SemiAnalysis's Doug O'Laughlin: OpenAI is making itself too big to fail by locking in every chip and hyperscaler partner

valuable asset that can be sold to a different firm or something? Anyway, uh we have our next guest, our first guest, Doug Gollin in the reream waiting room. Let's bring him into the TVP Ultradome. Doug, how are you doing? Good to see you. Good to see you too, guys. How are you all doing? Doing fantastic.

Great to see you. Uh what a day. What's uh what have your immediate reactions been to the uh to the AMD news? Was this on your bingo card for this year? Yeah. Did you see the person that bought like 6 million of AMD calls like on Friday, too? No, because was that you? Uh hopefully not. I wish. Hopefully. I wish.

That's not me. I would be in super big trouble. So, that's definitely not. Um, so I I I I'm actually going to talk more than just AMD specifically. The thing I think is really funny is that like look at all the announcements, man. I have uh a very pet theory.

It's kind of it's kind of lame theory if you think about it, but I think OpenAI is just trying to have everyone involved like Nvidia, AMD, SKHix, the entire memory chain, all the hyperscalers and effectively like you know it's everyone's problem now. Everyone is all vested interest in uh in essentially open AI. Yeah.

He's making himself too big to fail. Exactly. That's There's There's a world in the future where OpenAI needs effectively a bridge round and he can go to a bunch of different people with big balance sheets and say like if you guys don't pony up like we're all going down. Exactly. That's the plan, man.

Essentially, have you ever heard this phrase like uh it's like if you owe the bank a hundred bucks, that's your problem, but if you owe the bank a billion bucks, it's the bank's problem. Yep. That's that's essentially what's going down, I think. Um, so we're seeing like partnerships everywhere.

It's kind of it's kind of a crazy um it's kind of crazy time. Yeah. Uh, what's your take on people uh like overfitting to uh the dot boom and the Worldcom story? Do you think that's like overroought at this point or are there actually good analogies now that we're in like an infrastructure boom?

It's not just valuations because we've had bubbles in like software, private markets, venture capital before and those seem to have worked their way through so easily because it's just oh some fund that needs to like wind this investment down over a 10-year life cycle.

And it's such a small part of the economy even though the numbers retail. There wasn't a lot of debt. Yeah. Like like clubhouse was like $600 million which is just like nothing compared to the global. Who cares? Yeah, there there's uh there was a post that went super viral.

It was a screenshot that was tracking like the NASDAQ in in in the dot era to the NASDAQ today and it like lined up absolutely perfectly and then it turns out it was just completely fake news. Like we tried to recreate it a bunch of different ways.

There's some somebody like basically committed a chart crime and it went super viral and everybody's like, "Wow, it looks exactly it tracks exactly. You can't do that, man. It never works. Let's put it this way. It never works. Um, I have a post I just uh Okay, one, I wrote about the 2000 telecom bubble. Sure.

And two, I just wrote something that was mostly behind a payw wall, but like you could just look at the cover image and it's just like you are here. It's like the first interest rate cut was in 1998 in September. And I think that's that's probably our best an analog.

Like I think that that's probably the best one to analog, which means it gets a lot crazier from here. That's my belief at least.

So if we look back to 2000, all the analogies and eras back then, I think the one that is kind of scary and worrying is like the the vendor financing, like that's a lot of capital for for your vendor to be putting up and that kind of that's a little scary.

Like I I think it's getting a little circular, but like it's not it's it's definitely not tech bubble crazy because the valuation would have to be at 2x from here. That's kind of the thing that's like a little different.

And I think the other thing that's also kind of interesting is like last time you could argue the entire tech bubble was kind of very focused and concentrated on the like a very small amount of unprofitable companies that really didn't impact the economy when they all went bankrupt.

The difference is today who's the primary participants? It's the seven largest companies in the entire world. So that's that's kind of the the thing the push and take here that I think is really interesting.

And now given what we publicly heard with the capex announcements, it's going to be a lot of capital to get from here to there. And I think um I think it's going to be interesting, man. But honestly, if you look at everyone's balance sheet, no one's levered. That's the thing that I think is really interesting.

No one is levered at all. And like that's the that's the next leg. Yep. That's the next leg. Uh I was giving a a sort of a hypothetical example around the AMD deal to John. Uh, and tell me if this is just stupid or or too simple, but I'm walking around a farmers market. I see a stand. A farmer's selling some vegetables.

I'm not that interested in the vegetables, but then I think, what would make me interested in this situation? Well, and then I say to the farmer, "How about you give me 30% of your company and then I'll buy some vegetables? " Is that at all track with the the the the uh this deal?

Yeah, I think it's suddenly I'm interested if I if I if I can get 10% of your company, I'm interested in being a customer. Yeah, I think it's also also I want to I want to highlight something that's really interesting is look at the different deals between AMD and Nvidia.

Nvidia you essentially have to um you know they're like okay well uh please do an investment in us so we can keep buying your chips. Yep. Uh and then the vice versa AMD is like no no no you must invest in us so please buy our chips. Like that's the discount and the premium if you think about it, right?

Like Nvidia, you get the premium because it's like it's a better chip, higher gross margin. They're like you're getting so much of our money, you might as well just be an investor. And then on AMD's side, it's like, dude, you gota we don't want to we don't want this. So like like give us something in return.

Can you take me through the Yeah. Can you take me through kind of the view on on AMD generally over the past year? I know that semi analysis was pretty crucial. George H. was talking about some uh some like bugs that were in uh the actual like just trying to run large models on AMD chips was not going well.

Uh your team was pretty integral in that. It seems like they're moving pretty aggressively to try and catch up. Is is that is that gated by money? Like what's the what's the recent history of AMD in terms of like getting back in the game materially?

Um, so I do think they're much more competitive than they have been in a long time. Our team's definitely been working. Uh, watch for some stuff from semi analysis soon. TM I can't say anything more. Go subscribe. Uh, yeah, but we're definitely we're definitely continuing to do some work in the space.

I I think um, look on the smaller models it just on the imprint side it seems like there's not a lot of differentiation, but on the bigger extremely large world size there is.

So GB200 still law of the land but the Helios 450 rack which is going to be this next generation we're a lot more excited about mostly because they get the ginormous discount of the equity.

Um it's just kind of an insane thing what they've done man like it's such a big it's such a big amount of the the equity possibly on those penny warrants. Like I mean it's kind of crazy. It's like hey uh you know you buy $40 billion dollar of product you get $20 billion product uh $20 billion investment for free.

Like it's kind of this crazy like buy one get one deal. So, I mean, I think that that's um I I think I think they have a chance. They have a fighting chance.

The other thing that that was interesting is it feels like OpenAI can have various partners by AMD buy AMD chips, but then OpenAI gets to ex like get to count that towards purchases themselves.

In the Wall Street Journal article, they made it look like Oracle buys a bunch of AMD chips for a data center that opens warrants. Open to exercise the warrants and those count towards AMD purchases basically even if it's not on on OpenAI's balance sheet. Yeah, pretty much.

I think the way to think about it is it's a OpenAI infrastructure bet. Everyone who is in the OpenAI sphere counts they're and they're just trying to get the most money, the most power, the most pole into it so that essentially, you know, they're kind of creating gravity. Everything comes to them.

Honestly, there's like a really good analogy in 2020. Uh, Nvidia bought like 5x more capacity than everyone else in at TSMC and so there was no room for anyone to compete. Kind of the same thing. They're taking all the oxygen out of the room for anyone else that isn't named OpenAI. What do you think about this?

Uh, I I I feel like I heard this on Transistor Radio recently that uh GPT5 is still potentially running kind of like GPT4 class model under the hood, doing a bunch of test time inference on top of it to get you a better reasoning model. But like the core architecture actually hasn't shifted that much.

We're not we're So does that make it easier to port that model to AMD? Like what is the actual barrier?

We've heard about the CUDA moat, but is it something where like you can underwrite a hund00 million of new kernel writing to run on AMD even though it's a huge hassle, but it's because you're still amvertising that original training run that you did on NVIDIA to inference it on AMD going forward with this new deal.

like is there something where the the pre-train is stabilizing to the point where porting to a new architecture is more justifiable? I I think that that feels like a stretch honestly.

I wouldn't I wouldn't know one to one on this one like you know if this was a world with no reasoning then I would say yes definitely yes right um but I think reasoning really does change the game and how you or how you orchestrate these larger and larger context size windows and then multi-nodes because that's a really hard game too like multiode reasoning or multi-node inference is very very challenging and so that's I don't think you get a perfect one to one but like let's also be clear AMD has caught up quite a bit in the last year like it's a big difference and like That's I I think it's enough to maybe take a bat and like you got to also think about the bigger context here.

You know, we're talking like what $20 billion of revenue mostly 2027 story. It's kind of peanuts compared to the numbers that we're actually talking about from everyone else. They're still fourth place. Let's be clear.

Like this is this is commitment and I think you can see maybe a little bit momentum and inertia, but they're still in fourth place. Four fourth place uh with uh Nvidia TPU and Tranium above them. Yeah, even tranium even tranium is above them. Interesting.

How are people thinking about actually valuing open AAI at this point? It feels like you have all these like circular if you're I was trying to build a DCF for open AAI to underwrite you know investment at 500 billion.

It's like okay well now I own parts of AMD so I need to build a full DCF of that but AMD is linked to OpenAI and I have all these like circular transactions. Are are are investors like grappling with this or is it all just like vibe based investing what you've seen?

Um, I think it's I think it's a little bit of vibe based investing. I think the the real play here is uh TAM is big. It's really really big and it's going to get be a lot bigger.

I think the other thing that's really interesting is some of the agentic purchasing can really unlock very very very large parts of the economy that previously you could argue they had nothing in. Right? If you're talking about like you know the the partners they had available today for example uh booking.

com trip advisor you know we can be talking about you know and this is like some some 2021 math right but like 1% of all travel you know 1% of all groceries 1% of everything and you're like that's a big number that's a really big number um and so and you know that's that's going to be rev so so I think that that's the that's literally the plan I mean trying to become the super app and that means that they're going to be integrated with everything and you saw on the partnerships they had they try to integrate as much as as possible with every every player of the stack and so I think this agentic purchase push is going to you know if we start to see revenue from it soon which I think they're they're definitely shooting for will kind of open up the TAM story and start to monetize the free users because there's really only like a hundred million paid users it's not that much but there's hundreds of millions of free users and we're talking about a personal assistant who could plan for you.

Um, you can do something crazy as like, "Hey, I need a, you know, I I thought I lost my travel bag or something. " So, I was like, "Hey, I need a small dot kit for on the- go for everything. Could you just like, you know, make a travel size and order it and get it to my house tomorrow?

" That seems very like it feels like science fiction, but that is uh very much in in like the next 12 months. And so, hey, one to 3% of that transaction, that's a lot of money. And I think that that's the real way forward. is becoming so integral into everything that everyone has at stake in you. Yep.

So what uh so the the AMD order and the fullness of time is around 6 gawatt. Where do you expect that energy to come from? Do you expect Sam to start doing crazy deals like this on the energy side? He's already has Oaklo which is just uh you know up. He's not chairman anymore. He had to step down.

But uh I I I knew he wasn't direct. I mean, I'm assuming he's a major shareholder still. Uh, but so that's kind of one angle, but where else where's the power going to come from? Dude, uh, that's a trillion dollar question.

I feel like uh, definitely there's a lot of ways to But it's an important it's an important one, right? If you're trying to if you're trying to underwrite and say this is real, at least for AMD, it's like, okay, well, how real can it be if there's, you know, dude, it's okay.

if we can't plug them in, you know, there's a lot of ways. There's a lot of ways. Okay. So, um let's let's first talk about like like the grid is actually not built for um 100% production. So, there's a lot of slack in the grid itself because it's actually built for the highest day and the lowest day.

It's built to like have a 90 you know 99% uh margin rate u you know confidence interval on the absolute high and the absolute low. So the the peak of the power and so gas or not gas but generation throughout an an average day is at a much lower utilization than 100%.

So I think the real reason and and something you're going to start to see is that people are going to like essentially the grid is going to kind of start to become a lot more fragmented and there's going to be a lot of behind the meter p power generation that happens at the data center.

And so in that case what you do is you'd sip from the grid meaning that you can use the grid when the grid is not totally utilized when it isn't you're doing gas gen on site. That's probably the thing that is um most I mean it's gonna be huge.

It's going to be huge like natural gas uh you know energy consumption is just going to go through the roof. So that's going to be like the big driver I think to get these data centers online. I think it's going to take a lot of time. Um and also we're seeing like grids breaking right now.

Like grids are just like grids are not keeping up and that's kind of been um I think it's going to be a big story in 2026. But I I I also think that that's where we're going to see international purchases too just because like we're running out of power in the United States.

I mean, I think what what's also interesting is like even though we say we're running out of power, uh, when you have this big demand response, supply comes out of the woodworks.

So, we're starting to see stuff like gas gen, gas turbines, like all these different like little clever pockets of energy are coming up to then fulfill the need. I expect that to continue. Effectively, if you have enough demand, well, you know, supply will find a way. Yeah. Yeah.

That definitely feels like the next wave here is we're going to see a hundred million hundred billion dollar deal for a new Hoover Dam and more nuclear actual buildout. There was some news about this with uh some of the hyperscalers uh trying to recommission nuclear power plants, but it just feels like it's so slow.

It's not on the order of like in like what TSMC is fabbing next quarter. So we kind of lost track of it, but uh it does feel like that that's going to be the next chapter and maybe it'll be a little bit longer lead time, but uh you got to start thinking about it now.

Where where do you expect the leverage to come from if that's the next leg of the journey? Is it net new private credit fund formation? Is it is it just banks getting more and more active? Like where where's the capital going to come from?

Okay, so one, let's not forget there's positive free cash flow from all the big guys. Yeah, that annually is something like, you know, I actually have a spreadsheet, maybe even open right now. Um, talking about the total firepower.

We're we're talking quite a bit of capital is available to do this, like a surprising amount. I want to say on an annual basis, it's something like 250, maybe even yeah, 200 billion. But the thing that's also really interesting is none of the hyperscalers are levered at all.

There's a really good paper from Double Line talking about like would you rather loan to the United States government or would you rather loan to Microsoft. So the real like you know crazy town is if you start loaning to Microsoft instead of the United States government because Microsoft has net cash.

Microsoft also uh you know has become more creditworthy over time versus I would argue the US government has become less creditworthy. And also what's really crazy is that Microsoft has uh you know a five bips premium to the United States Treasury. So you're effectively, you know, their rates are risk-f free.

Um, and so what's possible, I think, in terms of like now I think the real issue is like how much liquidity could the actual credit market handle? I don't know if they can handle like $300 billion.

Like that's a [ __ ] ton of money, but like we're talking, you know, there there's things that can definitely come out of the woodworks. Like private credit is going to be part of it just because they're sitting on a on like an ungodly amount of capital and they have to deploy it.

Um, I definitely think the hyperscalers for me are the one that are most interesting. Like we're talking about I mean dude that's crazy just like like you say five years.

Yeah it's funny you say 300 billion because I'm thinking I I was just doing the math on 250 billion in free cash flow if if that's used to service 5% coupon payments.

Uh that's five trillion in notional debt like like value if you're just saying like what what like have you we want to know where the where the systemic where's the systemic risk going to come from? I don't know. Yeah. when they all run out together.

That's the that's the problem is like what happens is like it's just think of it as like one long debt debt fueled binge, right? But the thing is you're looking at the debt fueled binge and I can tell you confidently today there's not really much debt to go around at the big guys.

So, you know, I think I think we could right now today it would get and also I don't think everyone's going to get to like three times leverage. That's just insane. I don't think anyone wants to do that. Yeah, but I definitely think there's a there's a way to get to uh $800 billion of credit.

Like I mean it would be insane, but that's one and a half turns from Meta, Microsoft, Amazon, and Google. And the two that I think are most interesting is uh the founder led ones like Zuck and Sergey. They control the companies. They have voting shares. They can do whatever the [ __ ] they want.

So if they if they woke up tomorrow and said, "Get me, you know, get me a turn of Ebida on this bad boy," you they could get it. Now, um, that'd be really crazy because, you know, they make a crap ton of money. I actually have the number right now.

So, right now, I think, yeah, Meta's cash on hand at the end of last year was around 77 billion. As of the end of uh, June, it was down to 47 billion and just will continue to drop throughout the year. So, they they already did that. I think it was like a $30 billion. They're not even close. They're not even close, man.

I mean, so, okay, I have the numbers right now. Net debt is 17 billion. Net debt for Google is56. That means they have more they have more cash than debt. Oracle is 94 billion, which is a crap ton.

Like, you know, they're pretty much they're not tapped out, but they they really pretty much evid to go up for them to raise more. Meta's at 2. 4 billion in debt. Um, and then Amazon's 58 billion in debt.

So like Meta's Meta and Google specifically, which are ironically the ones who are founder, they have so much [ __ ] they have so much free cash or like they have so much c uh cash they could raise if they want to. They're both have they both have like they gush cash and they're both founderled.

Like that to me is, you know, that's that's the next leg, man. And and also you have to think about like sorry, who doesn't want to lose? Yeah. Why are you not putting Nvidia in that camp? Founder, they have 50 billion in cash. they, you know, are no are not levered yet.

Like they could also lever up and start investing more aggressively, right? Well, they're doing they're they're enjoying some customer, you know, demand guarantees which are which are Yeah. What's the story you're telling around around Nvidia? I think I think Nvidia's focused on a different level of stack, right?

Because Nvidia is like, hey, we are an infrastructure provider and you know, we're have like all these long lead time things need so much capital and time to get there. The thing that Nvidia wants to backs stop in my opinion is the Neoclouds.

And so you're already starting to see that with like some of the the Neocloud purchases that are h or like the neocloud backstopping that happens at like um Lambda or even Core to a certain extent extent.

What happens is they buy essentially flex capacity for research GPUs for Nvidia so they can get better at programming like gem essentially.

But I think the thing um yeah I think Nvidia is much more focused on their part of the infrastructure ecosystem to be a backs stop instead of purchasing GPU hours right they're not they're not in the market to you know they're a seller and so you know if they if they woke up one day and said hey we're going to just start buying this stuff in mass to make model you'd be like what the [ __ ] going on um I I I just wonder I just wonder if the Nvidia story like if they get in on this like debt boom it it instantiates itself as like they lever up and then and then they basically have instead of 50 billion to throw around they have like 200 billion to throw around and then that goes into equity purchases of companies that are doing uh that are do that are purchasing GPUs and it sounds really circular but like they are just another vehicle for credit to absorb like credit interest because they're such a big company.

Yeah, I do think that that's possible, but I think I I just don't think I mean you see that in some of the smaller investments, but I think they're really focused on the neocaus and the data center side, like the infrastructure around their ecosystem. Yeah.

Do you think uh do you think Satia might be feeling like he went a little risk off too early? Yeah, I think so. And if and and this guy over here is listening to Transistor Radio, so he uh Microsoft's back.

Like I think they're I think they're going to step back into the market in a big way like uh especially on theou after open AAI and also MAI they're starting to be very focused on like I expect them to be back in the marketplace now effectively but I I still don't you know so I think what happened is they did this giant buildout in 23 where they said yes to every single leasing opportunity possible and then they like there was a top down pause and now they're starting to pick up the pieces being like no no no actually now by pausing we have to pick back up.

So I think um Satia definitely feels like I don't know he was definitely in first place for a long time and would have if he just kept pushing I think he would have kept in first place and now you have like you know Oracle popping in and being like yeah like I'll essentially steal all your business if you let me.

We'll come in at a lower lower price. Like I'm just going to lever my entire balance sheet for it. But we're talking about like the four-year guide is a hundred billion dollars, man. Like that's insane. So that hundred billion dollars arguably should have been all Microsoft's.

And that's, you know, that's that's up to them. And they will tell you um in public meetings that like, you know, it's lower ROI for them. And sure, that's that's the case. But I think Microsoft is definitely being conservative and they have the biggest balance sheet.

So they're I I kind of don't expect them to be the first player. I think of it as like kind of like um what is it called? Like OPEC. You need a slippery person to essentially defect. And the slippery defectors is meta and Google in my opinion. Those are the two biggest.

And so when they start to really get into the game, all of a sudden they're going to be like, "Whoa, whoa, this is attacking my business model and I'm losing market share. " And that's when um that's when the other guys come back in in a big way. Yeah.

Uh do you sore is going to move the any of these like GPU buildout plans? Like Sam was getting kind of mocked for saying like, "Oh, we're going to have to pick between. " I was to be clear. Yeah.

I was the one saying I said uh I just thought it was I I mean I thought the line the timing of going out and saying yeah there's a very real possibility in the future that we'll have to choose between curing cancer and free education for everyone and then a week later we launch you know the most intensive slop. Yeah.

Um but but I mean we got some we got some news on API pricing. It sounds like it's 32 cents a generation. like it it it feels like it might not be moving the needle as much on the on the inference side or actually like like is Sora going to make them GPU poor? I I don't know if you have a view on that.

It's 30 Sorry, sorry. It's 32 cents per inference. What's the time frame on that? Uh it's Oh, it's per second. Per second. Okay. So, it's almost three bucks for a for a 10 for a 10-second clip. Okay. So, that's actually like on par with V3. That's pretty expensive. That's pretty That's pretty intense.

That's pretty expensive. And they give you what 10 credits a day or something. So, if there's hardcore users out there, they're generating $30 a day. What What is that? $1,000 a month. Like, it it it's it's it adds up. If you get a million DAUs on there, like that's a lot. Yeah, I think they're they probably have more.

They they have the potential to do much more than a million DAUs. I mean, they're they're they're definitely compute rich, man. The the Pulse thing. Have you guys Have you guys been doing Pulse at all? They they generate those images for you, and you're like, "Oh, a little personalized slot feed for me. " like that.

Um I know on on launch day we were reacting to it and John John was like John was very impressed. For me it was like three out of the five images that I looked at were like great and then like two out of the five I'm like you could have just not generated this one. This is just absolutely garbage, you know.

But yeah, um what do you what do you think about what do you think? We had a question in the chat from uh Bobby. He says uh asking about like what what do you think about public sentiment risk?

It feels like I've never seen more hatred of AI from from non- tech folks than than in the last uh week or so and and you're seeing data centers data centers just getting kind of like blocked.

Uh that feels like a potentially you know it's certainly going to be a roadblock at a bunch of different stages of this buildout. Yeah, I think that that definitely is a newer uh roadblock that is getting more real because let's be real, up until now, no one cared. It's like such a small part of the big picture.

Um, but I still think the reason why I think we're we're going to go bravely into the future is everyone else seems to be pretty on board. And I think the one that really helps is the Trump administration is like super duper on board. Um, they're like, "Hey, whatever.

we're going to tear off the [ __ ] out of everyone, but uh we're going to we're going to completely make up for the weakness in personal consumption by massively investing.

And you know, it's like we're going to get the biggest check from every company in the world, every nation in the world, and the only thing anyone wants to spend anything on right now is GPUs. And so they're like, "F it, we're building more GPUs. " And so you're you're seeing all these huge announcements.

And I think that that's the you know, that's that's kind of the thing that will kind of push it back. I mean, if Trump is being like, "Dude, from the government itself, I'm blessing you to buy more GPUs. " That's kind of, I think, the thing that really holds it together and makes it possibly much crazier from here.

Um, the public backlash thing, I think that it really comes back to you have to have that feedback loop work all the way to the right decision maker, not just uh some random person like, "Yeah, I'm seeing that read it on the news, too. " But like Mark Zuckerberg or Satia has to wake up and be like, "What am I doing?

" Right? Uh right now everyone in their camp all the closest advisers and trusted people are being like uh spend more spend more. The government uh you know Trump is Yeah. I mean the dinner the dinner everybody wanted to say the biggest number right.

It was even even Zuck threw out a probably bigger number than he meant to because he heard some it seemed like he just heard someone else say a big number and he didn't want to say a smaller number. Oh how much am I good for? How much are you good? He's like he's like on his phone right now. He's like your number. Okay.

I have a I have a question one one more question. Uh how long do you think XAI can maintain this level of inst you know BA basically like this level of investment without being able to back it up with real revenue growth.

I because it seems like this last financing they were having to tell like the journal or some other publication like we're not having any trouble fundraising which usually means like you're having not the easiest time if people are kind of poking around.

I I mean okay so the same way I'm not going to bet against Sam Alman's ability to raise definitely not going to bet against Elon Musk's ability to raise. Sam Alman is like the newest kid on the block, but if there's someone that like you should never essentially short ever in the history of time, it's Elon Musk.

And the other thing I think that's like much more important with that is Elon's back in the seat. He's the CEO of Tesla and he like actually has the compensation. So he's like, "Fuck it. I'll ride. I actually I'm very locked in. " Yeah. Um and you know, if we're talking about Mag 7, you know who has a lot of market cap?

Um Tesla. Tesla. Oh yeah. So things that's what I'm saying that thection Yeah. It it just feels like I wouldn't be shocked if a year from now XAI was a part of was merged in with Tesla. 100%. 100%.

And then that you know getting and that would actually be a Tesla stock would pump on that even though people wouldn't Yeah, dude. But Xi would would crush because then they're then they they they're all of a sudden attached to one of the largest financial entities in the entire world. Totally.

Like that's that's a you know we're talking trillion trillion dollar market cap, right? Like that's real [ __ ] like you can take that and turn it around. You can loan it out and you can you can move big checks, dude. It's very very non-trivial for them to raise 1020 billion off. Yeah.

What what I was getting at was basically this last financing where they're going and raising capital at 200 billion and those investors have just been getting pitched anthropic at like 170 with crazy with a crazy revenue ramp and a bunch of you know real real you know traction in codegen.

the investors are looking around being like, "Okay, like I'll I'm gonna I'll give you the cash, but like I I don't know the next time around if there's going to be the same level of appetite. There there's some really big pockets that back um Elon.

I I I wouldn't bet against the Elon sphere because you have to remember he made a lot of billionaires out of like you know, for example, Larry Ellison is always on the phone. I don't know if you knew that.

He's he, you know, he was good for a lot of X uh for X when they took down Twitter and so you can see a lot of big uh people come out of the woodworks and like Larry's flush now. Yep. You know, he's he's got a lot of equity to spend. Okay. Like, you know, these people can come out of the woodworks.

So, I definitely I had a I had a theory a theory that that we were kicking around earlier that I could see it because Larry's was a big investor in the original uh X takeover, right? There's this iconic text exchange where Elon and Larry are talking and he's like, "I'm good for a billion.

" And he's like, Elon's like, "Okay, you might want to do two or something. " Yeah, whatever you recommend. And so I could see a world where X, the social platform, does not go to Tesla and get spun into the basically Paramount, this whole like media conglomerate, right?

because I I just don't, you know, especially with XAI a part of Tesla, it's not like Tesla's sitting there and being like, we need to be able to pump timeline posts instantly into every Tesla in America. It just stops being quite as strategic, maybe. Yeah. Yeah. I mean, I I I don't disagree with that take.

That's a pretty good take, but I just think that like, you know, Elon will find a way. Whatever way that that that it needs to happen, if that's with Oracle, if that's with Tesla, like Elon will find a way. I believe is there is there significant asset value in just Colossus 2? Like it it is the biggest supercomputer.

It's this massive data center. It seems like it was very hard to build. It seems like all the hyperscalers would love to to to have it. Uh Sam would love to have it. If you just ran an auction for it, do you have an idea of like what the value of Colossus 2 would be if there was like breakup value there?

Um, you know, at today's value, I I mean, I would say it's probably probably like tens of billions. I mean, maybe not tense. I mean, let me I would say tens actually tens tens because it's we are we are are we including clauses too? Yes. Okay. Then yes, tense. Yeah. Yeah.

I I'm just thinking like like the XAI assets like sure the business isn't necessarily like printing profit with some massive consumer product market fit. Someone would be really stoked but someone would be stoked and there's some value.

It's interesting if you think about whatever NBS is just that deal with Microsoft literally scale that up and that's that's probably what you would buy on a TCO basis for Colossus like there's demand. So okay. Okay, man. You got one one last question. You got to go. Okay. Yeah, hairrained uh hairbrained theory.

Uh I want your feedback on this hottake. Uh there's a lot of discussion over the value of TSMC in Taiwan risk. Uh but we now have the golden visa. $2 million to bring someone to America. For a hundred billion dollars, you could bring 50,000 people.

Is there is there some world where we get into like AI talent wars but for process knowledge around semiconductor manufacturing and and we see some American entity start poaching TSMC employees in mass to get them to America. Um that's definitely possible. I just don't feel like it's probable.

People have been saying there's a supply shortage yet like TSMC has a lot of like has almost a captive audience dude. Like if you think about it, TSMC is like, you know, I mean, it'd be like working for the government almost, but like even more prestigious and killing the [ __ ] out of everything.

Like TSMC is like the company of Taiwan. Dude, honestly, TSMC engineers are like really underpaid as is. Like, don't tell don't tell them, right? But like um on a on a on a dollar basis compared to even American peers and a lot of people like like China for example, poaches very heavily.

So, I think there's definitely a chance, but I feel like you have to take the entire know all like it's not just the tool. It's the tool, the process, everyone. You'd have to take the whole fab. Um, but yeah, I think that'll be I think it would be sick, dude. I think if Intel Intel could do it, right?

Intel should do it. Intel should do it because they have the golden visa. They have the backs stop from Trump. They have all the investors coming in. They'll probably have an open AI deal soon. Who knows? And then they have all the money and then they can just bring people over.

Uh, anyway, we won't take any more of your time. Thank you so much for hopping on the show. Super fun, Doug. Always great. We'll talk to you soon, guys. Yeah, talk soon. Bye. Uh, let me tell you about Turbo Puffer search every bite. Serless vector and full search from first principles and object storage.

Fast, 10x cheaper, and extremely scalable. Uh, you also heard