have access to a lot more power globally. Yeah. Some of which may or may not be above board, but who knows? They have $1 billion in API revenue is enough to turn the company cash flow positive.

Yeah.

And they have a GPU payback period of 10 months and their hardware is depreciated every 3 to 5 years. So looking, looking quite good over the pond.

He also shared a lot of philosophy that felt sort of derivative of Western AI lab leaders' thought. I mean, he was focused on, you know, some of the same language around we don't want to monitor. We don't want to optimize for pure profit. We're on the path to AGI first. It's something that you've heard from a lot of lab leaders. Uh, he did say that he's very focused and he wouldn't be doing world models or image generation or video generation or even a chat app. He said he doesn't want to be the next Alibaba, is purely focused on the path to AGI, sort of taking this focused approach. And so we'll see. Uh, they've, they've sort of been out of the game with Moonshot sort of coming from behind with K3 and a lot of hype there. And then GLM, of course, as well. So it'll be, it'll be interesting to see what he has up his sleeve on the back of this investor call that everyone's paying attention to. In other news, let me tell you about Console. You see it on the laptops. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets.

Gong moment! Cognition is acquiring Interacton, the makers of Poké.

Huge.

Uh, I think Poké has been an acquisition target or been talked about as an acquisition target for a long time now. They've just made— been able to deliver delightful, novel products. Uh, I wouldn't have expected them to go to Cognition. I think a lot of people were thinking that they'd get picked up by a big, uh, consumer company, maybe a, you know, an Apple.

Weren't people saying Apple?

Apple would have been amazing. Uh, having the Poké team working on Siri would be, would be quite cool. Uh, could have seen them going to OpenAI or really any, you know, even, even an Amazon, right?

Yeah.

Uh, so a little bit unexpected here, uh, but not surprised that Scott saw potential in the team. And I'm very interested to see, uh, how these businesses end up actually merging. Will they keep Poké around? Will it be just sort of folded into Cognition? But either way, very excited to see what they build.

Yeah.

Let me tell you about Figma. Agents meet the canvas. Your AI agents can now create and modify your Figma files with design system context. There's an older article in the Wall Street Journal that we never got a chance to read through, but it's It's fun because it tells a story of a young person who basically started— literally listened to a podcast and started a search fund and made, I guess, millions of dollars. So did the meme. Yeah, I did the meme. Uh, this shortcut to private equity riches is minting young millionaires. And I saw Sal Khan from Khan Academy talking about something sort of adjacent to this, uh, because Khan Academy, of course, sort of open source education. You can basically learn everything that you learn in college from Khan Academy just watching videos, um, and doing their, their online open coursework. I believe it's a nonprofit.

It's still crazy that, that Khan Academy exists and, and plenty of people still turn it down.

Yes. Oh, they do, they do. Um, yeah, this person didn't vote, basically. Uh, Go-Getters are backing to buy— are backing to buy up HVAC outlet outfits and specialized manufacturers. If they make it back alive, there's a lot of money to be made. What a quote from the Wall Street Journal. So, uh, in 2015, Bakari Akil was a homeless college dropout with a single focus: figuring out how to get rich. Sleeping in WeWorks on the subway in an airport waiting areas, Akhil watched videos, listened to podcasts— let's give it up for some podcasts— and pored over personal finance books. He needed to own a business, he decided, but how to do it with— how do you do it with no money? Then Akhil read a Harvard Business School case study that described how MBA graduates were running around the country trying to buy companies with money from something called a search fund. He felt like he had discovered a secret. In the decades since Since Akhil, now 37, has bought 2 multimillion-dollar companies. He has spent each month of the past 3 years living in a different country. So he's doing the, the nomad lifestyle. He's running HVAC companies? I guess. We'll get into it. Uh, he's now worth 7 figures. Akhil is part of a growing wave of people looking for a shortcut to private equity riches. Armed with grit and determination, they are ditching the well-worn path from spreadsheet-wielding associate to managing director and finding ways to by HVAC and plumbing outfits, specialized manufacturers, and other small and mid-sized businesses. Financing from— for deals— financing such deals is far cry from traditional private equity, where firms use funds they raise from institutional investors and the ultra-wealthy, plus debt. Instead, these would-be business owners scrape together financing on a deal-by-deal basis. They use loans from the U.S. Small Business Administration, seek funds raised from specialized investors or cobble together money from friends, family offices, private equity funds, and SBA-licensed small business investment companies. And that's just the start. Convincing the owner to sell and executing an improvement plan is a daunting task even for the most experienced dealmakers. While these buyers typically—

you're overseas constantly.

Yeah, I mean, I imagine that he's overseas now, but during the heyday, he was like actually going and, you know, knocking on doors. But, uh, the reason I brought up Sal Khan was that he was, uh, advocating for instead of college, buy your child a business, or, or take a group of, of peers your child's age and they can buy a business together. Because if you look at the cost of college, yeah, you know, $200,000, $250,000 sometimes, If you get 4 students and they're effectively the executive team and you say, look, we're— we have a million dollars in cash, we're gonna lever something up.

So we're gonna have 200K of net income. Yeah. And we're gonna lever it up.

Yeah. And the cash flow is gonna pay your salary, so you're gonna have a job.

Yeah.

From 18 to 22 when you graduate, and your job is just to learn everything on the fly. And best case, you wind up with a a fantastic business that makes a bunch of money, but worst case, you're no worse off than the money that you would have spent on college. And so you get to learn all these skills on the, on the fly. I don't know. I don't know if it'll take off. It feels like Alpha School adjacent. Uh, but when the time comes, yeah, maybe we can get all the kids and say, I mean, you guys, it's hard to predict what the world would be like in 15 years or 20 years if you're just having kids now, but it doesn't seem that crazy to imagine. But I think, I think the aura of, of, of certain schools will endure and that will be the right path for some people.

For sure.

Anyway, I view these guys as home— the homesteaders of the Wild West, said Albrecht, who previously represented a big private equity firm and a law firm, Gibson Dunn. They're heading out without a dollar to their name, and if they make it back alive, there's a lot of money to be made. Many caught the bug at a top business school where entrepreneurship through acquisition courses are now a standard part of the curriculum. They are drawn by the promise of a more flexible schedule, a conviction in their own operational know-how, and desire to avoid grunt work that ends up lining the pockets of others. Fueling— also fueling their rise, a perception that it isn't as easy as it once was to make big money on Wall Street. Makes sense. A lot of the big private equity firms were founded decades ago, have founders that have, and partnerships that have accrued a lot of the value, and, and it's much harder to just scale up something new. Uh, private equity industry is struggling to profitably unload companies, causing mid-level employees dubious of their chances of receiving pay tied to deal performance to jump ship. Some 77 search funds, money from specialized investors to back individuals looking to buy businesses, were raised in 2025. 77 search funds, new search funds listed by the, uh, according to an annual study conducted by researchers at Stanford in 2025. That feels low based on how big the meme was all throughout 2025.

Maybe there's no, there's no way that that's accurate.

Yeah, maybe there's some that are, that aren't being caught in like the, in the, in the definition, or they, they, they have defined their company as a, as a true private equity firm instead of a search fund. So they said, no, we're not a search fund, even though they might actually be acting as one. But that is historically high, although down from the 2023 peak in— of 101. There was 101 search funds founded in 2023. Meanwhile, the number of firms doing deals as independent sponsors without a fund has roughly doubled to 1,400. So there's a whole bunch of different, uh, stuff going on. Um, there, there's other stories in here. Caroline Sabbat bought a Boston-area plumbing business in 2025 and merged it with a smaller one her and her husband had started. They got, they got joint degrees from Harvard's Business and Government Schools after nearly 10 years in the US Navy. He was a former Navy SEAL and he launched Minuteman Plumbing, Heating and Cooling with a master plumber he met through a Harvard mentor between graduation and starting his job at a consulting firm. They took over running the business —And grew it. All of a sudden, I'm working 15-hour days on this plumbing business. Caroline, now 35, who managed a crew of 9 as a flight naval officer. It was going well, but not well enough that you can light your high-paying consulting gig on fire. Caroline decided to buy a business to kickstart growth. She found one that they could finance with an SBA loan. These typically come with a lower interest rate than a bank loan because they are government-backed. They did have to personally guarantee the loan. Caroline also had to take out a life insurance policy. She bought the business for 1.8 times EBITDA, for a tiny fraction of what most private equity firms pay. So not too bad to get that market.

Yeah, you have to imagine it was very subscale otherwise.

Working for someone younger than me and having to live life one PowerPoint slide at a time was torture, he said. And he quit his consulting job. Anyway, there's another interesting Google news. Google did a— they released this massive, massive study. I think it's like 100 pages about the impact of AI. And they have very— a very, very unique dataset, obviously, because Google is vended— like, the AI is vended into Google Search and all the different tools and the API. And they say AI is helping workers, not replace them. We've been hearing glimmers of this and people sort of saying they're backing off of what they were saying before, like the massive job displacement. Maybe it's not coming right now. If you use your own eyes and ears instead of reading, uh, like thought pieces or sci-fi or something. Yeah. Oh yeah, that too. Yeah, yeah. You know, it's all over here. Oh, what? Okay, cool. Um, we will be joined by Lisa Su in a little bit, uh, but for now we'll go back to the Google study that says AI is helping workers not replacing them. So, uh, one of the looming worries about artificial intelligence is that it will automate away jobs. This, uh, worry certainly has been looming. Uh, a report from researchers at Google released Thursday says that so far the technology is mainly being used as an aid to workers. It's a yes-and technology. Uh, it is finding— it, uh, it is a finding that if holds true as AI continues to develop could point to a future where The demand for highly skilled workers is accentuated rather than diminished. Just because you're using AI doesn't mean it's going to automate your jobs, said Google economist Scott Strand, one of the researchers. The stakes are high for Google parent Alphabet. Google operates Gemini, one of the most popular AI platforms. Public backlash to AI is intensifying with stark fears about job loss and anger over the buildout of data centers across the country. Some large corporations that have laid off employees, this was your point,, in the past year have said that they were able to downsize because of AI, yet the US unemployment rate overall remains relatively low. And, oh, okay. Um, and, uh, the, uh, and economists remain split on whether the technology will ultimately eliminate jobs or simply make workers more productive. We can continue chatting about this in a minute, but let me tell you about CrowdStrike, your business AI. Their business is securing it. CrowdStrike secures AI and stops breaches. We are going to be adjusting our audio setup. I think we will be joined in just a few minutes. Are you ditching yours? Back here. Are you going full bandana for the interview? Is that what you're thinking? I don't know. Anyway, I think we are— You gotta get the backstory.

On the bandanas. You don't typically think of AMD and this sort of Western-themed bandana, but now I do. I mean, I don't know. John, did you know— Cowboys. —that we landed on the moon before we put wheels on luggage? That's impossible. That's what they're saying, John. We put a man on the moon before anyone put wheels on suitcases.

How is that possible? No one thought about wheels on suitcases?

Seem to let people— were just like that. 1970 was the first time that we thought, why don't we put—

it would be great to put those things from the car on my luggage. If only we could miniaturize the technology.

If it can be done, we should do it.

Yeah, we have to.

But if no one's done it before, it might be impossible.

It's sort of a moonshot project. That is, that is actually a crazy, crazy statistic. I mean, that's like the— what, sharks are older than trees? Isn't that a thing? I think sharks have existed in the ocean since before trees or something. That's one of those like, uh, like brain teaser things. Um, anyway, uh, what else is in the timeline? I'm trying to pull up this, uh, AI study because there were some interesting details in here. But, uh, the researchers characterized the use of AI as shallow in most occupations. Workers are only using AI for a relatively small slice of the tasks they do. Uh, we were talking to Travis I was talking to Elon Musk about this yesterday. Truck driver, what's the job? Is it to drive the vehicle or is it also to protect the vehicle? Is it also to do small repairs on the vehicle? Handle logistics with the vehicle? Driving is just one of the tasks. And so even if you have a level 4, level 5 self-driving system, you might still have someone in the vehicle for a long time. And that's like the most automatable task that people have been working on for 20 years now, and it's still not quite there. The other interesting thing that you were asking Ajne about, what is the next leg up? What is the next thing that causes another order of magnitude boom in token consumption basically, or like revenue or whatever metric you want to measure the AI boom through? Yeah. And there are just like diffusion elements, right? Where, you know, the number of people that are using coding agents is still small and you could just see diffusion that way.

But there is— Yeah, I think there's some data, US households that pay for AI products is still well below 2 digits, somewhere around 5%.

Yeah. I do think that—

Part of that is like there's just a lot of households that are not going to pay for AI, at least directly.

There is just the, like, going back to the meter chart, which has showed the amount of time that an AI agent can work on its own. You know, we're in this, like, reliable agent paradigm, but I think people might be undercounting what, you know, a willingness to run a prompt that will run for a week and give you a result. Like, we're not quite there, or maybe it needs to be more communicative. Maybe the system needs to get back and forth with you, but people just aren't, uh, it's, it's, it's, we're still very, very early adopter to, uh, the type of person that fires something off. Most people at least want to check in with the system in, uh, in 20 minutes, in at 1-hour increments.

So, well, you know who's optimistic about AI, John? Who? Friend of the show, Mark Zuckerberg. Oh yeah, Mark has launched a new campaign that is focused on AI optimism. Uh, in Axios, Meta CEO Mark Zuckerberg on Thursday laid out an optimistic view of the agentic future, arguing the company's focus on connecting the world will only be strengthened by new AI tools and technologies. Axios says why it matters: his position is framed as a stark contrast to some of Meta AI's competitors who, according to a video ad posted with Zuckerberg's comments, promote fear and a dystopian vision of the future.

Uh, I thought it was a cool video. I liked it. It's interesting to imagine watching it not as a fan of social media though, because it's leaning a lot on like, we've connected people, we bring people together. And I certainly see it that way when I log on to Meta platforms. I'm sending you funny reels and we're having a blast, but a lot of people see it as brain rot, right? And they don't, and they're like, oh, if you're gonna do what you just did to the next thing, I'm not certain, I'm not fully on board. So I don't know how it'll be received.

Yeah, I don't know how it's gonna land.

It's definitely the correct angle. The opposite is worse. Like clearly, like you wanna be optimistic and there's a lot of data to show.

I'm really glad that he's not like memetically going down the fear-based path.

Oh God.

Yes, that would be very, very rough working for some players. Um, anyway, so we'll see how it lands. Um, what else is going on? Paramount is moving into microdramas. This is the moment you've been waiting for. You love dramatic movies. I don't know, can I interest you in some microdramas? Have you tried any of the I watched the AI version. The real short, the true short.

Yeah, yeah, yeah, yeah.

I watched a true crime one and I felt like I was actually fine with the AI video and voiceover element of it. What I felt was missing was when you're tracking a true crime story and you turn on a podcast about that true crime, you know, whatever it was, I like being, led through a meandering idea maze of what is interesting to that host. So maybe they find the, the town and the details about the town where something happened particularly interesting, and so they're just putting in random facts that flesh out the story in a particular way. I felt like the one that I watched at least was very much like the Wikipedia level summary, or like the headlines just condensed, and it was too generic, and it wasn't giving me enough, uh, novelty and like new facts. So I— even though like the video wasn't quite there, it was obviously AI, it wasn't, it wasn't super visually striking, the audio was sort of, you know, jilted, uh, that all would have been fine if it had been someone who actually went and found some really deep, interesting, novel way to tell the story, or novel details that otherwise you wouldn't have heard of from the headlines. So I don't know, still early, but I don't think they're gonna get me. There's something about going to the theater. It's an experience. It's like, it's the anti-brand.

What if a theater were to turn the screen sideways?

They've done that. They've done that. There's, there's somebody in Brooklyn, I think, put together like a whole like, we're gonna watch Instagram Reels together, or like TikToks. It was like short-form film festival or something. Yeah, I mean, at some point you have to do a premiere. Uh, Kareem from Subway Takes was talking about that. He was like, You know, I've had this show, it's a massive hit, it's breakthrough. Like, where's my premiere? And like, maybe there should be a premiere for the next season of Subway Takes whenever that happens. I mean, it's also the hard thing with, uh, with social media shows is that there isn't really a season. It's just sort of always on. We sort of fake this with, uh, what are you laughing at?

Something I'm not going to read. Tyler, who's back in the studio, said, I've been watching this on Reel Short. I'm just not going to read the title.

Oh no, you can imagine it gets pretty clickbaity in there. You can definitely imagine that.

Uh, yeah, this doesn't look, uh, this looks like if Apple knew that this was on there, they might, uh, they might ask them to take it down.

Potentially, potentially. Yeah, I do wonder, uh, where the line will be drawn. There's a whole new, there's a whole new class of problems that Apple have to grapple with. I thought this hilarious. I thought this screenshot from var Epsilon was very funny, where it's just, I couldn't solve the Riemann hypothesis. I couldn't solve the Hodge conjecture. I couldn't solve P versus NP. I couldn't solve Navier-Stokes. This is just every really, really challenging Millennium Prize or unsolvable math problem. We've been working on our own conjectures. Conjectures.

We're creating the haze because all the conjectures— Cosgrove.

Yeah, all the conjectures are solved.

Well, I mean, we actually got to create new ones. We built it out, I think it was 2009. Yeah, I remember back in college. We've been letting it sit because we just didn't have the technology. Yeah, we don't have the capability. Yeah, we needed AI to advance. Yeah. And I think we're going to be close to cracking it. Yeah. But we should release the conjecture and let everybody kind of take their own shot at it.

Speaking of conjecture, there's a lot of conjecture about what OpenAI is releasing today. Tebow said unbelievably excited for what's coming together tomorrow is feeling Codex-y. Uh, people are speculating that it might be a Cerebris, a faster version of Sol. It's out. Turbo mode. It is. Okay.

ChatGPT Voice is now in the desktop app. Control your computer and direct multiple agents running in ChatGPT Work or Codex using just your voice. It's powered by GPT Live, so it can speak, listen, and coordinate work in the app at the same time.

And then also health is rolling out to US users everywhere. Health in ChatGPT. You can securely connect Apple Health and supported medical records to understand your information. What about sleep? Oh, got to get it on there for sure. There's also a new law introduced, a bipartisan AI kill switch bill following the OpenAI cyber incident. The bill enters Congress as lawmakers remain divided on how aggressively the federal government should regulate artificial intelligence. I'm sure this will be a longer conversation on a future show.

Um, uh, Shiel Monat is sharing some interesting nuggets in an information article on Stripe, which we will be getting to.

Let me tell you about Cisco next episode. Critical infrastructure for the AI era unlocks seamless real-time experiences and new value with Cisco and We have Lisa Su joining us in just a minute. We're about to bring her onto the show. And thank you for tuning in. Thank you for everyone who makes the show possible.

This is the moment we've all been waiting for. It is.

It is. And thank you to Ramp for making it possible. Time is money. Say both. Welcome to the show, Lisa. Thank you so much for coming on. We appreciate you being here. Such a busy day. Thank you you so much. Uh, please grab a seat. If you wouldn't mind putting that headset on, we will hear you loud and clear. Uh, how is today going?

It is a fantastic day. Yes. What have been the highlights? Well, you know, it's wonderful to see just everyone, you know, all of our community here. So customers, partners, um, certainly a lot of developers here, and You know, we're seeing a lot of excitement.

So take us back to 2014. How is 2026 different? Just a little bit different. Not just in terms of the AI boom, but as a leader, emotionally, is this a more stressful time because everything's happening so fast, or is this just a more exciting time? Like, what is your life like as CEO of AMD right now?

Well, I have to say, when I think about this entire arc, you know, 2014 until now, the thing that has been perhaps most interesting to me, it's like the world's like a different place. Like every year, like every 2 years, you see the tech trends are different. You see the customer, you know, sort of needs are different. You see the marketplace changing. And that's what just kept it really exciting for me because it's one of those things where like you're never going to get bored. It is certainly not less stressful, so I can tell you that. I think what's different about AI right now, I think the rate and pace of change is actually much, much faster. I mean, semiconductors has always been a fast-paced environment, but the rate and pace of change of the adoption curve of AI and then how the technology is changing and then frankly how the ecosystem is changing has, has us constantly constantly on the, hey, we got to go faster. Yeah, I mean, that's the, the primary thing is we have, you know, we're in a business where you make decisions, you know, 3 to 5 years in advance on your technology roadmaps. Yeah. And you realize, man, we can even go faster. Yeah. And that's what the market wants. What is—

and I think there's so much more external pressure in a way that the computing industry and the technology industry in general just hasn't had.

Well, I view it as actually a really good thing. I mean, people ask me that from time to time. And, you know, the reason I view it as a good thing is, look, you know, we're in a place where everybody needs compute. Yeah, everybody wants compute. You know, every country wants compute. Every hyperscaler wants compute. You know, all of us are using it. And so it is not, you know, we're the components under covers. We're actually, you know, helping drive, you know, sort of how technology is being consumed. And so I think that's the difference, right? It's, it's much more front and center, you know, versus just something that only techies know about. Yeah.

Uh, culturally, how is— what is the state of AMD's culture? Because there's one view where, uh, you're at the center of the AI boom, it's the best place to work. And then at the same time, things are changing, you have to move faster. There's a little bit of anxiety across every employee base around what my job will look like in a decade. How are things changing for the employees at AMD?

Well, I think I'd take a step back and say, hey, why do I wake up every day? What are people at AMD really excited about? And we're really excited about putting great tech out there. And I'm not kidding, you know, our mantra, like our number, number one mantra is, you know, build great products. And it's about how do we keep pushing the envelope on that. So you have a day like today or this week, which has been a culmination of years and years of work. And you think about launching Helios. You think about launching Venice. You think about bringing together the entire ecosystem. And you're like, it's days like today that we're working so hard for. And I think the culture is one of we want to be really the the best in the industry. I mean, our goal is to drive the future of technology, and, you know, we do it the AMD way. And what does the AMD way mean? Um, we do it in partnership, we do it in collaboration, we believe in open ecosystems, uh, we believe in using the best technology for each workload, you know. So this idea that, you know, one company is going to have the killer chip, I don't think that's the world we're in today. Right, the world where you say—

Was that always the AMD way? We had Travis Kalanick on the show yesterday, and he started off saying, you know, the Uber way, and then he stopped himself and he said, well, the Travis way. And he laid out sort of his approach. Was that something, was this approach something that you feel like you, the company had in 2014, or was that, or these sort of pillars, things that you felt like were important going forward, and you've added, added some over time?

Well, I would say what has been the foundation of AMD, you know, back to, you know, Jerry Sanders, our founder, has always been about pushing the envelope. And, you know, in some sense, uh, you know, being risk-takers, you know, wanting to have the best roadmaps out there, uh, that has always been the foundation of AMD. I think the last 10-plus years, you know, hopefully what I've brought to it, what Mark Papermaster, our CTO, has brought, what the leadership has brought, is a sense of not only are we going to put the best technology out there, but we're going to be predictable, we're going to be great partners, and we're going to do it in a way that brings together the ecosystem. And I have to say, everything about tech, like I was so happy to see, to have Tom Brown of Anthropic with us, Santosh of Meta, Sachin of OpenAI, you know, Jeremy at AT&T. Yeah, G2 was here. And what you kind of get from all of that is, you know, partnership is not just a word. Like, partnership is kind of our foundation. And I'm a big believer in 1 1 is greater than 3. So, you know, obviously we have to have great tech. Yeah. But we also have to be able to, you know, kind of, kind of see the future through the broad lens. And the way you do that is having, you know, great partners along the way. So that's very much, I think, the foundation of today's AMD culture. Yeah.

What were, what were some of the highlights of, of Tom's talk earlier? He was talking about how AI is actually speeding up their ability to adopt new hardware and bring new systems online.

But what were the highlights for you? Well, I'm super excited to be able to talk about it. You know, we've been— I kid you not, we have very much wanted Anthropic on AMD for a long time, and we just had to find the right intersection point, the right intersection point of our technology with, you know, where they are. I mean, you know, Claude has been incredibly, incredibly successful. And probably the highlight for me was, I think he told that story on stage about the MI355, and that's a very true story. It's a very true story.

And you've been busy today, but that's where what the internet's talking about.

Is that right? Okay, I, I haven't seen that, but I could tell you a few months ago, um, my guys were like, hey, we think Anthropic is, is on, you know, 355s and they're doing work on it. And I'm like, okay, well, you know, make sure you know what they're doing, make sure we're helping them. And they're like, Lisa, they don't really want our help, they don't really need our help. They're like, they're able to do it with Claude. And I'm like, wow, that's incredible, that's incredible. So he— I think he told you the, uh, the story, but it's just a lot about how the ecosystem has evolved over time, right? I'm a big, big believer in AI's incredible force multiplier. It's true across every industry, but it's especially true across our world, which is, you know, putting, you know, great technology out there. If we can reduce our time to market, you know, sort of the time it takes for from us to start an idea to when we actually finish and ship a product, if we can shave 3 months off of that or 6 months off of that, that has tremendous value for our customers. Like you heard, you know, Sachin say, like, more compute faster. Yeah, that is a frequent conversation I have— more compute faster. Yeah, and these are really complicated systems. I, I will absolutely say they're really complicated systems, and we need to make sure that, uh, we are able to put it all together. And that's where AI can be tremendously helpful on both hardware and software.

Yeah, you mentioned not focusing on just like the one single chip. AMD has a lineage in CPU, GPU, FPGA, many other systems. What are the advantages of having that breadth of product scope? And are there any disadvantages to that?

Well, I, you know, I think the interesting thing is, you know, every so often you hear like, this is a killer chip. Yeah, right. You guys have heard like GPUs.

You see that on the model side too.

GPUs are going to take over the world. Yep.

Like everything's going to be— CPUs are dead.

Yeah. We're going to move every single workload over. And the world just doesn't work like that. Yeah. And so our thought process has always been, you know, You know, people asked me very early on, "Hey Lisa, why don't you just focus? Like, you decide, like whether it's CPUs or GPUs, why do you need both?" And, you know, I think about each one of these questions deeply. And, you know, we fundamentally believe that there is no one-size-fits-all. Like, the world is a heterogeneous world. Like, you are different than I am. We have different needs. We have different workloads. Our businesses are different. You're going to need different compute. And I think the portfolio that we have, you know, CPUs, GPUs, we acquired Xilinx, so we brought the physical AI component in there. We have a rich, you know, PC ecosystem, I think allows you to truly pick the right compute. Now it's a little bit harder, right? We have a tremendous number of R&D priorities, but I think we've done it in a very you know, kind of, I want to say, smart way, in the sense that each part of our product portfolio builds on each other. Like, we talk about, you know, triplets being a big thing that we do. We do that across the portfolio. So we do that across our CPU portfolio, our GPU portfolio, and it, it just is an example of how we're able to leverage getting the right compute for the right workload. Yeah.

Uh, Talk about, uh, the other slice of the business, the other segmentation between, uh, consumer, prosumer, enterprise. Uh, how valuable, how important is it to, uh, have a clear chain of products in particular with, uh, talent? It feels like, uh, a lot of folks will start on a consumer rig and they might build some small model, or they might be doing CGI work or something else, and they're using a Threadripper to render something locally, and then eventually they move to the cloud, and eventually they start working in an enterprise. How is that going to evolve?

I think it is extremely important, like you said, because people access technology in different ways. Yeah. You know, not everyone is going to a, you know, big cloud instance to experience AI. And so, you know, the fact that we do have a consumer roadmap, a prosumer roadmap, our, you know, Ryzen, our Radeon, you know, frankly, we're putting a lot more emphasis on, you know, some of these other ways to access, you know, AI. So we made a big investment in AI PCs. I still believe that local AI is going to be one of the key enablers to truly get all of the tokens that you need in different places. So our Ryzen AI Max portfolio, I think physical AI is another place where, you know, we just launched these small robotics form factors so that people can experiment with it. So I do think it's quite important to have, let's call it, low barrier of entry. Like, I want people to experience AMD and have a wonderful experience. And they may end up, you know, being in, you know, the largest frontier model companies or, you know, frankly, the amount of startups that are doing just amazing technology is, is fantastic. And we want to support those as well.

How do you—

on that, how do you think about long-shot R&D? You have your existing roadmap, but we talked to new chip companies very frequently on the show. It feels like every week there's somebody new that has a billion dollars and they've, they're onto something and they're, and, uh, I can't imagine you haven't thought of it, or at least the AMD team hasn't at least thought of it. And so how do you think about, uh, how do you think about some of these investments in R&D that, that, um, are maybe like higher risk but, uh, uh, high potential over the long run?

Well, we are in a world where, uh, there is a long arc on R&D. Like, I really do like to say, you know, judge us on what we're doing today. The ideas were, you know, really thought of 3 to 5 years ago. Yeah, so chiplets, networking optimization, CPU and GPU, uh, putting that together in Helios. I mean, that's really unfolded over the last, you know, 3 or 4 years. And today, you know, we're thinking about, you know, what's you know, beyond, um, the roadmap. And I think both are very important. So we do have a tremendous number of new ideas, no question, you know, very active research team. Our team is working on, uh, not just MI500 today, MI600, lots and lots of ideas. We're spending time with our top customers talking about, hey, tell me where the models are going, what are you thinking the workloads will need, so that we can put that flexibility and capability into our roadmap. But I want to give credit where credit is due. There are a growing number of startups. It used to be that people only did software startups because hardware was too hard. Yeah, it took too long.

Yeah. And now it's easy, so everyone's doing it.

It is not easy. I'm joking. It is not easy, but it's appreciated. Yeah. That's what I want to say. Yes, it's appreciated. Yeah.

And part of this like software singularity that it feels like in some ways we're already in, VCs are just, they feel like more confident underwriting risk on the hardware side.

Yeah, I don't know that I would feel more or less, but I would definitely say that there has been a change. You know, people that didn't necessarily have the patience for hardware before. Yeah. Because it's a long arc on hardware. And now I think there's an appreciation that fundamentally, you know, hardware, you know, silicon software systems, when you optimize together, you're going to get a significantly better result. So my view is lots of good ideas out there. I think one of the things I pride myself on within AMD is we are quite, quite open to new ideas. You know, we've gotten to the place where I'm really happy with the acquisitions that we've done. We've actually brought in some incredibly talented individuals who have had a huge, huge impact on our AI roadmap. You know, certainly I think a lot of people know Anoush out there. So one of our acquisitions who is doing incredible job trying to make sure that everyone who wants to use RockM is going to get our support. Yeah, but a number, you know, our Pensando acquisition, our our Xilinx acquisition, our ZTE acquisition. All of these were to kind of bring the components in. Yeah, that help us, uh, build the full system solution.

Well, we know you're busy. Thank you so much. We have a gong.

We'd love for you to hit it to, uh, on this commemorate.

Do I get to do this?

Please, as hard as you want. Oh, there we go, right here. Thank you so much. And we have hardware companies, consumer hardware. They're hard, but chip in it. We have a D-Powered Chromatic from Mod Retro for the M64. Uh, you want to sign right here next to AMD? Perfect.

There we go. Is this one of yours, or is this, uh, it's Dylan's over there.

Dylan, all right, it's going in the museum. We're good friends of the team over there. We love, uh, love just consumer hardware and all these things. So thank you for powering it. Thank you for coming on and for coming on the show, and congratulations on everything. Thanks for having us. Thanks, cheers. We'll talk to you soon. Have a great day. Um, and that's our show, folks. Thank you so much for tuning in to TBPN live from AMD. Uh, we are traveling tomorrow. We'll be back on Monday, 11:00 AM Pacific. Thank you to the team here who flew up, and thank you to everyone at AMD for making this possible.

We really appreciate— dialing in these IRL shows.

Yeah, we're getting there. Uh, one step at a time.

The team makes it look easy.

Leave us 5 stars on Apple Podcasts and on Spotify. Sign up for a newsletter at tbpn.com, and we will see you on Monday, 11 AM sharp, for another one. Goodbye. Cheers.