A (2:37)

I think Ben Thompson wrote a pretty strong bull case for the, for the Intel US deal. Basically, like, you had to. You had to grapple with this idea of like, this doesn't feel like free market capitalism.

B (2:49)

Yeah, I guess. I guess the only thing is like, there wasn't. There wasn't this narrative around CPUs at the time that the deal was happening. Or at least it wasn't a very public narrative.

A (2:58)

Yes, yes. The bull case at that time that I heard most loudly was not the CPU boom. It was just overall chip, it was the 14, like the leading edge fab. That basically by having the US government as its shareholder, you see those dinners with all the AI lab leads and all the hyperscaler CEOs and there's a world where there's another one of those meetings and the government administration says, hey, we are backing Intel. I need all of you to commit to buy a ton of supply if intel actually goes and builds the leading edge fab, which is gonna cost them billions of dollars, but if they have the demand guarantees, then they will actually be able to go do it. Tyler, what was your take on that?

C (3:42)

Yeah, I think Leopold, sorry, Intel has long been kind of the Leopold take, which is that, like, you know, this is the clear. Like, this is the company you want to own in the kind of nationalization world of like, okay, Taiwan risk. What are we going to do if we can't make chips, you know, in Taiwan? Intel's like the very clear strategy. Yeah, maybe it's like not working right now, but like, at some point this is not just like economics. This is like, okay, this is, you know, national security. Like, it's extremely important that we have this capacity in the US and intel is kind of the only one that's even like, remotely in the conversation.

A (4:13)

Yeah, yeah, that makes sense. There were some predictions from AI 2027 and other folks in the AI forecasting world around the rise of agents that agents would arrive at this time. I didn't see too many folks who were predicting strong, valuable, effective AI agents really predicting the CPU crunch. But that's exactly what's happened. So AI agents need CPUs to do things. Training frontier models. That's still a GPU story. But running agentic workflows across data centers, orchestrating orchestrating tasks, rout jobs, managing memory, handling inference workloads, coordinating servers, increases demand for the boring old central processor. Intel's data center segment produced 5.1 billion in quarterly revenue beating the 4.5 billion that analysts expected. And Intel CEO Lipp Bhutan said the next wave of AI is moving from foundational models to inference to agentic AI. And that shift increases the need for Intel CPUs, wafers and advanced packaging. On the earnings call, he said that the CPU to GPU ratio is closer to one CPU for every four GPUs, versus one for every eight in prior years. And there is an interesting forecast from Lip Bhutan as well. But VK macro says CPU to GPU ratio flipping from 1 to 8 to 8 to 1 is absolutely wild. That's just a completely new world to what we've had so far. This is from Evercore. ISI's Mark Lapesis has upgraded intel directly from neutral to outperform and mentions that as AI workloads shift further toward inference and agents, the weight of CPU demand will rise sharply and the CPU to GPU ratio could flip from 1.1 to 8 to 8 to 1. Which is a massive, massive switch.

C (6:03)

Yeah, I mean, I don't know. Like, that's kind of a crazy ratio. It is, because, like, the less. That's the strongest frontier models is that, like, the models are going to keep getting bigger. You're going to need way more chips to inference them. Like Dylan Patel was on Patrick o'. Shaughnessy. I think it came out yesterday or the day before. And he's just saying the models are going to get really expensive tokens, are going to get super, super expensive. People are going to price down. So I don't know if I fully believe the narrative that you need all the CPUs. Yeah, you're going to need way more CPUs than.

A (6:32)

I mean, the.

D (6:34)

That's all, folks.

A (6:36)

It really lines up perfectly. His camera is the main one that fits perfectly. The Glasses land.

B (6:42)

I love a Friday.

A (6:43)

Yes, that's a good point. But there is this, I think, at least in the midterm, like the.

C (6:48)

Yeah, there's like, capabilities overhang and just

A (6:51)

like one really smart AI system that's being inferenced on GPU can spin off a ton of CPU workloads and do a lot of things that require CPUs. Yeah. Even in the SaaS apocalypse, like you Vibe coded Slack or whatever, it's like, well, that Vibe coded system is running on a cpu and if it took you like, I don't know, a thousand GPU hours, but then that System runs on CPUs that are running constantly for every user, you still wind up with creating more CPU demand out of the gpu.

C (7:24)

I think this Take makes a lot of sense. If you are long open source. We're going to have a lot of these open source models. They're quite small. You can run them on small amount of chips, but you just have a lot of them. You have a bunch of agents doing a bunch of stuff same time rather than the single model, you know, inferenced on a massive amount of chips.

A (7:41)

Why is the horse here anyway? Whether it's going to 8 CPUs for 1 GPU or staying at 1 CPU to 4 GPUs, it's still twice as good as it used to be because it used to be one to eight. And so that is bullish for intel. And Libbuton is making that clear to the market and to the investors on his earnings call. Then there's the wild card, the terrafab Elon Musk project. It's very exciting, but it's clearly further off. Elon Musk wants to build a massively vertical integrated chip manufacturing operation with Tesla, SpaceX and possibly other Musk companies needing huge volumes of chips for self driving cars, humanoid robots and even space based AI data centers. Intel is supposed to help design, manufacture and package chips for the project. Now the Wall Street Journal has a more cautious view. Today Elon is aiming for terafab to reach 100,000 wafers a month and then eventually million wafers a month, which is just insane scale. So let's put it in context. One million wafers a month is about 70% of TSMC's total monthly output across all fabs. TSMC's largest fabs put out roughly 100,000 wafers a month into production. So you're talking about 10 leading class TSMC fabs at intel just on the Terrafab project, plus whatever else intel is doing. So a pretty massive scale up. And TSMC CEO CC Way has said that it's just not that easy to build a fab overnight and get it up and running. So fabs take two to three years to build and then another one to two years to actually ramp. And we've seen this with TSMC Arizona, which we've been very excited about, but it just takes time and there is a bottleneck. And the bottleneck has been discussed at length. Ben Thompson wrote about TSMC risk and Sirtechery arguing that TSMC needs to spend more on CapEx. And I think that should be clearer now. We'll see what happens at the next TSMC earnings call because maybe they will be waking up. But overall the Intel. Intel now has a collection of plausible demand Stories. Even though the demand itself is not going vertical, the stock is going vertical on the back of these five key demand stories that are all pointing in the same direction, directly up. One, AI agents need more CPUs. Two, AI systems need more advanced packaging. A higher ratio of CPUs to GPUs. Intel can help with that. Three, the US government wants to a domestic leading edge foundry. This is just a national mandate. Four, Musk wants an impossible amount of silicon. And five, also, the hyperscalers want more supply. And so all of that is good news for Intel. There are more companies that are getting into the CPU design space. We talked about ARM recently and arm, of course, it feels like they will be going with TSMC in the short term, but who knows what happens in the long term? So as CPUs continue to be important in the AI story, all good news for Intel. So suddenly, investors are more willing to entertain a messy, expensive strategic chip story than they were five years ago. Intel famously missed mobile, which meant TSMC ran away with enormous manufacturing volume and left intel with a DOM problem. Volume was destiny, as they say. You can't grow if you can't build the fab that's on the leading edge. And you can't build the fab unless you actually have the customers. And so if you missed mobile, you just have this gap and you have to jump over it with the help of the government and a bunch of other people and the AI narrative and all these different five key pieces. So the pieces of the puzzle are coming together now, and that's good for Intel. It's also good for America's chip manufacturing prospects. So good news overall. And congrats to all the intel shareholders that were believers early on and rode the wave.

C (11:26)

Everyone, every US citizen.

A (11:28)

Every US citizen. Congrats to all Americans, every US citizen and every US debt holder. So even the international folks that own

B (11:36)

Treasuries are in better position today. You got to think about what this does to Trump's confidence level. He's like, I enjoy a bailout. He's like, I'll indulge Spirit Airlines. American Spirit. Yeah, I think he's excited about the potential.

A (11:54)

Sure.

B (11:55)

But why? Why stop at just bailouts, right? Why not? Why not start taking a position in hyperscalers? Like, look like I think I can move your market cap by 3x. I've done it before.

A (12:06)

Yeah.

B (12:06)

Intel was. Intel was falling apart.

A (12:09)

Yeah.

B (12:09)

Imagine if I. Imagine if he applied all his wisdom to a company already winning.

A (12:15)

Certainly possible. Jim Cramer's excited about it. He said in 13 months, Lip Bhutan took intel from a possible an unthinkable bailout candidate to one of the wealthiest companies in the chip industry. There is a big three of CPU, AMD, intel and ARM. And the agents need far more CPUs than these three can produce. So that means prices are going up. And they got the God candle.

B (12:38)

Yeah, look at this candle from Candle.

A (12:40)

We love a green candle. Should be white suits today. Well, before we move on, there's some news. Justin Bieber brought back Bieber back from the dead. Later in the show will tell you what this says about late stage venture re acceleration. We're going to tie those two together. In other news, for more than a year, Silicon Valley has buzzed about Cursor's growth and whispered about its margins. Now on the cusp of a $60 billion bailout. Laura over at the information.

B (13:12)

Wait, they called it a $60 billion bailout. Buyout.

A (13:15)

Buyout, sorry, buyout. We reveal the hot Vibe Financials $60

B (13:21)

billion acquisition calling it a bailout.

A (13:24)

It's not a bailout.

B (13:25)

Their margins were rough.

A (13:26)

We were just talking about it got warmed into my head. Cursor had negative 23% gross margins earlier this year. Amir says that is low for a company generating as much revenue as it is. This is the question of how sticky will Cursor be as a entry point to AI, as an entry point to inference demand, can they reroute? It feels like with Composer they have been able to retain a lot of customers. The company is still growing and we know Cursor users who have stayed with the product while changing inference based on what their plan gives them. So their subscriber and they will use whatever the plan, whatever gets the job done for them within the budget of the plan. And so obviously you see negative 23% gross margins and you're like, whoa. And then you look at the SpaceX deal and the XAI deal and all the compute that's sitting at Colossus 2 and you think, oh well, like what will the gross margins be once, once the inference is happening on a Cursor trained model or XAI trained model, are there going to be higher gross margins? It's pretty easy to imagine that the gross margins would increase. Did you want to play this? This piece from Patrick o'? Shaughnessy?

B (14:42)

Let's do it.

D (14:43)

So clearly Anthropic is in the lead, right? And OpenAI is cooked. What's interesting is because Anthropic bounds on compute and they can only grow it so fast and sort of to the point of Dario used to gloat about how OpenAI was being too aggressive on compute and Anthropic was more sensible in their scaling. And now Anthropic is like, fuck, we should have. I wish we had a lot more compute. OpenAI is able to pay the bills perfectly fine. In fact, they raised a ton of money to get incremental compute in addition to the irresponsible levels of compute that they were buying from Oracle and core, Reaven and SoftBank and all these people and Microsoft, you know, such as Trainium. Now they're getting Trainium as well from Amazon. But what's interesting is if you were to say, Opus 4. Six, you know, let's ignore models getting better over time. Let's just take diffusion of this technology. You and I may jump on the model immediately, day one, but other businesses take time and it takes time for people to learn. And so by the end of the year, let's say a 4, 6 opus tier model, the economy would spend $100 billion on. I don't think that's unreasonable. It's spending 40 billion right now. Anthropic won't have enough compute to do that. And so. And presumably OpenAI and Google will hit that tier soon enough, whoever hits that tier next. Sure, Anthropic may get to charge 70 plus percent gross margins, but if OpenAI hits it next, they charge 50% gross margins. They still get all of this incremental demand. And probably they also won't have enough compute to serve all the users. Sure, maybe Mythos is a model where if the world had enough compute, it'd be $500 billion of revenue or something crazy. There is such demand for these tokens and such limitations on compute, you know, and we see this with H100 prices skyrocketing and the useful life of these GPUs continue to extend. It's pretty clear even the Tier 2 Lab is going to be sold out of tokens, let alone the Tier 1 lab. The Tier 1 lab will have better margins, but the Tier 2 lab will be sold out, and probably the Tier 3 lab will also be close to sold out. Economic value that the best model can deliver is growing faster than our ability to actually serve those tokens to people via the infrastructure. And so this gap will continue to grow and the model labs will continue to have expanding margins until people in the hardware supply chain, infrastructure, supply chain, are like, wait, no, why don't I just jack up my margins?

A (16:48)

Oh, yeah, I love that.

B (16:49)

I love that sound. Bill Gurley, let's See what Bill Gurley. Let's check in with bg. He says, I find this conversation foreign along with the argument that we are data center constrained or energy constrained. Historically in markets, price is a leveler of supply and demand. If you have a constraint, you price higher, you don't have surplus demand.

A (17:06)

Yeah, it's a good point that they are. The big labs are losing money, but it doesn't feel like that the revenue of OpenAI and Anthropic are VC subsidized heavily. They're not VC.

B (17:16)

Yeah, that's what I'm saying.

A (17:17)

And their customers are not startups necessarily. It's like hedge funds and banks and stuff.

B (17:22)

So hedge funds and banks that are printing by betting on semis right now.

A (17:27)

Yeah, which is a different. I'm totally on board with like, well, you know, the Mag 7 are drawing down on cash flow and issuing debt and doing layoffs and they're funding the, like, they're subsidizing demand. Right. But the idea that B.C. dollars are the biggest drop in the bucket of subsidized demand feels like, I don't know, it just doesn't quite math out to me. The actual revenues and Demand from Fortune 500 companies is so high. And the really big dollars in all of these rounds, I mean, Anthropic just raised something like 40 from Google today. And that's not a VC subsidy. I mean, it is a subsidy.

B (18:12)

Yeah. I'm surprised, I'm surprised he's not talking about like big tech subsidies.

A (18:16)

Yeah, yeah. Maybe he is using the term VC dollars like broadly, which is like fair. So if he's using the term VC subsidies, VC dollars broadly, I think that does sort of. It is a reasonable point. But at the end of the day, you talk to most companies and most, you know, paid chat subscribers and they're just like, I like the thinking model, So I pay $20 a month. I like the pro model and I use it for, to create value. I'm getting that much value in like software for my business that probably doesn't even exist. Like all the vibe coded software that we use here, like, it's just not available off the shelf. We're building like completely net new products. And I think that that's like generally what's happening in the AI economy. There's been this like discussion for a long time. Martin Scarley was sort of debunking like, oh, is this like, are these all like circular deals? And he was like, no, like the economy is so broad that you have, you know. Yes, there are. Yes, Like Google might take a position in Anthropic, Microsoft might have a position in OpenAI. But you also have completely Main street customers and just average Joes who are paying, seeing ads, buying things. There's companies that are profiting off of running ads and other inference like there's like everyone.

B (19:32)

It's one big circle.

A (19:33)

It's. Yes. It's one giant circle of life that is, you know, actually self perpetuating because it is a true economy that hits 25 different categories. Well, you might think it's over, but Elad Gil does not. He says my view is the AI boom will only accelerate and is a once in a lifetime transformation. This is orthogonal to whether many AI companies should exit in the next 12 to 18 months as some may lack durability versus labs, new entrants or weird market ships. But he's extremely bullish. He also posted a funny thing about his new life plan. He said his new life plan is to move to Brooklyn, get a neck tat, ride a bike everywhere, cold brew his own coffee, also start drinking coffee. That's an odd thing to jump straight

B (20:27)

to the was this, was he like trying, is this like a cipher of

A (20:32)

some sort of he wants to.

B (20:34)

Is there a secret, is there a secret message embedded in this post? Maybe it sounds like he's advising his portfolio companies.

A (20:41)

Yes.

B (20:42)

I don't care how hard you're ripping.

D (20:44)

Yeah.

B (20:45)

A lot of you should probably exit.

A (20:48)

Yeah.

B (20:48)

Just given, given how, how much uncertainty there is.

A (20:52)

Well, later in the show we have a fun story. A bear wandered into a backyard and took a dip in the family pool. We'll take you through it coming up after later in the show, what this tells us about edge computing. Get ready for it.

B (21:06)

Has Jane street achieved AGI internally?

A (21:09)

I think so. They definitely have.

B (21:11)

$40 billion in revenue last year, more than all the big Wall street banks. And with only 3,500 employees.

A (21:17)

Yes. Let's give it up for Jane Street. Fantastic. We have some exciting news. Out of Thrive Capital, Josh Kushner announced a new fund, Thrive Eternal, a permanent capital holding company that will be concentrated in a small number of assets that we can own. And over many decades, across Thrive Capital and Thrive holdings, we are building and investing through a moment of exponential change, backing emerging technologies, the infrastructure that powers them and the businesses they can transform. Increasingly, he sees a fourth category. These are assets with qualities that cannot be replicated by technology. Iconic franchises and cultural institutions rooted in tradition. And he says his first partnership is expected to be with the San Francisco giants, an institution built on More than a century of shared identity and community and among the most iconic sports franchises in America. I looked it up before the show started. The San Francisco Giants, if you're not familiar, it's an American professional baseball team based in San Francisco. Baseball is a bat and ball sport played between two teams of nine players, each taking turns batting and fielding. The game occurs over the course of several plays, with each play beginning when a player on the fielding team, called the pitcher, throws a ball at a player on the batting team called the batter, and then the batter tries to hit it with a bat. And so they play this game, they sell tickets to the game, and that's how the San Francisco Giants make money. And Josh Kushner, with my internal, is

B (22:50)

getting it on the CVPN style show.

A (22:52)

Yes.

B (22:53)

For whatever you just described, I feel like it would be very cool. That would be cool if you had maybe most on screen graphics and to provide kind of live coverage maybe while that was happening.

A (23:05)

Extremely educational. This is a crazy story. Brandon Gorell. And our team was obsessed with it. A U.S. special Forces soldier involved in the capture of Venezuelan President Nicolas Maduro was arrested for allegedly betting on that operation, netting him $400,000 in profits. This is when the betting on yourself meme goes wrong, right? You should always be betting on yourself, but not literally if it's against the law. If you are involved in a government capacity, you should probably not be gambling on the outcome of your government work. Shame on you. But still, even Trump was like, he

B (23:41)

was betting on himself. And then he like referenced a baseball player that isn't in the hall of fame because he was betting on himself. But Trump was seemingly implying that that wasn't so bad.

A (23:52)

People are really, really having fun with the new image gen model.

B (23:56)

Oh, this is. I don't even know.

A (23:57)

We should pull this up because it's so rude, but this guy's effectively playing Minecraft through ChatGPT because he has a. Generate him an image. And then he says.

B (24:05)

And then he says, okay, walk closer. And it just generates a new scene.

A (24:09)

Pull it up.

B (24:10)

I'm sure. I'm sure there's some kids out there that are. That have figured out how to. How to play Minecraft in ChatGPT. But this is very funny. I found this post when it had 40 likes this morning. Sniped, Sniped, Invested early. Cohere is merging with a German AI lab called Aleph Alpha.

A (24:33)

Biggest fan of Cohere. I love Aiden Gomez. Transformer Paper alum. Death Grips Enjoyer. What's not to like? About the legend Aiden Gomez. We gotta get him on the show. But this is a congratulations moment for Cohere and Aleph Alpha GPD 5.5. Yes.

B (24:52)

Kits is saying it might be AGI because if you ask it, bro, the car wash is five minutes away. Should I walk to it or go by car and says car, bro, it's a car wash. Let the plot make sense.

A (25:04)

Wait, what?

B (25:07)

Like it just immediately clocks it?

A (25:08)

Oh, I like that. It's. I've seen this exact test done in normal language, like with a proper prompt, not the casual bro slang.

D (25:17)

And.

A (25:18)

And it's. It's. It actually is more. It feels more like AGI if it is able to pick up on the lingo and mirror that back. Car, bro, It's a car wash. That's very funny. That feels very remarkable. It also got the R's and strawberry correct. Right. And how many strawberries are in the letter R? 0. The letter R is not known for its berry storage. These are good answers. The Being a helpful assistant but simultaneously rejecting nonsense has particularly been difficult for LLM. So good progress here. We'll figure out more about what went into this and what the next generation of AI impacts are from this.

B (25:58)

Taj says is found a post five and a half years ago. Elon says the rate of improvement from the original GPT to GPT3 is impressive. If this rate of improvement continues, GPT5 or 6 could be indistinguishable from the smartest humans. Just my opinion, not an endorsement. I left OpenAI two to three years ago. I'm a neutral outsider at this point. Greg says thank you. A lot has changed. But he was accurate in his prediction.

A (26:26)

Funniest outcome is the most likely.

B (26:29)

And then the last thing I wanted

A (26:31)

best bros and do a buddy cop film together.

B (26:33)

The last thing there is a new app from X. Oh yeah, X Chat.

A (26:39)

X Chat.

B (26:40)

And I'm excited to try this out because the current. Yeah, the DMS on X have been pretty brutal. Yeah, right.

A (26:49)

I haven't had that much of a problem. I do have this weird bug where when I click the chat button on desktop, it sort of loads and then it resorts after it loads and sometimes if I haven't been in it in a long time, it needs to reshuffle several, several times and so that can be a little jarring. But overall it seems like they made the migration to encryption. I haven't. I don't know. I seem to get messages. It seems to work fine. It's cool that there's a new app. I probably need to be better about answering DMs. I have a lot of them that are unread, so having a dedicated app for that, that sounds cool. People were taking shots because the Everything app's supposed to be everything in one place. Instead it's three apps to get everything and the everything app. Sort of silly. Who cares? More apps the better. I like apps. So go download it, Go check it out and go like Nikita's post because he's been on a generational run. Also, GPT 5.5 is available in the API. Now that's breaking news, I guess from Greg. So anyway, we will see you on Monday. Have a great weekend. Enjoy the weekend. Leave us five stars on Apple Podcasts and Spotify. Sign up for our newsletter tvpn.com and we'll see you tomorrow. Flashbang have an incredible weekend.