SpaceX/Cursor Deal, Images 2.0 Reactions, Fake Bear Attack | Anil Chakravarthy, Naveen Gavini, Avlok Kohli & Ankur Nagpal, Joel Edwards, Renen Hallak, Darian Shirazi

You're watching TVPN! And today is Wednesday, April 22nd, 2026. We are live from the TVPN Ultra Dome, the temple of technology, the fortress of finance, the capital of capital. A bunch of major stories today. SpaceX and Cursor partnering up. More news out of Images 2.0. A bunch of news out of OpenAI and Mythos. A group of unauthorized users have been using Claude Mythos since the day it was released. There's a big scoop in Bloomberg. We'll go through. Bunch of timeline. We can also pull up the lineup and take you through who's coming on. We got Adobe, Build Forever, AngelList, Zancar, FastData, Gradient. We're going through all the news of the day. Well, uh, let's start with SpaceX and Cursor, who are teaming up in a very interesting deal. Is this a gong already?

This is gong-worthy because it's an option to buy the company, but a $10 billion breakup fee. Incredibly, incredibly— yeah, I think it's a win-win. It's a win-win. Yeah, it's a win-win. I think it makes sense.

It's a win no matter what. Well, let's go through the facts first. So SpaceX partners— pre-facts. Just immediate takes. We assume you already know all the industry.

Post-facts.

Do you know all the information? Do we need to give you any information? We'll see. But let's run through it. So SpaceX partners with Cursor to, quote, create the world's best coding and knowledge work AI. The deal gets Cursor, whose agentic coding model Composer 2 basically operates frontier-level performance, access to compute from SpaceX's million H100 equivalent Colossus supercomputer. And that is the correct term for Colossus. It is a supercomputer. There were some other terms that Elon was throwing out. Fantastic terminology from the xAI team over there. What was the other one? It was like AI—

Compute Gigafactory.

Compute Gigafactory. They're all hilarious and very good. I like all of these.

Computa. Computer data center. Were they saying computer?

I don't know. The xAI timeline, I was like, I was thinking about it and I was like, when did this actually start? Like, 'cause it just kind of came out of nowhere and just absolutely blew up. I wanted to actually review and reset on like the timeline of events here because it's gotten so crazy to the point where it's like now the thing that started as like Elon sort of being like, I wanna buy Twitter, is now like a Neo Lab with a massive supercomputer data center to coding agent and code review for the Age of AI, because don't forget, they own Graphite now, or potentially will. And a social media app.

Graphite, the space code review company.

Exactly.

I'm really, I'm genuinely so thrilled for the Graphite team. I'm thrilled for the Cursor team.

I'm thrilled for Scott Wu.

Yeah, Scott Wu is licking his chops. He's like, what are you going to leave out in the record?

He really is gonna have to live with it.

I also think we need to take one moment and just send some thoughts and prayers to SBF.

Oh yeah.

Imagine, I don't know how this works in whatever prison he's in, but I imagine, you know, hopefully every few days he can go and like talk to the outside world.

Yeah.

And you can imagine him going, you know, making the call.

Yeah.

And he says, you know, for the last few weeks, it's like, you know, sir, what would the mark be? How, how, how, how's Anthropic doing? How's my baby doing? And it's like, sir, sir, it's traded up north of $800. It's almost a trillion. And it's, it is quite possibly going north of a trillion. And then he's like, thank you. Hangs up, comes back the next day. I haven't checked in with Cursor a little bit. How are they doing? Sir, you may not want to hear this, but there's a $50 billion acquisition. They've agreed to sell for $60 billion. And if they don't sell, they're going to get $10 billion of non-dilutive capital.

Absolutely wild. Anyways, what a, what a, what a wild, wild turn of events. Anyway, let's go through the timeline because it's, it's interesting to revisit the flurry of news that has come out of Elon Inc. over the past just 4 years when this all started, April 14th, 2022. It's April 22nd now. So we're talking 4 years to go from just proposing buying Twitter. He made an unsolicited offer to acquire the company. He closed that acquisition after a bunch of back and forth saying, hey, maybe I don't want it. It was very clear that the stock would have traded down significantly from that $44 billion because this was when the interest rate hike happened and the end of ZIRP, and basically all software companies sold off significantly. And so we saw declines in Snap and Pinterest and basically anything that was— even Meta sold off like 40%, 50% post— although that was like an anomaly and they built back up. But the question about like, okay, well, Reality Labs is a what, $10 billion, tens of billions dollar bet on revenues that might come in like 10 years. Like it was so far away. And the revenue ramp on VR and metaverse projects was so slow that the market just had to discount those future revenues, even if they were still bullish on the idea of Meta winning the VR race, winning the metaverse at some point, it just wasn't gonna happen anytime soon. So you gotta discount that back at 6% instead of 3% or whatever your risk-free rate is in your DCF. And so everything sold off and Elon had lost locked in that pre-end of XRP price. And there was a big question about, could he get out of it? Were they gonna twist his arm? There was a little bit arm twisting. The arm was twisted. The arm was a little bit twisted, but Morgan Stanley came on board and a bunch of other folks.

It was almost broken.

But the people whose arms were twisted, Morgan Stanley and all the VC backers who came in, they wound up with SpaceX stock. And so they wound up doing very well 'cause they wound up with XAI stock and then SpaceX stock. And so it's all looking like, You know, as crazy it was, it was never bet against Elon. Like, that was the thing, was like, surely this is the time to bet against Elon. He's buying Twitter for $44 billion. It's a crazy idea based on the market and where things are, but everything sort of penciled out. So he closed the Twitter acquisition on October 27th, 2022, and he took control of the platform at the end of that month. That was Sink Day, of course. He comes in with a sink, something about the— what was the joke about the kitchen sink? He was like—

Let that sink in.

Oh, let that sink in. Okay, it wasn't like I'm doing the whole, isn't the whole kitchen sink another phrase as well?

Yeah, like throw the whole kitchen sink at it.

Okay, yeah, well he did that job 'cause he threw everything together. Yes, good job. Yeah, yeah.

Yeah, yeah.

So xAI came later. He publicly announced xAI on July 12th, 2023. And that's only 9 months after ChatGPT, maybe 8 months after ChatGPT. Google quick followed with Gemini and it was like, It was a very fast following, I think, to get to xAI off the ground.

And that was like, just for context, that was right around the time you first publicly talked about the potential for a tech livestream, right?

Oh yes, yes, I did. 2023. I think that was actually very close to the date. Yeah, I was making YouTube videos. I made a whole big YouTube video about the Elon Twitter acquisition, and I was sort of trying to justify it as a, you know, it was more about his desire for free speech than anything else. Like, you shouldn't look at it in financial terms. I think that still probably holds up. Holds. But there wound up being a whole bunch of other knock-on effects for Elon's strategy. Anyway, so he was, you know, coming out on July 12th, 2023 saying, "I'm back in the AI horse race. I'm competing directly with the big labs. I'm gonna go up against DeepMind still." Remember, that's why he founded OpenAI. He wanted to push back against DeepMind. "I'm gonna go up against Anthropic, gonna go up against OpenAI." And so the first major product milestone came Just a couple months later, November 3rd, that's when they introduced Grok, November 3rd, 2023. So it's been almost 3 years. The first real integration with Twitter happened on December 7th, 2023 when Grok started rolling out inside of X for Premium Plus subscribers, which was a new tier. It's actually crazy how many features—

Everyone got verified, remember?

Yeah, I mean, that year on Twitter was insane. Like just product feature, product feature, and they were shipping stuff that had been clearly developed beforehand. Like I think the community notes idea idea had been workshopped and even like engineered in the pre-Elon era, but he just got there and was like, "Ship that tomorrow." And like there were a lot of things like that that happened. And there were other things that got pulled back and review and like there were other pieces of the puzzle that were like not doing so well and he was just like, "Cut, cut, cut." And so he sort of like put it back in startup mode and it felt a lot more agile and it still does. And so that was the moment Grok finally became a part of the X product experience. In early 2024, xAI started shipping updates quickly. March 17th, 2024, 2024, it open-sourced Groq 1. March 26th, 2024, Groq access expanded to X Premium subscribers. 2 days later, they announced 1.5. Then on April 12th, 2024, they announced 1.5v, which added multimodal capabilities. Then Groq 2, Groq 2 Mini. By the end of 2024, Groq was available to all X users. And so 2025, xAI moved from being connected to X to absorbing it at the product level. On February 19th, 2025, xAI announced Groq 3 beta, then March 28th, 2025. So over a year ago, for some reason this feels more, more recent than that, xAI acquired X in an all-stock deal. So March 28th, 2025, effectively merging the AI company with the social platform. After that, X kept releasing faster model updates. They did Groq 4 Fast, Groq 4.1, then came the SpaceX deal. That was February 2nd, 2026. So the clean sequence is that Musk first proposed the Twitter acquisition, then founded XAI, then brought Grok into X, then merged XAI with X, and then SpaceX bought XAI. And so now Cursor is joining the team and—

Yeah, and this just makes, this just makes, it makes so much sense, right? Cursor needs compute, they need the resources, they need the capital to train a frontier coding model.

They've also never done a pre-training, I believe, whereas the XAI team has, right?

Correct. And the big, big thing is that the Groq brand has been through so much.

Sure.

There was an idea being thrown around last year that it had been banned in more workplaces than it had been adopted.

Yeah, yeah, yeah.

So more people had said, you cannot use this product in the workplace than were actually using it in the workplace. Cursor's a great brand, right? It's a brand that I think can probably expand outside of— outside of coding, right?

Yeah.

Say in the announcement, best to create the world's best coding and knowledge work AI.

Yeah.

So we're at a point right now where everyone is building the exact same thing. You got everyone on Earth building a box that you can tell to do things and it does things.

Oh, so you just want one company to do it all? Communist?

No, no, I don't. I think the competition is great.

You want the government to have it.

It's bringing out the best in It's bringing out the best.

It is.

But I think Cursor's like—

Drag out fight.

Cursor's always had a great brand.

I saw someone else— Wait, SBF replied to SpaceX's tweet. So that's me. Pull up, pull up the first post here and then scroll down. The SpaceX and Cursor are now working closely. That, that, that, uh, it should be like the first one in the, in the timeline. SpaceX and Cursor— SpaceX AI is what they're calling it now. Is it not up there? In Cursor, I can find it and send it in. Let me find this post and send it into the production chat.

Tuxedo Sam says, do you think they'll make Elon take his shoes off at the Cursor office?

Okay. So Richard Wu has some context. There you go. Look, SBF is somehow on Twitter. How can you be on Twitter from jail?

All these people in jail on X. I don't understand. No, it says it in his bio. Yeah, it says it in his bio. It says we can use BOP-approved phone calls, emails to tell others what to post on our socials.

Okay.

I think it would be a free speech violation to say like, if you are in jail, you cannot distribute information in the outside world in any capacity. But it feels like this was maybe some legislation that was created pre-internet. Yeah, it seems like having like a ghostwriter on your account is a really good way to like start, you know, kind of like influencing public perception, right? I think this is— Elizabeth Holmes has had enough kind of moments where it was like she, you know, her proxy posted something that was mildly kind of entertaining. And over time that just wears on people. And eventually people are like, you know, maybe, maybe, maybe she isn't so bad after all, right? So.

Okay, well, Richard Wu broke down the structure of the deal. He says the structure of the deal is pretty interesting here. I think what's happening is, one, XAI is having trouble training a state-of-the-art coding model, hence co-founder departures. They might have a bunch of idle GPUs. Cursor doesn't have capital to blow on a $5 billion training run to compete with Codex and Claude. XAI—

3—

XAI says to Cursor, Use all the GPUs you want at cost and get to a state-of-the-art coding model as long as we have the option to buy you. 4, Cursor also gets a free option. Train a model better than Opus and get bought out for $60 billion or get $10 billion that pays for all the GPUs you rented. Win-win. And I was reflecting on Michael Truel's, I mean, fantastic entrepreneur, remarkably young. But his aesthetic is very much in like the Stripe world, I imagine, because I feel like the one really cinematic podcast he's done is that one with— was it Patrick Collison? Is that the one I'm thinking of? And it's like them at the coffee shop, and it's so welcoming and warming. It's like, that is welcome in every corporate entity in America, right? Like, and to your point about like, you know, oh, like if your corporation is like, oh, what's going on with Elon and politics, blah, blah, blah. Do we really want Grok running around with Mecca and Annie and all this crazy stuff? But if it's just like Michael Truel from Cursor, he's so reassuring. Like, if I'm like Coke or GE or Ford, I'm like, yeah, of course, like, we love Cursor. That's great.

Yeah.

Very, very— aesthetics do matter. Vibes do matter. And I think it makes a lot of sense.

Matt Slotnick says he loves math.

Okay. What is the math here?

What he's talking about, Julian, he's— so Julian says at a 30x revenue multiple, at first glance, it appears that SpaceX is overpaying for Cursor. However, the deal deal is wildly accretive for SpaceX given it is expected to go public at more than 100x revenue. A 70-turn multiple expansion on $2 billion of revenue adds up. Wow. So, yeah.

This is a notorious thing in corporate M&A, and occasionally it's gone poorly, but there is like an old adage that if you can acquire earnings at a lower earnings multiple and maintain your current multiple, that is accretive to the stock. And so there's a whole philosophy around that, but the smart investors should, price each earning stream differently and having like a monopoly on launch capacity should be a higher multiple than a super competitive oligopolistic coding market, if that's what this winds up being. It's a tricky, tricky situation, but—

Will Brown, wow, Cursor just hit a $10 billion run rate.

They did, it's guaranteed, right? There's no way that they won't make $10 billion this year. That's not even a run rate, that's just like, it's locked in, it's more than contracted.

I don't know. Yeah. And it's very meaningful because it is effectively an exit. And then it's, you know, by itself, they can, you know, they'll be in a position to actually reinvest that. That's like a nice little Neo Lab kind of Series A, basically.

But Ken says, no, you got to multiply that $10 billion by 12. Because in that month where they get the $10 billion check from SpaceX, It will be $120 billion run rate. They will be the largest AI company in the world that month. And then they just have to figure out, they don't collect until end of year if they're still independent, otherwise it's more like $25 billion. Okay, well, so Fleet Invest has some other thoughts. Where you are?

What is this post from Patrick?

Wait, wait, which one, which one?

This is, Patrick is quoting. Scott Wu.

Oh yeah, Last of the Mohicans. Scott Wu. I love it because, uh, despite this does not look without images too.

Without images too, we just— this, this kind of asset, I actually think this might be done the old-fashioned way. There we go.

It's good at cutting out.

Thank you, watermark.

Yes. Uh, he says, here we go again, Last of the Mohicans. Uh, yeah, Scott was the last one standing in this particular category. And yeah, he's licking his chops thinking about who might want to hop over to Cognition, Cursor, Cognition, Windsurf, Devin, Cursor, Zombie Corp as the ghost ships continue to line up on the shores of Scott Wu's territory. What else should we go through with the Cursor? Cursor's lots of compatibility was contingent on—

I think that's it. I'm very happy for everyone involved. Yeah, uh, some news from Space News Inc.

This is very interesting.

China backs orbital data center startup with $8.4 billion in credit lines.

They're pilled. They're pilled.

Elon is interested.

$8.4 billion in credit lines, that is a lot of money. A Beijing-based space startup has secured early-stage funding— early-stage funding to the tune of $8.4 billion. What are we doing here? Uh, and extensive credit backing is part of a broader Chinese push towards space-based computing infrastructure, Beijing Orbital Twilight Technology Company Limited. We don't know how to name companies like that in this country, seriously.

This is the new meta.

Yeah. What is Cursor's real name? AnySphere? That's a pretty good name. I like AnySphere and I like Cursor, but doesn't hold a candle to Beijing Orbital Twilight Technology Company Limited, also known as Orbital Chengguang announced the completion of a pre-AI funding— pre-A1, or is that pre— pre-A1 funding round April 20th. The round saw participation from venture and industrial investors including Hisong Capital, CITIC Construction Investment Capital, Cathay Capital. They got everybody. It's a murderer's row over there. At the same time, Orbital Estate said that it has obtained strategic credit lines totaling $8.4 billion from 12 major financial institutions, including the Bank of China. Are they gonna build rockets with this? I mean, I don't understand why this would be so capital intensive if they're not gonna fab the chips and they're not gonna launch the rockets.

Well, they're gonna put a lot of GPUs in space.

I guess they're just gonna buy a lot of Huawei chips or something? Are they gonna go to Huawei and get some special stuff?

Yeah, I mean, this is pretty interesting because I feel like, you know, generally the main bull case for space data centers is that, the like regulatory environment in the US is going to be so hard to build data centers.

Yeah, yeah.

Like, you know, supercomputers. But yeah, normal land. Oh yeah, yeah, supercomputers. So you got to send supercomputers to space. Yeah, because there's less—

we can do data centers in space. We just— we, we— if it's on the ground, it's a supercomputer.

Okay, yeah, that makes sense.

Because, because we don't want any more data centers on Earth. It's Earth Day, by the way. Congratulations, Earth. You've done fantastically.

Yeah, but the whole thing, like in China, it's like they, you know, you can just like build things there, right? There's like very little regulatory overhead.

Yeah.

So it's like, what— like, I feel like the space data centers, yeah, a lot more sense for the US.

Yeah.

If you're worried about— if the, if the bull case is regulatory, which seems to be— that's like generally consensus.

Yeah.

I was laughing about how, you know, how there's all this fear about like data centers using water in America, but over in China they have the Three Gorges Dam, which generates electricity from water, and it technically uses a lot of water. The water is not destroyed or anything, it just passes through the dam, generates electricity. But they're using—

they're using an ocean of water.

Exactly. The Hoover Dam, that's a nightmare. If you just like— there's one frame where you don't want the water to be destroyed or made unpotable so you can't drink it anymore, but there's another one where you think like water has individual rights and should not be used at all. Like it should not be used, it should not flow through a water wheel, it should not flow through a dam and generate electricity at all. It should be left to its own devices still and just chill, maybe. Anyway, Orbital, has incubated— was incubated by the Beijing Astro Future Institute of Space Technology, which itself is backed by Beijing's Municipal Science and Technology Commission and the Science Park Administration. The institute leads a consortium of 24 organizations. Uh, the rationale for the constellation in November briefing: so large-scale data centers have expanded rapidly worldwide, but further growth faces major obstacles including heavy land use— weird— Soaring energy consumption, weird. Limits on atmospheric cooling, these things usually don't apply in China, but maybe they are skating where the puck is going and maybe they're thinking that they will need to change their direction. So they're planning a dawn-dusk orbit about 700-800 kilometers above the Earth aiming to achieve a large-scale space data center to support space-based computing by 2035. Wow, thinking in decades over over there. So they're going slow. They're getting there. An initial phase spanning 2025 and 2027. Wait, why are they talking about last year? We'll focus on core technological changes in the first computing—

They're manipulating time, John.

The experimental satellite was slated for launch in late 2025 or early '26, but it does not appear to have launched. They have, they have like decent launch capacity with the Long March rocket. I think that they just are not amazing at landing it, but you got more money, you can just YOLO more rockets up there, I guess. Problem solved. Although I would be surprised if there's as much pressure to not build supercomputers next to the Three Gorges Dam. Anyway, moving on. More images out of ChatGPT Images 2, ImageGen 2, ChatGPT Images V2. These are crazy. Has anyone done that rune test of the, of the The guy driving the car into the, you know, the whole thing where he was like, it's AGI if it can understand this meme. I wonder how, I guess you can't because like that particular meme has been saturated on the internet. And so it no longer is a challenge.

I'm gonna have it make it make that. This is a crazy image. I am a tiny man.

What is this? MySpace.

My son was born.

This is truly never.

They handed me to him.

Never forget where you came from. Blink-182. It really packed so much stuff in here. It's almost too much. I have been noticing the— Gabriel over there is pushing the model into chaos and seeing what happens. Cows are flying, horses are flying, trampolines are flying.

The next one by Ethan Mollick has like every image benchmark in one.

You were asking about this. You said— and explain the history of this.

So originally, like, when like DALL-E first came out, the like image that everyone was like, oh, this is crazy, it was an astronaut riding a horse on the moon.

Yeah.

So then, it's like, "Okay, that's like very easy." And then it became—

Well, it's very hard to Photoshop that. It's very hard to go get an actual picture. Sure, yes. Because if someone's just like, "Make a picture of a dog," and it's photo real, everyone's just like, "Oh, who cares? We have a picture of a dog." We don't have any pictures of astronauts riding horses on the moon.

Yes, but there's people riding horses, right?

Yes.

And so, like—

It should be able to process.

There's a sense where it's like maybe it's not like— it doesn't need to fully understand like everything that's going on because there's a lot of like references it can pull from. So then it became, can a horse ride an astronaut?

Yes.

Right. It's like the reverse.

Yes.

And early image models would always just do the flip. They would just put the astronaut on the horse.

Yep. Because that's more logical. Yeah.

So there's a number of these things where, like, if you ask a person to draw it out, it's like, okay, you just think logically, like, this is how it would look. But there's just like no reference images online. So another one was like a full wine glass.

Yeah.

So you think of a full wine glass as like it's only like technically, you know, halfway, three-fourths full.

I never understood why that one didn't work. I saw the whole video.

There's just a few images online of a full wine glass being like—

Oh, full to the brim because people typically fill them halfway. And so in the training data. Okay, got it.

Yeah.

Yeah.

And then there was the, you know, what is this image?

This is not the one.

This is not the one.

This is not the one from the timeline. That one's way more aggressive than the— this is the one. This one looks aesthetic and it still looks weird.

The only thing is, is it— would a horse's belly really look like that?

It is sort of like a humanoid horse.

It kind of looks like a human that—

But it does have the wine glass completely filled and then explain the clock.

The last one you pulled up was Grok, I think.

Okay, explain the clock. What's going on with the clock?

Yeah, that one's just another like understanding thing where like, can it, if you give it a time, can it do the clock?

Yeah.

Like the correct time. There's also one where you do a bunch of clocks on different time zones.

Yeah. —and it couldn't do the clock. It couldn't do the clock. It has two hour hands.

Yeah, this one.

You know what that means? We're moving the goalposts. Do it. Moving the goalposts. The goalposts.

Do it. It's time. It's time.

It's time to move the goalposts. We gotta, we gotta one-shot this. It's 99% of the way there, but it's not, it's not complete yet.

It's not complete. You guys have more work to do.

No, no, Sam was talking about this with Ashley Vance. He was saying that like, you know, he thought the job was finished with like ChatGPT images because it was really good. And then they worked a lot harder and they realized that there was more to do. And this is like the Carpathy thing about like, yeah, you get the self-driving cars to 99% and then it takes another year to add another 9 and then it takes another year to add another 9 and then another 9. And it just takes forever to improve these things because we demand perfection. We do not accept 2 hands on a clock. Keep grinding, folks. Keep grinding. Anyway, there are a lot of other fun posts. Semianalysis. I thought this was just a real image of Dwarkesh taking a selfie and then it was recontextualized via a meme. It is in fact an AI image and it doesn't even look like the new model though. Hi, I'm Dwarkesh. I grew up all over the US and now SF-based and always down to nerd out about AI, science, and history. A little about me. I host the DarkHash podcast, study at UT Austin, just published a book on the history of AI scaling. Let's grab coffee or do a fun activity this summer. Because of course, this is a meme template that is very popular and has gone viral with a lot of new people that have been hired and moved to San Francisco.

So the other thing I've noticed with images too is that I think it has fully, the new model actually has taste. I agree. And it has fully democratized like high-quality lifestyle and product imagery. Yeah. And I'm shedding a tear just a little bit. Yeah. Because you used to be able to tell like the kind of somebody who's working on a CPG company, you could kind of clock their ability. Yep. As a founder based on the quality of their images. Right. Because maybe they're not super creative. Themselves. Maybe they haven't raised a bunch of money, but if they're scrappy, they can meet the right photographer, they can put stuff together. And now it's like everyone just has a great product photography. Yep. And I saw it this morning with— there's that Andreessen Back company that does like electric scooters. Oh yeah. They were just like ripping out a bunch of images that look like they spent tens of thousands of dollars doing them, but they're clearly like images too, which is cool. It's just like funny that we've entered a world where— Yeah, I've seen some people.

I think that the end result will be like you will see more opinionated and creative and more people— like there will be a collapsing around like everyone will copy Apple or linear. But then on the flip side, you will see people that are doing things that are really unexpected and those things will be copied. But if the brand can run through and actually establish itself as like, "Oh, it has this unique aesthetic." it won't matter that somebody can recreate it. 'Cause plenty of people did that with the red antler stuff. It was like there were brands that really owned that and carved that out as like their aesthetic. And then there were a bunch of copycats and no one really liked those. I was reflecting on the fact that like for a long time Midjourney felt like it had a very unique aesthetic, like ArtStation and like painterly and sci-fi really well. And then ChatGPT images, GPT-1 and some of the other image models just felt like stock photography. And now I feel like I'm starting to see more stuff out of images with ChatGPT that feels more opinionated and has a stronger aesthetic. And it can do more of like the sci-fi stuff. Although sometimes it leans in a little bit too much to the photorealism. I think you gotta get kind of crazy with the prompts to actually get something that's abstract because I was taking some Midjourney prompts and I was putting them into ChatGPT images and I was noticing —like, it was just putting out stuff that was like—it was like—like, the prompt was like, make a video game of, like, a car racing. And it just looked exactly like a video game, because it just looks like a screenshot instead of, like, the idea of a video game.

Okay. Last one we're going to pull up, and then we have our first guest. Yes. Post from Ben Hilak. It says, nightmare blunt rotation. Oh, yeah.

It can do 360 images. They can do 360 images. 360 images, which is really cool. You got Sam Benson. So it generates a full equirectangular image, which is something you could view in VR, and then, uh, and then you can load that into a, uh, uh, panoramic, you know, uh, stereoscopic image, uh, generator. Uh, I wonder, I wonder when 3D images will come. And also, I mean, the, the big question is like, uh, people are going to want to animate these. What What's the downstream toolchain for actually turning this into video? Are we going back into video at some point? We will see. Anyway, people are having a lot of fun with it. And it should help with like front end and a lot of other stuff. It seems like it's done very well on design and layouts and all this stuff. Well, we have a perfect person to talk about AI imagery, AI, generative AI, creative tooling, and more because we are joined by Anil from Adobe, where he is the president. So let's bring in Anil. How are you doing? Welcome to the show.

Thank you for having me on. Great to be back on TVPN. And congrats on your acquisition. Great news for you. Thank you.

Thank you. It's great to see you.

Great to see you again. I would love to just get a general update on what has changed with Adobe. What have you been working on since the last time we spoke?

Yeah, look, I'm coming to you from Adobe Summit. We are here in Las Vegas in a big show, 14,000 people. You were just talking about Jensen. Jensen was here on Monday and really talking about how NVIDIA and Adobe are working together. And you were just talking about 3D. I caught the tail end of your comment there. And that's actually one of the big things we announced is how do you have digital twins so that you can take the digital twin and carry it all the way from product design into marketing and customer experience. So, you know, with HP, for example, they have 15,000 new SKUs every, every year with all the products they have. And they're working with us and with NVIDIA and to bring it together into, into marketing. So we are super excited about, you know, what we call customer experience orchestration, which is how do you use AI? How do you use all the software you have to deliver personalized experiences? How do you bring the right content? The right data about the customer, and then put that into the channel that they care about and really get the most out of the customer lifecycle, build loyalty. That's what we're— that's the big topic of the conference here.

Yeah. Walk me through that HP example, because if they have a SKU, I imagine that there's a CAD file at some point, and then they might want to instantiate that in text or imagery or video or 3D rendering, and they might want to go over to one of your 3D products or bridge to another. How important is Adobe at the center of like asset management and actually creating some sort of abstraction on top of what the actual core item is versus just translating from one sort of output to the next in the way that someone might take a PNG from Photoshop and drop it into Premiere Pro.

No, that's exactly right. So they start with the 3D image that they have, or the 3D, as you said, a CAD file. Yeah. From that, what we do is we create the digital twin. Okay. And then we apply the brand intelligence on their behalf. So in HP's case, for example, if it's a new PC that they are releasing, they have all kinds of things on what they want when they are It's a laptop and the laptop is open. What's the kind of image they want on the laptop? What's the kind of lighting they want on the laptop? And then they release it all over the world. So there's pictures of people. They always want to show a business professional in the laptop. That's the— along with the laptop. They want them showing them using the laptop to do something. So they have all kinds of brand guidelines, of course. But then there's a lot of a lot of tribal knowledge on what makes an ad successful. It's not only for them. Then they have to get all of that traveling through the entire marketing campaign because one of the big problems has been, you know, in the current world, they use the 3D CAD files for everything in manufacturing and so on, but the marketing, the content starts with a photo shoot. They're not actually using the 3D CAD files. Profile that they have the source of truth. And that's what we're enabling them to do. Use the source of truth so that the images are completely high fidelity, they're brand-preserving, but then you can generate everything around that. You can generate the background, you can generate the foreground, you can generate the users, and then of course you can do what we call transcreation, which is the translation into different languages. All of that can be generated based on their brand guidelines and what works for their brand.

So is your relationship with NVIDIA that you want to marshal enough compute that you can be delivering inference to a customer? Because I remember using Content Aware Fill back in the day, it ran locally on my device. As the models get more compute intensive, they eventually won't run on device, they'll run in the cloud. I know you have API partnerships, but it is their vision to deliver inference deliverance to the customer all in one go. Exactly.

It depends on the campaign. Let's say HP is running a back-to-school campaign, then they want a set of imagery, they want a set of things that they can transfer over to their channel partners, along with accurate product data. Or if they're running, for example, a holiday campaign, a different set of imagery, and so on and so forth. So depending on the campaign, depending on the geography, depending on on the customer segment, making sure that they have the right campaign, that is obviously the content, but it's also the product data, making sure that that's super accurate because it's got to match what exactly the customer's buying. And then anything that's relevant for the particular channel partner, it could be through their own direct channels, but it could equally be through CDW or through other channels, and they want to make sure that the entire thing carries over. So that's what we're doing for them is not getting the workflow right, getting the Adobe Brand Intelligence, which is one of the big things that we announced this week at Summit, getting all of that together so that they can have a single seamless marketing workflow and get the benefit of both AI and the 3D technology together.

Jordi Hays, you guys announced a big buyback this week and, and that, that's been exciting. What do you think you need to do from a results standpoint to show that you guys are an AI beneficiary versus a, versus a victim? Right. Because the narrative is obviously against you. Yeah, the market has, has spoken. But at the same time, AI is a powerful creative tool. You guys have massive distribution into, you know, millions of individuals and businesses. And so I think people are waiting to actually see, see the story in the numbers and there's like only so much you can say to kind of like give people confidence. But how are you thinking about it? Yeah, exactly.

As you said, we announced a $25 billion buyback. That is a signal of our confidence of, you know, this is going to continue to be a very profitable business for years to come. And we wanted to make sure that everybody in the market knew that. And, you know, look, I think to your point, First step for us is AI adoption and making sure that AI is delivering real value, whether it's to individuals and real value to users or to enterprise customers like we were just talking about. And for us, what that means is how do you really embed it into the tooling so seamlessly, whether you are a marketer, whether you are a creative, or whether you're a business professional using Acrobat, that AI works for you flawlessly. Honestly, and it does what you want. I love the example from the last show that you had, which is sometimes you do want the horse riding the astronaut. That was an interesting concept. But, you know, that may be— we have a lot of customers like Home Depot that just want sawdust and nails, and that's what makes their campaigns authentic. Sure. So depending on what the customer wants, depending what the user wants, making sure that AI is helping them get value, not AI is getting in the way. That's our focus right now is just making sure that customers, whether they're enterprises, individuals, small businesses, students, understand that AI is critical to the next generation of, of creative platforms and marketing platforms and tools. But it's in the context of what they're using. It brings the UI, the agent, the agents and the reasoning, the models and the data all together to work for them. So that's— I think we've made a lot of strides on that front here at Summit over the last couple of years. I think as, as the results show, one thing for us is I think from where we are in the market, for the market even to see us continuing to the performance that we've had, we're a double-digit grower. You know, we have very healthy margins. Even continuing to see that for the next couple of years, I think will change a lot of minds. And then I think We also see the opportunity through customer experience orchestration to reach a real inflection point. So that's, that's what we're looking to do.

Jordi, you can— Yeah, I was going to ask, how are you? What's your M&A outlook? It seems like there's a ton of really talented teams building, building creative tools that could, could fit into Adobe. Sure. Hopefully, you know, you're spending a lot on the buyback, but hopefully there's still some dry powder. For some talented teams. We do. We have plenty of dry powder.

We haven't overextended ourselves. In fact, we are really pleased. We just announced that we have received all the regulatory approvals for SEMrush and we are in the, in the, in the waiting period to close it. So in the next few weeks we'll be closing SEMrush. We're super excited about that because SEMrush, along with our portfolio, solves a real problem for marketers, marketers. I mean, we heard a lot from customers at Summit. Every brand is like, how do I show up in the right way through OpenAI, through ChatGPT, obviously, but through Claude and Gemini and Perplexity and Groq? How do I— don't even know how consumers— they know consumers are going there in droves. What they don't know is what are they even prompting for that I should be I should know about. If I'm a brand in the cosmetics world, what are they actually prompting? I don't even know what they're prompting for that I should be aware of. Then, how should I make sure— what do I need to do to make sure that I am included in the response? Then, what do I need to do to make sure that I'm included in the response in an accurate manner? Then, how do I get that traffic coming to me? Or how do I close the sale right there if I— if it's something that I can close right there? This is a new world for marketers. They have— marketers have no idea how this whole thing works. I mean, they don't know, for example, what happens when you put in a prompt into ChatGPT. What is the queries that get kicked off? Where does it get kicked off to? How is that information assembled? How does it impact their brand. They don't know the first thing about it is what they're telling us.

And they're saying, "I don't be helpless." I'm interested to see how this category plays out because there's obviously a bunch of companies doing, they call it answer engine optimization, AEO, or GEO, whatever they're calling it. But the big question was always, this feels like a very natural thing for someone like a Semrush to enter in a big way. They already have the distribution, so it'll be an interesting battle. Battle between Semrush and some of the upstarts like Profound and some other players.

But I think the battle will be over pretty soon, according to me. Yeah, there we go.

I think James— I think James at Profound, we should have you guys both on and go at it. That's great.

I want to talk about video. So, I mean, we saw OpenAI pulled back from Sora. There are some really amazing models that are happening. Some of them are in China. There's a whole bunch of IP stuff. But, uh, what I'm interested in is that, um, uh, Photoshop continues to be an important tool even in a Gen AI world, uh, because there's so many fine details. Having Gen AI is just one possible tool is valuable. The models are getting better. We're 3 years into this boom, but video is moving slower. It's more inference demand. And I'm wondering about how you're taking lessons learned learned and strategies that you applied to the evolution of Photoshop, how that integrates with different model providers, different tools, different in-house solutions, and how you're planning to bring that to After Effects and, and Premiere. Because this feels like a place where you would need even more opinionated editing. Like, we're nowhere near text box editing, like, you know, to like 40-hour a video into a movie. Like, that's just not— like, we're so far from that.

Exactly. So look, this is where we believe Adobe Brand Intelligence will be a huge differentiator because that's what then gives us the right information to pass on to the creatives to say, this is how— this is the kind of how you use Adobe Creative Agent and this is how you want to let the AI take over. And this is where you want to be opinionated. I mean, what we have done is provided that flexibility. We know we announced over 30 integrations now, you know, whether it's Runway or Flux and obviously Google with both Veo and Nanobanana, etc. So we've been very open and we're very open to integrating any other models that come up as well. The key, though, is the creative themselves. If you're working for a brand, I mean, if you're working for— I mentioned Home Depot earlier, if you mentioned working for Coca-Cola and so on. They don't have enough information by themselves unless they really get a really super detailed creative brief on what exactly to— where they need to take control, where they can let the AI run. Yeah, we believe we can automate that and provide the right information at the right time to the creative. They can bring the brand idea, they can bring the hero asset, they can bring the creative idea, and then you can apply the Intelligence and exactly to your point, you can be a lot more opinionated in what you produce.

Yeah, yeah. It is very fascinating that we got like fully generated video before we got a tool that puppeteers Premiere Pro in a way that can sync up music and edit to beat and make cuts just on normal non-AI generated footage. But that's just the nature of computer use and whatnot. So I'm sure it's an exciting time. Lots to build, lots to work on. Congratulations on all the progress. Yeah, great to get the update. Yeah, thanks so much for taking the time.

Good to see you. Thank you for having me on.

We'll talk to you soon. Cheers. Have a good one. Thank you. Goodbye.

Up next, we have— One quick thought. I think right now, already, the great image models, the Nanobananas, the images to— it is like you're just talking to somebody who is a digital guy or girl who is very good at Photoshop. Like, that is the experience today. And it was not that way 12 months ago. It is that way now. I no longer need Photoshop. I don't need to go to Fiverr to say, like, hey, I need somebody that's good at Photoshop. I don't need to text, like, Ben and be like, hey, can you help me Photoshop this? And you can just actually just prompt your way. You do not need any UI anymore. That's true. That's true. For the average, for the average. For, for, yeah.

I mean, we are, we are crossing into the future.

And I'm just saying that's gonna get, that's gonna get there with video where you just have a timeline. Yeah. And you can click on a scene or a segment.

But speed is really, really important here. Like, like if you, if there's one word that's wrong, on image and you need to regenerate it, you're going to wait 45 seconds and it will be faster to have done that in just a traditional workflow. Now it's getting faster.

It's so fast now. It's very fast. It's so fast.

But if you're editing a 2-hour podcast and you say, oh, I want you to cut to Jordi Hays in minute 45 and it has to go regenerate the entire thing, like that is slow. You're going to need to puppeteer the tools. You're going to need tool use. In here, I think. I don't know. Just because the—

I was just shocked because I was playing around with generating product photography. And you're able to say, put the text here, turn it around, put it at a different angle, more shadow, less shadow, all this stuff. And it's wild what you can do without any UI. But without further ado, we have Naveen from Build Forever.

He's the former Pinterest CPO. Welcome to the show. How are you doing?

Hey guys, thanks for having me. Thanks. I like the bomber, Jordi. Nice piece.

Thank you, thank you.

It's a one-of-one, uh, New York Stock Exchange. Oh, NYSE. What date was the New York Stock Exchange founded? You tell me. 1792. 11 Wall Street. Wow, that is a very old—

great address. Great, great year to get into business.

Great year to get into business. Best day to get into business. But it's great to have you on the show.

Yeah.

First time on the show. Please introduce yourself and the company a little bit.

Yeah, totally. So my name is Naveen Govini. I'm the co-founder and CEO of Build Forever. And honestly, we're just a group of engineers and designers and people that love building products that people love to use. And we've worked, as you said, I was early at Pinterest and we worked on a number of consumer products that people use around the world today. But our focus has been on a product that we launched yesterday called Extra. And it's really a reimagination of email as your life's inbox. And so for many of us, we spend all day kind of working, sending email. And when we have time to get to our personal email, it often feels overrun, overwhelming, and just like, I don't know, I personally gave up on my personal email. I had over 250,000 unread emails. And so we were just kind of looking for a solution of just making making email more manageable, more fun and delightful. And so with Extra, it keeps track of what's important for you so you never have to miss something. It helps you clean up your email, organize it around your life, and it does things proactively for you in the background and helps you get stuff done. And so this endless to-do list feels not so overwhelming and it just feels delightful and joyful to open up your email every day.

Is it a For You page for email?

Yeah, essentially. Yeah, that's how you could think of it. That's a good analogy. Yeah, so everything that's important in your life—

We turn every email that you have into a short vertical looping video.

It's not quite that, but you're wise enough to understand what I'm talking about.

No, I know what you're saying. It's basically like, yeah, just actually picking and choosing. You know, it's always been so silly that inboxes were primarily based around who messaged you last, not based around what was actually important or actionable or anything like that. Yeah, and I feel like we need to move— like, the idea, the whole idea of inbox zero is just such a pre-AI idea where you are just getting flooded with more and more and more and more and more spam. And I think products that can help here are great. Also, uh, it feels like you can kind of just Yeah, I'm curious. So you're built on top of Gmail. I don't know if you're supporting other email providers. Google is, like, aware that AI can be powerful in email. They're working on a lot of features. But I feel like this is one of those things that is going to require, like, really, really rapid iteration. And so I feel like you probably have this beautiful window of, like, maybe 3 to 4 years where they're kind of just slow to figure out some of this stuff, and you can probably build a cool business in the meantime. But where do you want to— where does this go? How does this— how are you thinking about this as a venture-scale opportunity?

Yeah, totally. Well, I mean, maybe starting where you kind of left off, I think one of the challenges in email is historically for the past 20 years, all of the innovation in email, if you think about it, has gone into the enterprise. And it's gone into really sending emails faster, right? So that's like kind of all the email clients that you've seen have been like, let's help you get to inbox zero and send faster. And when in reality, like our different take on email is that when you think about your personal email, not your work email, the amount of emails you send is actually very little relative to the amount you receive. And so the entire experience actually needs to be rethought of from information consumption, not as a communication tool. And I think that's the big pivot with email to think about on the personal side is that it's more around information, information consumption, delivering to you the right information at the right time. And then from there, there's an entire business to be built on how do you actually get things done for people in their, in their lives? How do you help them along that next point in time in the journey without them having to do work? And so you're seeing, you know, a whole ecosystem of amazing companies being built on the agentic side around fulfilling different actions, but all of them need kind of a distribution entry point of like, how do we start that journey? And what is the personal data that we actually help achieve some of those actions upon? And so I think email is a really great place for that. And if we can build an experience that starts to make that seamless for a user where they don't have to go and set up an API key or connect to a service and they can just seamlessly launch that from an experience that feels native to them like email, it could be a really compelling experience. And I think, you know, we previously built a very big business with ads at Pinterest and other companies. And so this idea of being able to show people products and services at the right time could lead to a really incredible consumer product that's also accessible to the masses.

Ads. I was not expecting that. Well, that was Pinterest. No, I know. But, but you're saying— are you saying, like, in theory, if you build a really—

Yeah, you pay to get to the top of the inbox. I mean, everyone says I'm sending you 1,000 emails, and if I advertise and pay, I can get more surface area, maybe?

[Speaker:JORDI HAYS] Also, I think people— I mean, our stance and many people's stance, I'm sure your stance is like, if you get a really nicely targeted ad, it is additive to a product experience. And so I think in the world in the future where somebody's using Extra for their email, and you decide to turn on ads, you could probably serve of, you know, really high quality ads users.

Yeah, some of our beta users really love actually finding deals and products from brands that they actually love that often get buried and missed in their inbox today. And when we show CMOs and marketers the product, they absolutely love the idea of their products being displayed natively, not buried in a promotions tab. You know, their CTRs are through the floor with kind of the existing tree treatment in many people's mailboxes. So if you're actually interested in a brand and you've bought something from them before, the idea of being able to see other complementary products or things from that brand in the future could be very compelling. And I think we're thinking about multiple tiers. Like, I think there's obviously a lot of folks that, you know, may just want to not have ads and pay. And so I think there's, there's lots of different ways to go about it. We're still early in the journey, but right now we're focusing on getting as many people to have a joyful inbox experience as possible before we start to monetize.

What's the beachhead agentic experience? Is it just unsubscribe?

People love coming in to clean up their inbox. It's crazy. I recommend everyone that's listening in, like, if you're tired of your inbox being overwhelmed, 5 minutes with our onboarding flow will give you an inbox designed around your life. And you'll feel a lot lighter and cleaner after using Extra. But I think a lot of people—

Is that specifically for the person that's sitting on like 200,000 emails or if they have like I probably have like 10 emails in my personal and like they range a lot from like need to write a thoughtful response to need to do some flow in some web app or need to send a payment or need to do— and I'm like, I don't know that I would hand over the keys to an agent for most of those, but I would definitely love a For You page on my like promotions updates tabs in Gmail, which are like basically newsletters and sponsored stuff. And then I would love— because Gmail's done like an okay job with the unsubscribe button, but it misses it a lot. But then there's some services that are just like, "We're breaking the law and like you have to call us to unsubscribe from our emails." And I'm like, "I'm not gonna like chase this down, so I'm just gonna keep getting like, you know, random our privacy policy updated for some telecom service usually.

Yeah. So I think with Extra, you get an amazing for you page with the information that you want. And I think honestly, a lot of people just feel the sense of control and clarity of being able to see it in a different view of like, here's what's important, here's what I need to do today. And, you know, it's distilled down to just a few items versus this long list of things things that you have to do in your inbox. That's cool.

Is the app, uh, live on the App Store?

It is. Um, yeah, we went live yesterday.

Uh, amazing. Your life's inbox, here we go, happening now. I'm getting it.

Thank you. There we go.

Very excited to try this. It's great to meet you. Uh, well, thank you so much for coming on. Thanks for coming on, breaking it down for us. Congratulations. Excited to give it a try. We'll talk to you soon. Cheers. Goodbye. Up next, we have Avlok from AngelList. And I believe we might have someone else joining as well. Ankur's joining as well. Fantastic. They've both been on the show before. You know them, you love them, and they have a big announcement today. AngelList is launching USVC, a public venture fund alongside Naval Ravikant. How are you doing?

Yo, yo. Hello. Feeling great. Welcome to the show.

Big day. Big day. Break it down for us. What is USVC? All right, perfect.

I'm going to let Ankur break down what USVC is.

Yeah, we're working on getting a 4 up. Just give us one second. Okay. We can just hang out. Great, great, great hair. Did you just get a haircut?

Here we go. We got everybody. There we go.

Yeah, there we go.

Yep, got a haircut just for TBPN, you know?

There we go. I thought so. Great to see you guys. Welcome to the show. Massive day. Ankur, it's been too long since last week. Too long.

Since we've been here. Every Wednesday afternoon blocked out. But no, today's launch is really fun. I mean, we couldn't reveal that much last week, but really the last month has been gearing up towards this. And I think Naval mentioned it when he first started AngelList. This was such a big part of his vision of what this company could become. and it's really fun to finally get this out there.

Let's get right into it. How are we describing the product? Yeah, yeah, let's talk specifically about the product. This, the, you know, everyone is aware of the companies that are a key part of it, but yeah, talk about kind of the decision-making that went into the product, how it's different than other products in the market, all that stuff.

Cool. Yeah, so the idea with USVC is we wanted to open up access to the asset class of venture. Right now, I believe it's very siloed. You need to know a lot of people. It's very, very hard for most of America to have access to the place where most of the wealth is created. So that's the goal with USVC. Can we build sort of a high-quality way to index into venture as an asset class? Asset class. Um, what's unique about this structure is there are a lot of private market ETFs. People may have seen the Destiny ETF, the Robinhood ETF, all of that. The difference with this is that this is set up differently where it's a closed-end tender offer fund, which, I mean, sounds like a lot of heavy words, but what it really means is when you're investing in this, you're actually holding something very close to the underlying business. So the price of the price of what you pay and what you redeem at is roughly equal to the price of the underlying companies. It's not driven by FOMO or like markets being super excited or down. And I think that is structurally very different.

I don't know if I've looked at that. And, you know, part of the challenge with some of these other products is they can have great assets that make up the fund. But if you want to pay, but it's hard to justify if they're trading at 5, 6, 7 times nav, it doesn't matter how good the underlying assets are. No good investor can kind of confidently say, I'm going to invest in this fund, unless you truly think that the assets are kind of mispriced fundamentally by 5x or 6x. Yeah.

Yeah, that's exactly right. I think the way to think about USVC is it's meant to be a vehicle that you can think about over the long term. So rather than trying to think about, am I getting it at the right nav? Am I getting it higher or lower than it's valued? You can just trust that when you invest in USBC, you're getting it at the right nav and you can hold it for the very, very long term. That's why we made the decision, or that's why USBC made the decisions that it made.

So how do you, how do investors actually onboard? It's very easy. Right now.

Yeah, it's very easy. We'll get, we'll get you all to invest soon, but it's fully self-serve. USVC.com. It's been, yeah, then that's sort of been the innovation here. Like historically, these types of funds required people to be accredited investors, but now access is truly open to anyone with as little as $500.

How are you thinking about number, number of bets? It's, it's open-ended. So, so it sounds like uncapped, but is this the kind of thing where you want to— I feel like Is it you want to concentrate on like 20, 30 core positions similar to a venture fund? I imagine this isn't like you want to spray and pray at seed, but maybe you would if it's the right founder. How are you thinking about it? Yeah, so I can talk about portfolio construction.

You'll see a lot of the names we've gone out with today are the names everyone has heard of, you know, xAI, OpenAI, Anthropic. However, long term, I think a lot of the long-term alpha is not even necessarily in these names. So we absolutely will also start investing in a lot of seed stage companies. The idea is these are bets. It's not just about the next 1 year, but bets that will pay off in the next 3, 5, 7, 10 years. So about a third of the portfolio, the goal would be to have it skew early stage as well, just because we want to be finding these companies before they're actually discovered.

Yeah, so similar to a kind of a platform VC where they're gonna concentrate a lot of their capital into later stage established winners that are more stable, and then you can still get exposure to the early stage of super promising companies. That's very cool.

Can you talk about the concept of direct ownership? How many layers of abstraction? I think that there's just like FUD generally among investors retail investors that I've talked to, that they will see a name that they're excited about, and then they will find a product that sort of proxies that name, but it's 7 layers deep, it's very confusing, and they don't really know how to process it. And mostly, they're not even necessarily worried about fees or anything like that. They're more worried about like, I thought I owned SpaceX, and then it just turned out I didn't. And we've talked to Matt Grimm at Anderol about this. There's been all sorts of stuff.

Yeah, people promoting Anderol SPVs that don't have access. Don't have anything.

And it's just like, yeah, and I think that messaging is important. So can you talk us through how you think about ownership, what rights you actually do need, what rights you don't, how you think about delivering exposure to the end user or the end investor?

I'm going to let Avalok take this because AngelList as a platform has drawn the line on what they will accept as an SPV. So, Avalok, why don't you take this one? Sure.

Yeah. One of the unique things about USVC is that it, it runs on AngelList. So we're able to have it take advantage of all of the infrastructure of AngelList. And one of the things I tweeted out a few months ago was AngelList, as an example, had banned all layered SPVs by default for the reasons that you mentioned. In fact, when we did it, we lost revenue because of it, but that's because we didn't know who the underlying owners were and we had no way to actually even trace it. And so we banned it. So in general, when it comes to the infrastructure that we're running at AngelList, we are always thinking first and foremost about ownership of the underlying assets because is we're running infrastructure for funds, SPVs at scale. So from that vantage point, you can think of it as USDC in the way it thinks about it, acquiring assets as the exact same ethos, which is you want to only look at assets where you actually know where the underlying assets are coming from rather than these layered SPVs that have fees that are not disclosed and carry that's not disclosed. That's something that we effectively banned from AngelList a while ago.

Makes sense. What is the pitch for founders? Why should they take USVC money? I can give reasons, but how are you positioning it?

I think that this is the most fun part. And you're gonna see this soon where, okay, you know, you'd argue that Anthropic doesn't really need more promoters right now, but we're going to be taking some bets on companies and startups that have very loyal, very passionate communities, but they may be a $50 million business, $100 million business, something in that range. And to them, it allows them a very scalable way to actually let their community share in the upside. And that's going to be one of the magical things here where it's not obviously an IPO or anything, but now we can have a vehicle for people to get their communities involved in their upside. And for most founders, it's, you know, generally pretty cool to be able to say, hey, by the way, if you are a customer and you want to participate in this, you can invest in USVC.

There's a $500 minimum. Is there a maximum?

There's, there's no maximum.

So load up, load up the truck. What happens if Elon, Sam, Dario, these guys are like, you know what, I want to just team up with the other guys? Let's throw $50 billion collectively into each other's companies. Through USVC. Through USVC. Yeah, let's just all create a keiretsu. This is the Japanese way, correct? Yeah, yeah.

You did strike something interesting. The nature of this fund is different, right? Effectively, it's an evergreen fund, which means we can accept kind of unlimited amount of capital, but from a deployment perspective, Every time that happens, effectively, we have, you know, we have to kind of come up with a strategy for that. Yeah. So it's a very different structure.

Yeah. How should investors think about the time horizon if you're, if you're going— I've talked to people that have LP'd into VC funds and they're like, I don't even know if I'll see it before retirement. I mean, there's people who got into like early rounds of SpaceX and they're, and they're like 20 years into like, it looks great on paper. But I still don't have liquidity, and that's a benefit, but it's also a drawback. And I think going in with eyes open is important, but how should the investor, do they think about this as 10 years journey or something that they're throwing in retirement bucket mentally, or should they think about it, obviously it's not something that they're gonna be trading in and out of on a day-to-day basis, but what's the right mindset to go into this with?

So I'll caveat that this is not a liquid instrument, right? Like we've been told to repeat, this is not a liquid instrument. Yeah. However, the goal with this type of structure is at our discretion, things are looking good and this is our goal. We will have up to 5% of the fund available for redemption every single quarter. Sure. So the idea is that people theoretically, once we get into a good rhythm, them can kind of pick and choose for how long they want to be a part of this journey. Sure. Again, it's not the most liquid instrument, but from a timing perspective, there's— it is, you know, people can kind of come in and come out. However, you know, only through quarterly redemptions. Yeah, that makes a lot of sense.

Very cool. Jordi, anything else?

No, this was great. I'm very, very excited. This is the most angelless product ever, I think, even though you guys are more than a decade in now, right? Yeah. And yeah, I'm excited to see where this goes. I feel like, I imagine you guys are gonna be spending a little more time on the East Coast with this. I guess you're already over there, Ankur. We're already in New York, and I work from the West Coast, so we have good coverage.

Yeah, good coverage.

The website is usvc.com. There's a beautiful video of Naval Ravikant.

And I have no doubt we'll have many, many of your new portco, Founders on the show. We can't wait. So keep us in the loop. Excited for it. Great to see you guys.

Have a great rest of your day. Congratulations. We'll talk to you soon. Uh, uh, this, this launch video is very funny because, uh, there are clips of Naval and then there are clips of Founders, and the first, uh, the first clip sort of makes it look like Naval is talking to Steve Jobs if you, if you just have the sound off. Uh, but what a, what a fantastic portfolio. Uh, go check it out usvc. There are 7 companies in the portfolio currently: XAI, Crusoe, who's been on the show, Anthropic, Sierra Technologies, Brett Taylor, Lagora, OpenAI, and Vercel. You have 7 different companies here represented. Uh, go check it out at usvc.com.

Well, uh, wow, before we go into the next guest segment, I I'm trying to— this hasn't been posted by any, any super reputable sources yet, but there's a number of sources saying that Thoma Bravo, actually according to Reuters, is nearing agreement to turn software firm Medallia over to creditors, which would be a $5 billion loss. Oof, that is rough. Absolutely brutal. We'll dig in here more next.

Anyway, We have Joel Edwards from Zancar in the waiting room. Let's bring Joel in to the TVPN UltraDome.

Joel, how are you doing? What's going on?

Good. Hey guys, welcome to the show. Thanks for having me on. Great.

Thanks for hopping on and thanks for cleaning the camera. I appreciate that. Always want to look sharp on TVPN. First time on the show. Please introduce yourself and the company.

Hey guys, Joel Edwards. I'm a geologist by training. I'm from Zancar Geothermal. We're a group that we wildcat for new geothermal resources out west to make power. Okay.

What's the best? Incredible. What's the best geothermal? Actually music to my ears.

What's the best geothermal?

If I got one more AI agent, I would— No, I'm kidding.

Name every geothermal resource. Like what are we actually, we just concretize it for me. Like what are we talking about here? What is you striking gold look like? Striking gold.

In your head, have you guys ever been to Yellowstone National Park? First national park?

Yeah, I think so.

Seen the Old Faithful geyser, any of those thermal features? Okay. These are the types of systems we're looking for, except for the systems we're looking for are hidden or blind. Okay. So there's nothing at the surface. There's no geyser, there's no hot springs. Okay. Nothing. But there are massive geothermal resources at depth that you can tap with a well field. You can pull that heat out of the ground and you can put it in a turbine and make electricity. Interesting.

Yeah, take me through the geology 101. Where's the heat coming from? Why is this underground? How do we get it out? Is it just, there are just like pockets of hot air, hot water? Really dumb it down for me.

There are pockets of hot water underground. Okay. Not hot air. And they're fairly discreet, fairly local, but when you find them, they're incredibly energy dense. So, think in your mind like volcanoes. Yeah. Right? That's a manifestation of a geothermal system. Okay. Think hot springs and so forth. There are parts of the world of the crust where there's just a lot of heat. Yeah. And that gets manifest in these different thermal features. And you can drill into these things and You can pull steam or hot brines out of the ground and then you put it into a power plant just like, you know, all thermal power plants, nuclear, coal, combined cycle gas. Yeah, they all put heat into turbines, turbine spins, generates a generator and you make electricity and so forth.

Yeah, it's that classic meme. We discover a new way to generate power and we are going to use it to boil water and generate steam and spin a turbine so we get more electricity. And this is like Exactly. And we do every time we discover something new, we wind up making a decision.

What is the business actually gonna look like? Is this whale hunting where you're really trying to find a single site in the next 24 months and then you'll spend years?

Do you wanna commercialize it or do you wanna sell it off to another company at that point? Like how full stack do you wind up going here?

Yeah, we already own and operate a power plant, geothermal power plant in New Mexico. So we sell electricity. To an investor-owned utility, and then they service Albuquerque and Santa Fe and so forth.

Interesting.

So the business is to find geothermal resources, build a power plant, and then sell the electricity to a customer. Yeah.

So this is part of that. Yeah. Sorry. Sorry. Is there— is there— this feels like eco-friendly. Is there pushback to geothermal energy production generally? Like people are worried about earthquakes or something? I don't know. Like, it feels like this is— there's a—

Yeah, people are going to find a way to find anything. So what are they going to hate about this?

Like, are you in— like, are you renewable? Are you clean energy? Like, what category are you in, do you think?

Yeah, right now we are beloved by both sides of the political aisle.

So you show up in D.C., enjoy it while it lasts, buddy. Yeah, exactly. Somebody will find a way. Triple glaze.

We're one of the few like bipartisan things where you show up in DC and everybody loves you, shakes your hand. It's a really exciting moment. Cool. There's, there's attributes about geothermal that people love. It's baseload, there's no emission profile, it's domestic, so all these resources are in the US. Yeah, you know, it uses US supply chains, it uses drillers. Everybody loves drillers. Well, not everybody, but a lot of people love drillers. Yeah, and so forth. So it has a lot of those attributes that speak to both the left and the right. Sure. And then probably Another attribute is the underdog attribute. Everybody loves a good underdog. Geothermal is an underdog. We're a niche, small industry. We have a lot of the same setup that oil and gas had 100+ years ago, where it looks like there's a lot of resource out there, looks like a ton of potential. It's still niche. Is it going to scale?

Yeah, it's just one of those things like rare earths are not actually that rare. It's just hard to find, hard to get to, hard to industrialize. —refining nearby.

How much geothermal energy is there? Like, are we talking like tens of gigawatts potentially if we were to put in like AI data center parliance?

Yeah, I think tens of gigawatts is near-term super duper realistic based on the tools and the tech that we have today. Yeah, so exciting. Yeah, obviously there's people talking hundreds of gigawatts, terawatts, We'll see what happens, right? As things, as industries scale, nonlinear, unpredictable things happen, right? Costs and technology and all these things change. So, it's hard to predict what's going to happen 10 years from now, but basically, there's two things that drive scale for resources. It's the number of resources out there in existence and then it's your ability to extract the resource, and that's really technology dependent. So, geothermal, where Zanscar really focuses on on finding more geothermal hotspots. We have a thesis that there's a lot more out there. We've already found a bunch of sites. We've announced a few of them. There's more to come. And then there's other groups that have focused on how do you get more out of these systems, and those are in the unconventional bucket. It's like, it's like the oil and gas play in the Permian in West Texas. Yep. Right? They came in with unconventional well fields. They drill horizontal wells. They stimulate. They get more hydrocarbons out of the rock. Yep. That process is happening in geothermal right now as well. Okay.

So you have the exploration piece, finding more, and then getting more out of it. Yeah. Double-clicking on the exploration piece, it sounds like AI can be used to do more computational discovery, actually interpret data, but what's the gold standard for actually obtaining the data? Like what sensor are you using to detect the potential of a geothermal you know, discovery system?

What's the secret sensor? The trick in geothermal is that there's not a single type of geologic data that tells you if a system is in existence other than just drilling a well into the thing and measuring.

So there's nothing you can do, just like radar or like pull something through the ground and find it that way? Unfortunately, no.

I wish it was that easy.

Okay, you got it.

In oil and gas exploration, they use seismic reflection imaging, and that was kind of their silver bullet toolset to find a lot of these reservoirs. In geothermal, we use a mix of datasets to try to find these hidden systems. Interesting. And that's where, that's where like AI tools really flex because they can, they can handle the hyperdimensionality problem really well, better, better than humans. So we basically build a sandbox in which the models learn how to find systems. And building the sandbox piece is the really hard piece. You have to put the right data in there, you have to put the right learning, objective rates and so forth. That's the piece that we live in.

We build the sandboxes for the models. Interesting.

Last question for me. I feel like you guys are bending towards the meme template, like the world if it ran on geothermal energy and it's like, yeah, there's like, utopia, flying cars, everything's green.

Hypothetically, friend of the show, Mark Benioff, he's out in, in Hawaii, he's big on AI. Could we put a cap on a volcano and capture all the energy out of a volcano and use that to power Salesforce AI?

Sure, why not? A cap? What does the cap look like?

I don't know, I'm just imagining a physical— Something to trap the heat. Something to trap the heat that comes off of the volcano. Maybe you wanna drill into the volcano and extract the lava that way, but has anyone actually, at least theoretically, theorized about capturing energy from volcanoes? They seem very powerful. They're pretty powerful.

We have a lot of active operational geothermal wellfields on volcanoes today. Really? So the main island of Hawaii has a geothermal wellfield. They've drilled into the side of that volcano and they make electricity out of it. That's crazy. So I guess if that's the cap. Yeah, yeah, yeah. I guess the cap is the wrong metaphor.

Little mini caps. Yeah. But they are extracting energy from the volcano. That's remarkable. That's right.

Wow. Yeah, so there's a lot of geothermal operations, power plants in Indonesia, Japan, Philippines, where a lot of these big volcanic systems live. In the US, we have some volcanic systems generating power in California. Yeah, yeah. And then most of the rest of the systems are generating from cracks or fracture zones in the crust. Sure, sure.

Yeah. That's very exciting. Amazing. Did you raise money?

We raised some money, yeah. How much did you raise? We got some cash.

I wanna ring the gong for you. It's exciting.

The gong is coming. I knew the gong was coming. Yes. We raised $115 million a few months ago and— Sounds good.

And just closed out a $40 million credit facility. Congratulations. Thank you.

Wait, $40 million what? Credit facility?

It's a credit facility, yeah.

Massive. Yeah. It's all good. I'm really glad you're doing this. This is amazing.

I'm really glad you're doing this. Appreciate it. You know, people talk about AI bottlenecks and AI, right? You know, talent, chips, you know, energy. And there's sort of like this cycle in technology live streaming right now. And right now there's sort of like a talent bottleneck and an equipment bottleneck. But, you know, if we keep doubling from here, there will be an energy bottleneck in a year. And technology live streaming. So I'm glad you're doing this work.

Thank you guys, appreciate it. Have you ever been on a drill rig? No, I have. You guys have not lived yet. I mean, this is the American experience is to be on a drill rig.

I know, I know. Tell us when you're in Southern California. We would love to go live from the rig. Yeah. So anyways, thanks for hopping on.

We'll talk to you later. Thanks guys, appreciate it. Cheers. Goodbye. Up next, we have the founder and CEO of Vast Data, just hit a $30 billion valuation. Congratulations. With a $1 billion Series A. The biggest ever. Absolutely massive news. How are you doing? Welcome to the show. Good. How are you?

Thanks for having me.

We're good. Would you mind just giving us a little bit of the journey, a little bit of the background? I'd love to know. All the things that went into building the company to this point since it's such a massive milestone.

Of course, it's been 10 years in the making. We started in 2016. We wanted to redo infrastructure. We thought AI required a new infrastructure stack way back then. If you remember 2016, it was the very, very early days of AI. But it was already clear that you needed fast access to a lot of data to train these massive systems. And now we needed to infer and to fine-tune and for reinforced learning. And so we've been on this journey with AI throughout its various evolutions. We started from storage building a massive—

Was there any points on that journey where you were like, maybe we were too early? Did we pick kind of the wrong— category, or was it kind of like steady— you felt like there was steady progress being made and you had line of sight to getting to where you are now the entire time?

We definitely did not have line of sight when we started. If you looked at my presentation that I built for myself, the last slide had a picture of a brain on it and it asked, "Can we build a thinking machine?" I'm not sure we still have today line of sight to doing that. That, but it's been our North Star. And over the years, we always partnered with the organizations that were on the bleeding edge of AI. Before generative AI, it was hedge funds and life science institutes that were doing massive analytics using GPUs. And then of course, once generative AI hit, it became the labs and the AI clouds that became our biggest customers.

Do you have insight into how hedge funds are scaling compute resources? I just saw— didn't Jane Street invest $1 billion in CoreWeave? And it seems like I saw another company that they funded that was doing custom silicon, and it feels like they have incredibly advanced compute stacks that are a little bit more opaque because they maybe just have a different recruiting flow or different public messaging flow. That is exactly right.

In fact, Jane Street is a customer of ours, both directly and through CoreWeave. And so we know them well. We play in that space now for 7 or 8 years. I found that before the big AI labs, the hedge funds were the ones that had the biggest GPU clusters. Yeah, they're ahead of their time and they're very secretive. We don't know much about what's going on there.

What about the social networks and the social platforms? I feel like ad delivery is also potentially underrated as a source of like large-scale machine learning, large-scale compute clusters. We've seen, you know, Meta trained the Llama model on the Reels-sized cluster that they built as sort of a bonus. But how big is that? Of a business for you and how do you think it comps to the hedge fund world?

I think the hedge fund world is a lot more secretive, but I would dare to say that it's bigger in terms of size. Um, there are very few of these Metas out there. Um, and there are a lot more hedge funds.

Oh, interesting. Okay. Um, well talk about, uh, what the next phase for your company is on the back of this financing round?

So we're trying to fill in that software infrastructure layer. Uh, Jensen gave a really good analogy of a 5-layer cake: power, chips, infrastructure, models, and applications. We're that middle layer. We want to abstract this new hardware away from these new models and applications. We want to make it easy, make it secure for everybody to to generate AI agents and use AI in production. And especially as we move from the labs into enterprises, I think it needs to be simplified. And it's always been that software infrastructure layer, what I call the operating system, that took a new technology from new hardware and a killer app to making it widespread where everybody can use it. And I think we're at that moment with AI and we're trying to do our part in accelerating the adoption of AI.

How have you been processing the CPU shortage? This idea that like, as agents get more and more robust, they need more CPU, they need to be fed, you gotta feed the GPUs. Is that something that you're working on particularly, or do you have a view on the overall, just CPU shortage, which sort of blindsided, I think, a lot of people in tech who'd been following the GPU ramps, but then they're saying, oh well, what's going on with the CPU crunch?

I think these AI factories are bigger pieces of technology than we've ever built before. It includes CPUs, it includes DPUs, it includes GPUs. In some cases, it includes TPUs. Sorry, what are all these products?

DPUs? Can you break that down? D, data processing unit. Oh, data. Okay. Yeah.

Yes. And so all of these chips are in high demand. We're seeing a NAND shortage that's extremely severe right now. We're seeing a memory shortage. And so I don't think there's anything that is of abundant supply right now. The fabs need to be built out. Nobody was expecting seeing this type of curve in the growth of AI. And I think it's going to continue for definitely a few years to come, if not many years to come. And so the hardware vendors need to build out a lot faster than they have been.

Yeah. Can you walk me through at a deeper level your relationship with CoreWeave and how that fits together? I think a lot of people think of CoreWeave as servicing, you know, the companies that are training and inferencing models specifically, what are you getting out of that partnership?

So CoreWeave has been an amazing partner for us. They're the biggest of these new AI clouds like, uh, Enscale and Crusoe and Lambda. You see them around the world also as sovereign clouds. Yeah. And basically, uh, they're building this new stack. Yeah. And they're making it easy for end users to consume this new stack in a cloud environment, or just the software builders. Sure. And so we try to provide them with as much software as we can such that they can turn that around into cloud services, whether it's a storage service or a database service or a streaming service. As we add more and more layers of this stack, those are the things things that are required from those end users.

Okay, so we, I mean, we were just talking to the president of Adobe. It sounds like Adobe is doing inference and they're training some models and they're doing a lot of different things. They might go to CoreWeave and then Vast Data might be powering the software underneath and they might not even necessarily have a direct relationship with you because they're getting just access to the best possible product from CoreWeave directly effectively. Is that right?

It is right. Adobe specifically does have a direct relationship with us. Most of these larger enterprises tend to have both a cloud environment that they're leveraging and on-prem and hyperscale. What we've built, what we call the DataSpace, basically allows them to do all three of those and have one abstraction layer that enables a seamless experience. Interesting.

Given your exposure to the, to the world of hedge funds, are there a bunch more Situational Awareness-esque funds that have absolutely printed over the last year or so, but just done it a lot more quietly and the West Coast kind of hasn't picked up on it?

That's a really good question. I actually met those guys from Situational Awareness a couple weeks ago. They have a really interesting view into the future. And I think they're capitalizing on that view for every phase as it unrolls. I am not the right person to ask because those funds tend to not have as much infrastructure. The ones that we work with tend to be more quant-driven. Situational awareness, I think, is just a few very smart people. They don't need that much compute power.

Yeah, yeah, it's more macro research.

Doing it the old-fashioned way with guts.

Yeah, yeah, I love it. Conviction. Yeah, yeah, I mean, conversations and research and having a contrarian view. Well, congratulations on the round. Thank you so much for taking the time to come chat with us.

Yeah, great to meet you. Fantastic. Appreciate you breaking it down. We'll talk to you soon.

Thank you. Have a good one. Yeah, can you imagine situational awareness with Couple gigawatts of compute under the hood. Anything's possible at that level. I was— What was that copy trading profile? You know the one I'm talking about where you can like create the Nancy Pelosi stock? Autopilot. Autopilot. Autopilot has a tracker for situational awareness.

Every time they post about it, they use this like AI image of him. It doesn't really look anything like—

No, no, no, no. But there's so few images of Leopold on the internet that that AI image has just become like what mainstream media will pull from for him because they just assume that that's actually him and it's not.

But it's just very funny that it's a man who has only appeared on Dwarkesh. Yes, there's no other. There's only video of him.

SBF just posted his full holdings. If lawyers hadn't fire sold, he's up 165x on Anthropic, 72x, 27x on Solana, 8x on Robinhood, Genesis Digital. Anyway, uh, he's taking a victory lap. We have our next guest, Darian, from Gradient Ventures, in the waiting room. Welcome to the show. How are you doing?

What's going on? Hey, how's it going?

Good to see you, Jordi.

It's been a while. Good to see you. It's been a long time.

Good to see you too.

Uh, where are you calling in from?

Yeah, I'm in our office in Jackson Square. Um, we just, uh, moved in here about 8 months ago, and we're right next to Thrive and Obvious and IVP. It's like, uh, you know, a VC community now. We could live here even.

There's apartments upstairs, you know. So amazing. Uh, since it's the first time on the show, I'd love to get some of your backstory, some of your journey to venture, uh, just a little bit of like how you wound up in this particular seat in Jackson. Sure.

Yeah. I mean, I, I grew up in Silicon Valley. Um, my father came here in the '70s. He worked in in tech as well.

What was he doing in the '70s? Like, what was tech like in the '70s? Semiconductors. Semiconductors.

Semiconductors. The true Silicon Valley.

Yeah, yeah.

True Silicon Valley where we actually made stuff. Yeah, we actually made stuff. He was industrializing, now we're re-industrializing. But so he was working in semis, okay.

Yeah, and then he sort of fast-tracked and like coached me through everything on, you know, what startups were, venture financings, all those kinds of things. He was actually fortunate because in business school his, his closest friends were Vinod Khosla and Bob Kagel. Um, probably everybody knows Vinod, but Bob was one of the founders of Benchmark. And so he had been steeped in, you know, all these different kinds of communities that were very, very venture-focused and startup-focused. And the industry has evolved a lot. Then I was fortunate because I actually got an introduction to Sean Parker when I was 17 through my cousin Hadi Partovi, who I think you've had had his brother on the show. Yeah, yeah, yeah. And he introduced me and I visited the Facebook office and met Mark, and Mark was like, "Oh, you should come work here." I think they were struggling to hire people at the time because I wasn't exactly the best engineer, but it was a fun time. No, we built a lot of stuff. We launched News Feed, Photos, et cetera. And then I went to college.

Wait, wait, really quick. Is the story, is the apocryphal, potentially apocryphal story of like everything was a single file and you would yell, "Hey, I'm changing the code." Is that, do you think that story is real at all?

Oh, I mean, like when we wanted to push to the web tier, we would just rsync the latest folder on Dustin's laptop. Wow. It was like, that was the push process. And then we introduced Subversion, actually, there was a guy that I worked with, Aaron Sieg, who was brilliant, who we put together like Subversion, checked in files in an app, we created templating, I worked on that project too. But when we got there, there was like tons of SQL interspersed with HTML. It was hilarious. But that is what startups are about, is it's super messy. And that's what I learned there. Although I also learned that only one in a billion are as anomalous as Facebook. Not every startup can go that way, but also hilarious.

Yeah, did that warp your kind of like, where, I don't know the full story.

Did you come out and you were like, everything I turn, everything I touch has turned turns to gold, and then it's like, okay, normal VC world, like there's gonna be failures.

I don't know, I feel like I'll talk with founders sometime, and it's their first company, and it's going so well, and they don't even know that, like, how anomalous it is. And sometimes I'll just be like, you have something really, really special here, and stay locked in, because the odds of this happening, you know, another time in this way are very low. Oh, totally.

I mean, I ended up starting an enterprise software company, which was much more difficult, cool. I remember meeting Aaron Levy when he was starting Box. He was in a bunk bed with Dylan in a house in Berkeley, and they asked me if I'd invest, and I thought they were crazy, and I didn't, and that was a huge mistake. So, you know, but then there, there are positive stories. Like, I met most of the founders of Palantir and invested personally early in Palantir, and that was— ended up being an incredible company. Or, you know, uh, you know, when Luke Nosek was talking about SpaceX like almost a decade ago, and I was like, you're crazy, we're gonna rockets to the moon, like, we're not gonna do that. And ultimately, you, you, uh, you really just have to surround yourself with really smart people. Fortunately, I've always been surrounded by people much smarter than myself, and so I think that's why I'm in this chair. Um, ended up being a founder, invested in a lot of great companies, like first investor in CaseText, which got bought by LexisNexis. Uh, one of the first investors in Lagora, which is now on a tear. And then at Gradient, we've, you know, seeded Lambda Labs. My partner did that. Deal. You know, I joined here about 8 years ago with the goal of investing in AI, mainly because I didn't get crypto. And I was like, I don't understand why crypto is interesting. And so I joined the nerds at the AI fund.

Yeah, during the 2021 era, like, you guys, you know, you were Google's early-stage AI fund, but you were still like, as a fund overall, some of the partners were like branching out and would do different stuff, obviously. But you guys now get the benefit of like, we were doing all these companies, like, you know, almost a decade ago or a decade ago, and now they're like, in quarantine.

What was the enterprise software company that you built? And like, coming out of consumer social, why enterprise?

Yeah, I mean, I was always obsessed with data. Like, I actually thought that the edge that Facebook had was how clean the data was on everyone's profiles. People would just give us their birthday, like what classes they were in, like all this information. And it was very well organized. And this is before LLMs where, you know, you couldn't organize and massage that data very easily. Now you can, things can be amorphous and it's easy to process, but before you couldn't. And so, you know, I was very interested in, you know, small business data. I thought that we could compete with Dun Bradstreet. It was a company called Radius that we scaled to, pretty significant ARR, and we sold it. It was a huge, it was a long journey. It was full of crisis, full of learning. And actually, that kind of helped me with becoming a VC, mainly because I really empathize with like how founders are going through a terrible time. I mean, running a company is not fun, regardless of whether it's going well or not. Like the best companies in the world, it may look like everything's going well, but the mess is just massive. Masked by growth. It's like, that's the difference between a good company and like not a good company. And, and in general, like, founders I really have a lot of respect for because I decided not to sign up to found another company after my own. I decided to go, go to the dark side and be a VC, which sometimes I say is a cop-out. But, you know, being a founder is really what— it's really the most difficult job. And if you can sign up for it multiple times, I mean, that's incredible. That's a sign of like just incredible Or a masochist.

Yeah, that too. I'm so curious— I'm so curious what you're excited about investing today. And I know it probably comes down to just backing great teams, because that typically works well. But we're in such an interesting time right now, where it feels like all software, all enterprise software is converging on the same kinds of— of simple use cases, this natural interface to get agents to do things. Like, everything is converging. To just the text box. And you guys were— you did Lambda. You did a NeoCloud 8 years ago. That was like the best time to be investing in a NeoCloud. And then now, it just feels like you have the SaaSpocalypse in the public market. A lot of early-stage companies that are getting investment are still SaaS. So I'm curious, do you still believe in SaaS? What kinds of opportunities are you most excited in?

I believe in shorting SaaS if I can. That's probably what I believe in right now.

I generally think that software is—

Bullish on puts. Oh, that's a good laugh there.

You're bullish on puts.

Yeah. Yeah, I'm bullish on Salesforce puts is what I'm saying. But, you know, I think, I think that the reality is that we are doing a lot of soul searching like a lot of seed funds. You know, there are a lot of things that are still very interesting and investable. Like, we just invested in a robotics company that I'm super excited about that has a hardware component. Yeah, I wouldn't touch hardware with a 10-foot pole 5 years ago. Now I'm very interested in it. Um, you know, we've invested in, you know, tools that help you manage coding agents. We've invested in video editing solutions. We're investing in an AI hedge fund. Like, we are investing in a lot of really interesting things that are outside of the software as a service scope. And the real reason is just that if you want to build anything, you can in a few hours. It's just so easy to. And I think that that is a major change, not just for software companies, but generally for the bottom 40% of work that you sit behind a desk ask for, maybe the bottom 60%. And so we're spending a lot of time thinking about ways in which we can think about new areas. Now, we have the benefit of having done this before because when we launched in 2017, right after the transformer paper came out, there were no AI companies. There were none. And OpenAI was a nonprofit, Anthropic didn't exist. DeepMind was still like a startup. A small division within Google. Like, you know, now, you know, so we had, we had to go and hunt and find companies like Lambda, like Rad AI, like Streamlit, like Writer, you know, like a lot of these businesses. And, and so we're getting back onto what we had done before, which is scouring and doing a lot of outbound, going and meeting founders, getting to know founders, and, and really slowing our deployment pace until it's very clear where the value is gonna accrue at the seed stage. And that, I think, is the smart thing to do right now. I think that software in general is really in jeopardy, and we've made some software investments, but we believe they have strong moats, because they have network effects, or they have data network effects. But in general, like 80% of software companies are really in trouble.

[Speaker] Yeah, talk about the pipeline for founders and how it might be changing. I'm sure you've been tracking, like, there's this crazy fall-off in the number of computer science grads, and that used to be just fertile ground for, it used to be general casting, right? Central, you know, central casting for, you know, VC-backed founder was like Stanford CS, and I don't know about Stanford specifically, but it seems like we're getting a lot less or fewer computer science grads, and I'm wondering if you envision a future where you'll be pulling more physicists and mathematicians or theologians or philosophers into entrepreneurial stuff. I mean, Mark Zuckerberg did not study computer science as his primary major, right? Wasn't he a philosophy major, I think, or something? Psych major. Yeah, psych major. So like, there's plenty of examples that it works.

Honestly makes a lot of sense for social media founder. Brilliant, brilliant.

But I mean, do you see there being like a pipeline crisis? Is the contrarian move now to study computer science computer science at a time when coding jobs are going away? Like, walk me through how you're processing this idea that, you know, our best and brightest aren't going to, you know, just learn to code and become computer scientists, you know, at the same rate. Oh, I mean, absolutely.

You talk to any scale-up or, you know, tech company today, the mid-level engineers are managing coding agents in English. They're not even coding. They're managing, like hundreds or many dozens of coding agents at once. I think this, at the senior level, the engineers are probably okay because we do need more code reviews and evaluations because a lot of these coding agents are spitting out bad code at volume. And that will, that'll get improved over time, but I think it is a challenge. But I do think for entry-level positions, we've told a whole generation, including my generation, that you should go to college And like, I don't think that that's necessary, right? Like, college was originally formed, especially liberal arts, was formed for the process of being creative or thinking or philosophy or things like that or writing. We've told people that they can go to college and get a good job at a tech company, all those kinds of things. I think that that's kind of in jeopardy currently. My hope is that we focus more on like real trades, right? Like welding and plumbing and a lot of the, or, you know, you know, ceramics and pottery, like these are going to be like more respected work lines of work in the future as knowledge work just gets completely replaced by AI. I don't have a great answer for you, but I generally do worry about this because I don't know if colleges should be teaching computer science to everybody. That being said, I think everyone should take a course to understand how AI works, how these LLMs actually work. I think that everyone should do their best to read transformer paper, you know, everyone in the world because it is kind of the Magna Carta of this generation. But you know, I do worry about this. I worry about, you know, workforce displacement and how that transition works and if our government is even capable of helping people transition from one phase to another or one era to another.

[Speaker] Yeah, I'm just wondering like, is the next great like pottery entrepreneur going to go through pottery trade school or like be a business major and be able to like marshal $1 billion of capital to like build the pottery empire of the future? Like, do you go deals guy route and then tack on the trade? Or do you pull from the trade school? Maybe it's a moot point because you'll know it when you see it and you meet the person, but it's a fascinating world to imagine because for the longest time it was like, oh, you wanna do something into pottery? Well, it's gonna be a SaaS company and you're gonna have to know computer science. And now it's a little bit different, right? I'm like somewhat right here. Do you think—

I think actually— oh God, go ahead, Jordi.

No, I was going to take the conversation another way, so finish, finish.

Oh, I was just gonna say that like, imagine being someone that's incredibly good at like some form of ceramics. You would create your product and then AI would market it, take photos of it, would, you know, design it, buy ads for it, do the back office and do the finance for it. That's actually the future that I think we're gonna end up in. Now, ceramics is kind of a weird example, but you know what I mean? Like a car mechanic, like an antique car mechanic or a welder. Those are examples of like how AI would help those people. So I think we'll have many, many more entrepreneurs that are not building unicorns. They're gonna be building, you know, smaller lifestyle businesses.

Yeah, I completely agree. It feels a lot like what happened when everyone got an iPhone and a YouTube account. There were a lot more creators. There's like this, or Shopify, same thing. Like you're democratizing these like long-tail businesses that like previously it would be like, oh, you wanna, like during the dot-com boom, there was like, oh, you wanna sell t-shirts online? Here's a $10 million check so you can buy some servers and rack them and just have a website. Teespring. Yeah, well. Remember Teespring? Teespring was sort of a meta platform a little bit, but yeah, I mean, it's a good example of like a lot of expensive infrastructure doing something pretty basic. Sorry, Jordi, what's that?

I was curious, do you think you'll write more checks this year than last year or vice versa?

Hmm. I think we're probably—

Have you gotten to, or have you reached a point of frustration around valuations? You've been through, you know— Cycles. Cycles enough to kind of know when things are getting probably overheated.

Yeah, I mean, we don't feel as though valuations are out of control at the precedency stage. Where they are out of control are probably in the later stage, you know, like Series A, Series Bs, even really late stage are trading at incredible multiples. I mean, you know, I think that we did this robotics company at like sub-50. We've done, you know, this other multi-coding agent company at sub-50 as well. Like, we're pretty focused on keeping valuations below 50. And I think that capital-constrained companies are the ones that end up being very successful because they sort of figure out how to build better products and become more efficient. It's when you overfund companies, like some of these new Neo Labs that are raising billions of dollars, people like start making bad decisions when they have, when they're overfunded. And so I think that where we play, which is not funding the next contender to Anthropic or OpenAI, the valuations are pretty favorable. Yeah.

Jordi, anything else? Thoughts on solo GPs today. Is it overrated to be a solo GP? Yeah.

And should there be a trade school for becoming a GP?

Should we take millions of plumbers? That's how we solve workforce displacement. We need to create millions of venture capitalists. Everyone who's doing pottery.

You know, we keep coming back to the pottery example. If you are a top-level ceramicist, which is the actual term, it is incredibly lucrative, incredibly prestigious. Like it is.

I didn't know you had this. Pottery.

Oh yeah, ceramicists are real— like, it is an art form and it is incredibly high skill ceiling, uh, similar to—

not as high of a skill ceiling as venture capital. Yeah, not as high as an art form.

Yes, yes, it is an art form. But yes, uh, thoughts on solo GPs?

Yeah, I mean, I, I met one this morning. I've invested in a bunch of them. I, I think that the sub-$50 million solo GP is a great business if you can raise the money You can get allocation in rounds. You don't have to compete. It's when you get above $50 million that it's really challenging because you're going to be ownership constrained. You're competing against someone like us, and, you know, we have— we have— we write a larger check. Um, I think also another thing is that the solo GPs are really effective at feeding companies to the multi-stages and to, you know, pure-play seed funds like us. Um, and so I think it's actually a fantastic place to be. Yeah. Um, but it's just going to be really rare for the solo GPs to break out one-size-fits-all beyond that. So you have to be comfortable with being, you know, a company, a fund of one with no help. And you have to be, because your fee structure is not going to be able to support much more than that. Yeah. Um, and so, but I am bullish on it. And especially with these AI tools, you can really diligence companies without needing more people. You can run your back office entirely without needing other people. Um, and so I'm in general, like, pretty, pretty excited about them. They have struggled to raise more recently, right? Like, LPs are not deploying capital at the pace that they were when solo GPs were breaking out in 2021, 2022. Um, and so I think that you're going to see a lot of them probably go join funds, and a lot of the really good ones with good returns be able to raise their next fund.

Um, do you think, do you think more of them should be thinking like, okay, I'm going to do a $30 million fund, a $30 million fund, and then a $30 million fund? Because it feels like a lot tend to be like, OK, I'm going to do this $25 million fund, and then I instantly want to ramp up as much as possible, and then suddenly have a different strategy. Suddenly you're actually competing head to head, where it feels like if you do want to break out and become a brand and a platform over time, if you have like 3 really high-quality funds back to back, but deploying them on a shorter time horizon, maybe it could be a better route.

Yes, faster deployment cycles, smaller fund. Because they're just sort of liquidity constraints across LPs. But a lot of the fund-to-funds I've noticed are now focused on sub-$75 million funds. And so there is a market for that. Excuse me. And so I would generally focus on keeping your fund size consistently small at $30 million, maybe deploy over a shorter time horizon. And then you can also get LPs to, you know, pre-commit to multiple funds, which is something I've seen people people do is that, hey, I'm going to do this every 18 months, do you want to commit, you know, some amount every 18 months? And, and that actually ends up being a better story for people than raising one bigger fund. Um, and, and so I, I, I generally do think that the faster— that the, the smaller fund on a faster cycle will yield better results. Um, because you also have to put a lot of companies in the portfolio. Like, if you run portfolio modeling, which we do quite often here, at our fund size, it's usually 40 to 50 companies per fund in our Monte Carlo simulation that yields a 4 to 5x fund. In the smaller funds, it's really like 80 to 120. And so you have to be writing a lot of checks and get to conviction quickly and be willing to own less. And then you can run SPVs into the, into the follow-on rounds, which I think has been really successful for a lot of people. But the smaller the fund, the more work in my mind. Because you have to constantly be looking at deals every week. You have to be putting capital out every week into different companies in order to make the math work out.

Yeah, well, tell us about the latest fund at Gradient. What's the structure? How much did you raise?

I wanna hit the gong. Yeah, we raised $220 million.

[GONG] [APPLAUSE] I think we already talked about it. We talked about deployment and how many—

It's also possible we already hit the gong for you. Did we? Well, just, you know, whenever it got announced. But great to see you. Thanks for coming on. Appreciate all the insights. And let's get your new robotics bet on the show as soon as they're ready. Yeah, as soon as they're ready.

We'll have to talk to them. That'd be amazing. Yeah. Have a great rest of your week. Have a great rest of your day. And we will talk to you soon. Thanks. See you. Goodbye. Thank you so much.

Great to see y'all. Later. Yeah!

Anything else going on in the timeline? There's a bunch of stuff, tons of images. Where else, where else do we want to go today? What other— oh, I wanted to get your take on this. Imagine if Codex existed in 2005, and it's this very retro— people are doing Imagine if Codex existed in 2012, and it's the, it's the Apple design cues from the early Mac OS 10, and then the Windows XP aesthetic. Do you think the Windows XP aesthetic is overdue for a comeback? This is in the middle of the GPT Image 2 feed right there. This aesthetic. It's got to come back. This and a Mountain Dew. It would fix you. This would fix me. No. I, I, I, you know, I have some memories.

Imagine playing Halo while you just look over and this is just running there. I don't know, you're just drinking a Mountain Dew.

For some reason, I don't know, a lot of the Windows features that were launching during this time were a little bit rough. This was when Mac was starting to take off. And these releases, Windows Vista, Windows post-XP, post— what was it? I don't even remember the other versions, but Windows 8, I think, was one of them. They were starting to be less loved. They were starting to be a little bit more like, is it worth the upgrade? Is it a good vibe? I mean, it's not nostalgic for me yet, I guess is what I'm saying. But there are a couple other important warnings in here. One of them, let me find it. I know you had thought about potentially putting on a bear costume and trying to defraud your car insurer. And if you in the audience were thinking about dressing up like a bear and attacking your car, in order to collect an insurance payout that's up to $141,000, think again, think again, because humans, this is in the New York Times, humans who used a bear suit to defraud car insurers are sentenced to jail. You might be thinking, I'll just throw on this bear suit and I'll attack my car and then I'll call my insurer and I'll collect a check and it'll be easy money, easy money, get rich quick scheme.

No, no. 3 Southern California residents were sentenced to jail after masterminding a scheme in which they staged fake bear attacks on their luxury cars, then collecting more than $141,000 in insurance payouts. To carry out the attacks, the residents had a person in a bear suit climb into the cars and use claw-like kitchen essentials to leave scratch marks. The Los Angeles residents then filed claims to defraud 3 different insurance companies. So I don't understand why they had to be in a bear suit. Like, were they filming? Were they filming?

Yes, there is security footage. It's okay. Okay. No way.

Yeah, yeah, we're gonna pull up the security footage. It's absolutely, uh, yeah, Rolls-Royce Ghost. Yeah, a 2010 Rolls-Royce Ghost. I know someone who has one of those. I don't think he's responsible for this, but, uh, they are dressed dressing up like a bear to get into the car and rip everything and destroy it so that you can file an insurance claim.

Hey, I'm just a bear, I'm gonna just get in this car.

The schemes are really out of control these days. That is a wild, wild one. The California Department of Insurance began its investigation called Operation Bear Claw after an insurance company flagged a suspicious claim about a bear rifling through a 2010 Rolls-Royce Ghost in Lake Arrowhead, California, where there are in fact bears. So it's not that unreasonable. Uh, the defendants claimed that a— that the bear had damaged the interior of their Rolls-Royce with photos they submitted to the insurance company showing scratched seats and doors. Video footage submitted with the claim and released by the department shows what appears to be a bear crawling around in the back seat of the Rolls-Royce and swatting at at the dashboard. Investigators later discovered fraudulent claims submitted to two other insurance companies that claimed a bear had damaged the same interior. They were just two Mercedes-Benzes just scratching everything. That is like the most perfect scratch line. Like, it's so suspicious. Like, it's so clearly— like, I feel like a bear's claws would be way more organic and like, also hopefully doing more damage.

That looks like the weakest bear ever.

Yeah. Because it's not a bear, it's a human. After its investigation, the department concluded that the culprit was not, in fact, a particularly nimble bear. The defendants were arrested in November of 2020.

So I would— I would— my first thought is not super believable that the bear walks up to the Rolls-Royce, goes, opens the door, goes around. But a few weekends ago, yeah, I was hanging out at a former guest of the show's house. He has goats. Mm-hmm. Friend drives up in his car, parks, and leaves the door open. A few minutes later, what's in the car?

A bear? A goat. A goat?

A goat. Goat was in the car? Got into the car and was just like— no way. And this was a— this was a Land Rover Defender. Okay. It was like quite a— quite a jump up. Goat jumps up and is just hanging around, like, actually just—

was there damage to the of the Land Rover Defender?

I don't think that— if there was, he didn't bring it up. But there may be some hoof marks, some hoof scratches. And this is all a complicated, uh, this just seems like such a, such a wild scheme.

Here's a real bear going in a car from Tyler. Okay, they really can open the door.

I, I feel like they had to have seen this video and thought, finally.

I thought— I, I feel like most cars will automatically lock the doors. You know, this wouldn't happen with those— everyone loves to hate on the flat door handles that all the electric cars have, but bear's going to struggle to open a Tesla door for sure. This is, uh, is, is the bear still in there? Is he— he's— wow. There's, there is, uh, there's a history of bear visitation. What do they know? Pasadena, where, uh, there was a woman who was on on the local news and said that a bear came and swam in her pool. And she seemed like oddly happy about it or like excited, like she sort of liked having the bear come and visit. And it felt like she wasn't exactly like building the strongest fence and maybe was sort of excited about the idea of a bear coming for a dip in the swimming pool. But people do odd things with animals. People dress dressing up as bears. Scientists also apparently have been giving salmon cocaine. I have no idea why, but the New York Times has a post. These salmon got high on cocaine. That wasn't the craziest part. Scientists in Sweden made an unexpected discovery. I think Tyler read this piece and can spoil the ending because—

What are they doing over there, Tyler?

What is the unexpected?

I think they tested like what happens give the fish cocaine?

Okay.

And I think the answer was that they swim much faster and they can swim for longer.

Okay. So, and that was unexpected. That feels like—

imagine being the scientists that like didn't get a grant for their research. Yeah. You know, and they were working on something. I'm really close to breaking Alzheimer's. And they're reading this article and they're like, wait, like I was gonna do Alzheimer's research and then these scientists just wanted to give hard drugs to fish. Nearly twice as fast.

Nearly twice as fast. I guess that's somewhat interesting, but this does feel like something that could be one-shotted. Interesting discovery. If you just—

no way, there's just no way you could have—

you had to prove it.

You had to prove it.

Yeah, I mean, I guess it's relevant in the, in the field of environmental toxicology.

As much as we're joking, this is a good quote: while it's unclear if swimming faster and further while under the influences harm these fish, experts say it's probably not great.

But, uh, yeah, I mean, I, I guess if you're an environmental toxicologist, it is useful to have various data points on what happens with contaminants in the environment to the fish populations specifically. And so I am willing to, to believe that there is good that will come out of this other than just a viral, a viral post. He wasn't trying to bring a Halloween hoax to life. He wanted to see how salmon in the wild reacted to pollution from the illegal drug. And there might in fact be pollution if the You know, if importation goes poorly and narcotics are delivered to the water supply, that would certainly not be good. In recent years, there has been an alarming rise in the number of waterways polluted with the drug cocaine, prompting scientists to wonder how fish might be handling their highs. As it turns out, fish indeed get wired. In a study published Monday in the journal Current Biology, Dr. Brand and his colleagues showed that coked-up salmon— This is in the New York Times. Coked-up salmon swim— isn't there some account that just says like, first time it's ever been printed in the New York Times, and they just screenshot like words and phrases that feel funny in that particular— when they're shown in the Gray Lady in print. Makes them swim faster, travel farther than their sober counterparts. The study prompts additional questions about the effects that human drug habits may be having on salmon and other freshwater fish. Well, we certainly want clean waterways.

Speaking of things under the sea, Red Lobster brings back endless shrimp for a limited time starting April 28th. Luke Metro says this is an EA cause area. Hostile takeover of Red Lobster. Make it vegan. Make it vegan.

Make it vegan. No, they gotta make a play to make it come back. And this seems like something that customers were clamoring for. So I don't— have you ever been to a Red Lobster? Nope. I don't think I've ever been to a Red Lobster. I've done the Outback Steakhouse.

There are things in life that I don't think I will ever do. No. And Red Lobster, I believe, is one of them. Ben, have you been to Red Lobster? I've been there. Yeah. You— what's your— they're big in Minnesota. Uh, it's fine. I think they're all gone now.

1 to 10, what are you rating it? Like a 5. Like a 5. I think before it was perfect.

No 5. No 5. You gotta pick a side.

4. 4. 4. Okay. Subpar. It's sub-5, in fact. Anyway, thank you for watching. Anything else, Jordi Hays, you wanna run through?

No, it has been an honor and privilege to—

Well, this is kind of cool. ESPN launching a sports-only trivia show called ESPN Jeopardy. I like this. I thought this was a really, really good idea. I can't believe It took them 50 years to come up with this concept, but apparently Joe Buck is closing on a deal to host the show. ESPN Jeopardy! is expected to air on Hulu, Disney+, and may make a run on linear TV on ABC or ESPN streaming platforms. Very— this seems like, you know, so, so obvious in hindsight. I think it'll be very, very effective. A lot of sports fans love trivia and now you can watch it.

Boeing flashback. We love you. Goodbye. See you tomorrow.