Ep. 044 - Is NVIDIA Over? EMERGENCY Pod with ChinaTalk | Dylan Patel, Jordan Nanos, Jordan Schneider

All right. Hey, Dylan. Hey, Jordan. Thanks for joining. Did you guys hear the uh CEO of the world's largest and most important company just resigned?

Nuh-uh. [laughter]

I I mean, there's some allegations, some some pretty serious conduct violations here. He was uh apparently you can't sleep with your subordinates and uh assets and keep your job.

Yeah.

I think the trillion dollar class might be fine, but once you get to five trillion, then you have problems with this sort of stuff. [laughter] Then then woke 2.0 really starts coming after you. Oh, 3.0. Yeah, we're we're rotating back.

I mean, I don't know. Like, you you like look up to some guy, this guy, and then like, you know, he does these things. It's not fun, you know? Yeah.

You can edit you can add a well this I think it's a crossover one right so we we should talk about the future of AI a little bit too but we open with match and rasa let's start with match match and rasa so match act for those who have not been tuning in to my matchack shilling over the past few months is this idea that um uh basically du uh advanced duv uh lithography alongside EUV which already hasn't been able to get into China um shouldn't be able to go into allegedly shouldn't be able to go into China and Rasa. Uh slightly different. It is uh uh aiming to to close the quote cloud um cloud access loophole um by which uh you know Alibaba, 10-centent, bite dance today currently can legally access um GPUs which are banned from being ex from being uh imported into China by setting up these you know cutouts in Malaysia which you guys have done such a great job of sort of untangling. Um, Jordan, do you share uh uh Dylan's skepticism about uh Rasa and his potential fallout? Or maybe Dylan, why don't you why don't you expand on on why you think it's uh such a bad thing?

Um I think you know the what is the US national interest is it it is to a maintain a lead on AI as large as possible and b control China's access to AI as much as possible. If your goal is to lead, maintain a lead as large as possible for as long as possible and maintain control over China, you do not prevent China from having any compute because that actually means that the US will have no compute too. If you full ban anything in my world, I believe that China will just also start enacting economic war. I don't know if you remember in 2022 I stayed up all night and I wrote um the October 7th chip restrictions regulation um and I said China and USA are officially economic war. Um or I said it's a new cold war. It's an economic war. Um and in there I found like 20 loopholes that I posted about that very night. And I'm just one dude, not an army of lawyers. Um and was hoping they'd get patched. Many did not get patched and they still are not patched. Um but like one of the key aspects of this original restriction which I think is you know mostly true which is um there's the civil military fusion within China um the Chinese military and state will and can apprehend any resource to for military and intelligence purposes even if it's a completely commercial thing right we've seen it with things like doin and and and uh Tik Tok already and that's the most consumer you can get um so clearly they do utilize it um at the same time they will also enact and respond in certain ways. We've seen this with the trade wars. We've seen this with some of the stuff on the auto tariffs. We've seen or auto um you know shutting almost shutting down American auto plants, things like this um threatening rare earth minerals and things like that. There's been a lot of actions that China has threatened or done some and then pulled back um to show their fangs. Um, and so if you ban chips completely from China, including rental and deployment, there is no reason for China to not just blow the whole thing up, right? You have to have some tit for tat. Now the optimum thing for American policy is what is the minimum amount of compute that China can have with the most amount of American oversight to where they will not blow up the supply chain because we have we do not have the stomach to disentangle the supply chain very clearly. We just don't over and over and over again. And so given that you cannot have China have no chips period both internally and foreign. that will blow up this blow this all up. But to minimize the amount of compute and maximize the control, you can have Biden style diffusion laws where it's certain companies and yes, it's a basically a bill written for hyperscalers, but hyperscalers can rent to Chinese companies and companies with American sort of where America has like power over to be like, hey, turn it off at any moment. You want all of their compute to be like that. Um if and if you allow Alibaba or Oracle or Bite Dan or Alibaba or Bite Dance or whoever it is Moonshot to rent compute from American companies outside of China and that is their main access of compute. This is supremely awesome because China will not want to send their data externally in many cases and if they do probably stuff that the US intelligence apparatus can can handle. And two, if [__] goes tits up you can turn it all off. Whereas, if you turn off remote security access, then there's even more incentive to a smuggle, b um, you know, pre, you know, ramp your own domestic industry. They're going to do that anyways, by the way. Um, which is why you need the match act. Um, but c you you you sort of really prevent any any like reason for them to continue the status quo, right? which is a lot of Chinese industry is dedicated to making American compute at least in the electronic space for American companies. Um you know the largest PCB supplier to Nvidia is a Chinese company for example. Um and it's not just Nvidia of course. Um so I think I think you have to come out come in it with like a like a game theory perspective and the best thing for America to do is allow remote access of GPUs because the alternatives are much worse.

Yeah. I mean they didn't in 2022. They didn't in um uh well 2022 the the loophole the rules had so many loopholes. Yeah. They didn't Okay. They didn't in 2023. They didn't in 2024. Um and then I think like look Trump like kept doing the war in Iran even though he [laughter] lost him 10 Senate seats. Like I think there I think there is a aspect of um pain especially in a sort of second term lame duck when um you know all all the cabinet is leaking against Lutnik because they're pissed and they've already you know there's an Axio report that said that like China's already breaking the trade deal and they're already doing economic warfare against the US. I don't think that, you know, escalatory dominance is necessarily something that should be completely assumed on on the part of of the Chinese government.

The other argument I would make is that in Rasa um if China has access to compute outside of China, they are now paying humongous margin dollars to the US ecosystem. Right? If you take a megawatt of compute outside of China, um there's multiple layers of margin stacking. First of all, Nvidia is making, you know, um, you know, $7 million of profit per megawatt that gets deployed. In addition, the cloud provider is making another 7 to10 million of profit if it's selling at today's pricing. Um, there is a tremendous amount of profit that we get to get which companies like Anthropic, OpenAI, Google, um, you know, Amazon, Meta, etc. do either less you know they don't have the cloud or they're deploying their own chips that margins the cost structure is so different why does China win they have economies of scale and their cost structure is better why does American AI win we have our cost structure is better and our economies of scale is higher and so you know allowing them to ship all this money out right if bite dance wants to compete with meta in the global market on video recommendation algorithms meta has a fourth of the cost of compute potentially um if you don't let bite dance directly by the chips, but you do let them rent them.

I think that um there's also a reflexive nature of this, right? I'm just gonna I'm just gonna devil's advocate this [__] for a while because I I think we're generally pretty close, but it's fun to do the devil's advocate just so people hear. Um you know, and this is this is sort of what Jensen's argument was to some extent was, hey, you know what, AI uh AI diffusion across the US is largely aided by Chinese models, right? If we look at total AI ARR of the AI ecosystem, um you know, if you add up all the application layer companies that are running on open models, it's actually well north of $10 billion, right? It's not like it's some like tiny little amount um of open model revenue that is going to American companies. And Deepseek makes Deepseek for Flash and their revenue is less than [__] base 10 and fireworks and and and all these guys serving the Chinese model. actually America is making so much money off of the Chinese models and it's like amazing and it lets us acrue and invest more and more and more. Um you know yes bite dance bought a bunch of ads on Tik Tok and Zuck didn't really see the threat immediately. Yes, of course everyone's shortsighted. Everyone makes mistakes, right? You know Jensen got a little horny, right? It's his short-sighted mistake at the end of the day. Um but but it's um

It's fully legal. I don't think it's not I don't think it's kind of legal.

I don't think that's why he's losing his job. I I think he lost his job because there is some evidence that he had, you know, had that they knew the chips were going into China% potential the affair.

Yeah. And and look, man, I mean, circumstantially around Jensen, um, his Taiwan sales manager is indicted for smuggling uh, chips out of Taiwan. His closest OEM partner, Super Micro, had their COO uh, and co-founder or, you know, head of sales, whatever, Wall-E, arrested. Um, one of their sales managers is currently a fugitive in Taiwan. So, it's not necessarily like [laughter] uh out of the blue that they are cracking down on this and that people around him are uh being caught with clear evidence, right? Like the picture of of uh the Super Michael Taiwan guy with the haird dryer in his hand moving the stickers from the fake servers onto the real ones or from the real ones on the fake servers audit in September of last year. It's pretty like damning evidence that they're aware of this.

Well, I think the the evidence that's required that needs to pass test would be Jensen actually knowing about this and Jensen being complicit in it and doing anything. But either way, if I had to guess, I bet the board has evidence here, right?

If I had to guess, the board has evidence on this. I mean like how do you explain no longer reporting Singapore revenue? Yeah.

Yeah. like the circumstantial evidence is there and when I I think this is the the big thing is like people even in the case of his these affairs they would point and say well it's not technically illegal except the spirit of his actions is against the interests of other people who have a different opinion about the relationship that the US should have with China and the technology that they're working there and a convenient time right right around the midterms is when he gets whacked so

Yeah. Basically, I mean, it's pretty clear Lutnik always you, you know, and and Trump wanted to make deals and they used export restrictions as a deal making thing. But like, you know, in another sense, right, like it's the [__] law, right? You can't negotiate with the law. You make a law, you make a law, right? And you you enforce it. Um, and so there's clearly like issues there.

Really, the diffusion laws aren't like they didn't remove them. It was super weird. Like he had like a comment and there was like a thing and a footnote, but they did not they never repealed it in 2025. They you know uh

Wait, sorry. These are laws in effect. I didn't realize the diffusion. I thought we would have to repass them.

It's not really a law in the traditional sense. It's like a executive action. No, no, it's it's like guidance to the BIS basically.

Yes. And then he never published like a real Kesler never this Kesler. He's just he I don't know what the [__] he's been doing. Apparently, he shows up in the office, but like you know, he never passed a real piece of guidance saying we are no longer doing this. he just like made a footnote somewhere which is like by the way guys don't worry about this like we're going to bring a new thing for you soon and that was in May of 2025 so no it's it the level of like indecision and and incompetence that we've seen and just like you know cognitive dissonance on the part of of this uh policy process is um really breathtaking but anyways I want to come to Jordan what's your what's your roster take where do you want to put the temperature on just how many chips we should let uh China have access to remotely

Um the the one thing about Raza is that that that makes it clear you know to me is like like you said China is trying to get off GPUs. So like in our tracking in February um GLM was released with support for seven domestic accelerators not just Huawei but they included more threads. Hyggon, Cambercon, Kungan, Meta X, Enflame Deepseek published a mega kernel with Huawei Ascend for their V4 model. Uh Mtoan claims they trained longat entirely on Ascend. Like people don't even train on AMD, you know, it's it's like only Nvidia or TPU that you can train on in the US. So that's happening like with or without the remote access laws being passed. It's just more about motivation. I think it's a lot about inference as well. Like are you going to let bite dance compete globally with inference capacity from these providers on Nvidia GPUs or you just going to like not let them into the US and not let them have a business or you know in Southeast Asia or anything like that or make them serve it from within China? That's that's an open question.

The the main thing I think that people just don't understand is that they think like if you remove one item at all of at the top of the supply chain, they won't do everything. And I've used this analogy a hundred times, but it's the jigsaw puzzle, right? You remove one piece, they'll [__] engineer the one piece. If you blow up the board, it's harder. Um, at the end of the day, this is an economic war. Um, if you think you can get away with something without much blow, you know, without huge ramifications across supply chains, you're not going to be able to solve the issue. You have to just flip the table if you want it to happen. The best way to flip the table is to prevent them from buying relatively low value or low total market cap value goods and make them buy the expensive stuff. Don't buy don't buy ASML equipment and make chips and make models. Make them not allowed to buy the ASML equipment. Don't let them buy chips. Make them rent the chips. Don't let them rent the chips. Make them buy the AI. Right? You know, make them transfer up the value chain where the value chain is much larger and larger and larger. And if you do that, then China has an economic disadvantage, massively so. And this is a scale game. Um, and therefore, the West will win. And so I think like that's that's the key reason why, you know, look, they're going to be pissed about whatever you ban. Ban the thing at the bottom of the totem pole and chop the legs out underneath them. Don't just attack only the top spear, the AI aspect, the chips. Focus on how you make the chips. Focus on the equipment to make the chips. That's what's more important.

Well, I mean, there's a lot here. I think to start, right, it's not just like, yes, is maybe bad for progress, but that that the load star from a, you know, national interest China talk perspective is is relative capability, right? So even if this is something that um you know if if we're if our ecosystem is 20% cut off but like the delta between US China is getting uh tighter then that's you know maybe a worse thing than if the ecosystem is 40 or 50% cut off um at the different kind of levels however you want to slice it um but the delta is is is increasing over time now you know you mentioned the the the Kieran chips versus iPhone and you know where um where Huawei is versus um Nvidia chips but the question the question is not just like capability, it's also scale, right? And that's where it comes back to the equipment. Like yes, it sucks um or it's not ideal that um you know, China's been able to break through the sort of 14 nanometer line which we like kind of sort of tried to draw in the sand um in in in 2022. But you know this again it's a scale game, right? It's not just that you can make a thousand chips, you need to make a hundred thousand. you need to have megawws upon megawws of like advanced hardware in order to be able to um both

I mean, it absolutely, absolutely isn't. Like, there's old blog posts of him talking about how he's not interested, and you can kind of read in his writing like a vision of the world which is much more kind of unified and holistic than what you see out of Xi Jinping in his speeches. And you know, his frame of the world is this, you know, struggle for commanding heights. But like, you know, Liang Fun doesn't run the country, right? She does, and he's been doing it for 15 years, and he's going to be around for a long time to come, and that's the reality that people need to operate under. And Jensen clearly didn't, like, he did not buy in to this framework. He didn't think that sort of China advancing at the same clip as the US was something which was a problem, and there is enough of a consensus among America's elected officials that that's wrong, that we've come to the place that we're at now.

Um, Dylan, any closing policy thoughts? I do want your guys' take on like, just how bad this is that Jensen is now off the table for the future of American innovation. And we should talk about open models. I mean, he was the big champion, right? Like, Dylan, you were talking earlier, like, I still don't understand quite why just someone didn't spend a few billion dollars and make something better than Kimi or DeepSeek and open source it, but...

I guess that was supposed to be money. Not that easy. It's not money, it's talent, man. Like, you know, like all of the most talented people primarily work at Meta, Anthropic, OpenAI, DeepMind, and Meta TBD, right? And then, you know, those who don't are split across a hundred different startups. Um, there's no startup... I mean, like, yes, Thinking Machines has a crack team, but is Thinking Machines as good as Moonshot? Like, no. Like, Moonshot skill-wise is probably just as good as the US labs. They just have a big deficit in compute and deployment.

Um, you know, so I think this is like, if we don't have a scale advantage, then you don't win these. You know, yes, you know, these smaller labs have spent a lot of money on models, but DeepMind, Grok, Meta have regularly fallen behind the frontier of Chinese labs, and they're spending more. So it's not just, oh, spend a few billion dollars and you can make a model that's better. Like, that's, you know, the innovation and the quality of the engineering and researchers at the Chinese labs is just very...

So we just... We just got to get Alex Wong on the case. Is that... that's the only zero to one we've had which has been successful in the past three years.

I mean, they hired a ton of great researchers and have a ton of compute for sure, and spend a bunch of money on high-quality data. It's part of it. I don't know. Look, um, I think there's... so there's two things that I see the discourse shifting towards, which I'm a little like, not completely against, but just trying to be more nuanced on. One is like, ramping up export controls in the form of RASA and MATCH and the diffusion framework is like an obvious good thing, and the US should be immediately doing it. Like, I'm not sure people are thinking through all the implications, which is what we tried to discuss here.

And then the second thing is people are like immediately, oh, Jensen leaves Nvidia, and it's like guaranteed the company's going to zero and can never run without him, and it's such a unique structure. And this is all true, but I think that in that case as well, the more nuanced take is like, in the short term, Nvidia is fine. Like, they need to deliver on commitments to customers and build a bunch of GPUs with the supply chain that they've got. And like, nobody set them up for success as well as Jensen could, but now they just need to execute on the stuff he set them up for, like Rubin and Rubin Ultra. And I think a lot of people know what to go do. I think they know their marching orders. They know their responsibility. They know the mission.

I think the really big test for Nvidia is like, Jensen has been incredibly good at retaining talent, especially amongst his executive team. There are very few high-profile departures from the Nvidia executive team. A lot of people retire, and basically only Renee Haas has gone to ARM and like gone on to be a CEO anywhere else. Like, there's no EVPs, SVPs, VPs that like really leave that often from Nvidia. This is a time where, you know, Jensen and the mission isn't at the helm. Somebody needs to continue to give them compensation and direction and keep them and retain them. And this is probably when the sharks are in the water. The VCs are going to come in and offer people billions of dollars to start their own chip companies, their own NeoLabs. Hey, you should come join us over here.

I mean, we're going to see in the medium to long term whether Nvidia can operate without Jensen. I think that question is up in the air because if a year from now we've seen like a bunch of departures from Nvidia, I'm going to be really concerned about their future. If a year from now everybody's still like rowing in the same direction and Colette is driving things, or they've got somebody else, like especially a technical leader who has emerged like in Buck's... um, maybe they transition Michael Kagan out of CTO or something. I don't know, whatever that looks like, then I would be coming back and saying, hey, I'm, you know, much more confident in Nvidia.

But do I mean, I really think people are underestimating the dominance they have in the supply chain, in the sales and support that they have, all the relationships with the Neo clouds, all the investments they have in the labs and all the startups, the ability they can use their balance sheet to backstop everybody and make them be partners with them, they can invest in everybody, they can acquire all these companies as soon as they're successful. Like, it's still Nvidia's to lose, and I think there's a very, very high chance that they're still doing just fine two, three, four years from now.

And another open question is like, okay, say like some people get poached, they raise big rounds, and they start doing new things. Is that really like the worst thing? Are we sure that's like the worst thing for the Western ecosystem to have, you know, different people trying new bets? Um, the other question is like, we're going from a founder to someone who is not the founder. And I imagine there are just like, aside from executing on the roadmap, there must be some like, you know, big wacky out-of-consensus things which Jensen probably would have done over the next five or ten years, which maybe wouldn't have paid out for another five or ten years, which I guess potentially wouldn't have happened.

Um, and you know, to what extent does that happen, but like half as much, to what extent is that just like, okay, instead of it happening within Nvidia, it happens from, you know, some EVP's new startup? Um, that's kind of the question that I have going from like, you know, the Jobs to the Tim Cook era of Nvidia.

I'm certainly not sitting here saying like this is a really good day for Nvidia. Like...

I don't understand why Nvidia is not down a lot more.

Yeah. I think it's really bad for sure. But people like, I'm talking to people and they're like, Nvidia's at zero, and I'm like, you use the Apple versus Steve Jobs analogy. If you bought stock when Steve Jobs went on leave and then passed away, you'd return a ton of money with Apple stock since that time.

Yeah, but didn't he come back and like...

No, gone. Gone 2.0. I think we're... I think, you know, Apple, this is Jobs leaving in 2015, not leaving like '90.

Sorry. Okay. Got it, got it, got it, got it, got it. I thought you meant the original gone.

Okay. Apple in 2015, looking back is...

Really?

Steve Jobs, pure visionary.

negotiating every supply chain contract, and Steve Jobs is not doing that.

So, Dylan, like, how do you think they... if they start faltering, like, on what dimensions will they start faltering?

Um, I think Rubin, you know, if there's any manufacturing issues with Rubin, people are distracted and Jensen's not there, things get solved slower, things get solved by committees, things don't get escalated up. I mean, I don't know how many times at SemiAnalysis if we [__] email Jensen for a thing to get solved. Like, what the [__]? Like, why are we emailing Jensen to fix something on some open source software, and yet that's the only way we get it solved, and we're irrelevant, right? Like, compare that to like actual companies in the space. How often is it Jensen going out there and solving a problem? From what I've heard from everyone around the ecosystem, it's a lot.

Um, how many times is it like, you know, Elon tells Jensen something, Jensen goes and tells his team, and the team fixes it because the team was trying to work with the team...

Yeah. All...

and it doesn't work out.

Yeah, exactly. But I'm saying, I'm saying Rubin could slip. The yield curve could look worse. Feynman could slip. Feynman could be not as aggressive. You know, one of the popular stories I love saying is like Volta didn't have Tensor until three months before tape out. Jensen is the one who [__] decided this, right? There's bickering little people, and Jensen's like, 'No, AI is the way. Let's do [__] Tensor.' Like, this is... and it's like, 'Fuck everything else. I don't care that, you know, we're three months away from tape out and you say it's going to take six months to incorporate the change. You're going to do it, and you're going to do it now,' and stack on a hundred of these decisions.

Feynman's going to be much worse than it is, and it's going to come later. Rubin's yield ramp could be slower. Rubin Ultra could be slower. Oh, the spec on Rubin Ultra would take longer to do, slash might not even get done, and therefore the volumes that Nvidia can ship are lower. Like, there's a bajillion things that you could decide. XYZ things on software. Oh yeah, like DeepSeek open sourcing all this software and DPP and like all the stuff that Moonshot's doing and making KDA work, that shit's going to take forever to get up and running at high efficiency at Nvidia relative to before.

Maybe it's only a week, but each of these things and decisions and him driving and leadership and the supply chain giving them 70% of their supply instead of, you know, much less because it's Jensen. Um, you know, all of these things are exacerbating, and it could be like in 2029 the best chips aren't Nvidia's and they don't have the majority of the supply chain.

Still the fair watermark. Nvidia is still the leader. They are still driving the industry. If Nvidia slows down, it's not like someone else accelerated. Nvidia slowed down, and so someone else is the best.

No matter what, the industry is progressing slower. Hardware progresses slower without Jensen at the helm of Nvidia, period. Yes, there are competitors, but...

Man, this is pretty morbid for all of us here. Um, I mean, it's sad. This is very Shakespearean.

Um, dude, I was so sad. I've been a Jensen fanboy since I was a teenager. I don't think people like understand how much I [__] like look up to the man and love him. Like, you know, it's crazy.

Dude, the memes are fire. This is probably some of the best memes in like, you know, this... the memes the past few days is up there with like Republican primaries in 2014-15. Like, the memes are that [__] good.

Yeah. If I could snap my fingers, Lisa Su, but I don't think she'll do it. Charlie Kas...