Ep. 044 - Is NVIDIA Over? EMERGENCY Pod with ChinaTalk | Dylan Patel, Jordan Nanos, Jordan Schneider
Jordan NanosDylan PatelJordan Schneider
- The episode frames Jensen Huang’s reported departure as an export-enforcement crisis with consequences beyond NVIDIA. The panel links the crisis to alleged chip diversion, but Nanos distinguishes circumstantial evidence from proof of Huang’s knowledge; Patel qualifies his theory that the board has evidence with “If I had to guess.” Legal overseas rentals and illegal physical smuggling remain separate questions in their account.
- Patel argues that remotely switchable U.S. compute gives Washington valuable leverage over Chinese AI. A blanket cutoff could provoke retaliation against supply chains American servers still depend on; renting through U.S.-controlled clouds preserves the ability to “turn it all off.” Schneider disputes assumed Chinese escalation dominance and thinks Beijing might welcome RASA as pressure for domestic substitution.
- Chinese GPU rentals can increase American profits while keeping Chinese competitors’ compute expensive. Patel estimates $7 million of NVIDIA profit per megawatt plus $7 million–$10 million for the cloud provider at current prices; Meta could potentially operate at one-quarter of ByteDance’s compute cost. He also puts American open-model application ARR “well north of $10 billion,” while Schneider asks whether scarce gigawatts would create more value at U.S. frontier labs.
- MATCH draws the clearest consensus because equipment controls constrain China’s manufacturing scale. The panel favors restricting advanced DUV alongside EUV; Patel’s jigsaw analogy is to “blow up the board” instead of leaving one missing piece China can engineer. Schneider says Ascend is comparable to Hopper and describes Chinese developers supporting domestic accelerators, making the equipment bottleneck increasingly consequential.
- An American open-model champion requires concentrated talent, compute and data beyond a large budget. Patel rates Moonshot “probably just as good as the US labs” skill-wise, with compute and deployment deficits; Schneider adds data. NVIDIA’s Nemotron spending failed to beat Chinese open models, supporting Patel’s answer to the few-billion-dollar fix: “It’s not just money.”
- Schneider gives NVIDIA a high chance of remaining healthy for two to four years, while Patel sees execution risk compounding. Supply-chain dominance and the Rubin roadmap support the near term, but executive retention is Schneider’s one-year test. Patel’s downside case includes slower yields, roadmap delays and a possible loss of chip leadership and supply-chain majority in 2029: “Hardware progresses slower without Jensen at the helm of NVIDIA, period.”
- Succession exposes unresolved questions about technical leadership and state influence. Patel warns of possible “civil war” between finance, technology and sales, with finance in charge. Schneider favors Colette Kress, while Nanos hopes she does well. Schneider’s analogy to NVIDIA’s “nationalization” draws categorical rejection from Patel and Nanos, who characterize the intervention as enforcing export law while leaving lawful chip sales and public advocacy available.
1. NVIDIA’s leadership shock becomes an argument about enforcement
- The speakers take Huang’s departure as their starting point, then separate personal-conduct allegations from alleged chip diversion. For Schneider and Patel, smuggling is the more consequential issue; Nanos says the evidentiary test is Huang actually knowing and being complicit. Patel’s inference stays an inference: “If I had to guess, the board has evidence on this.”
- Schneider describes an alleged Supermicro scheme: a Taiwan employee using a hairdryer to transfer server stickers so fake machines could pass an audit in September of the previous year. Schneider says the employee is cooperating with prosecutors. Nanos asks how to explain NVIDIA’s cessation of Singapore revenue reporting; Schneider maintains that establishing Huang’s knowledge is the necessary next step.
- Patel insists overseas GPU rental is “fully legal,” drawing a boundary between Malaysia-based cloud access and chips physically smuggled into China. Nanos frames NVIDIA’s choices as a conflict between the “letter of the law” and the “spirit of the law.”
- Nanos and Schneider say the Biden-era diffusion framework was left unenforced rather than formally removed; they describe BIS guidance, not a conventional statute. Schneider points to a May 2025 footnote promising replacement rules that never arrived. Patel is surprised: “I thought we would have to repass them.”
2. Patel wants Chinese compute dependence to remain switchable
- Patel’s national-interest framework has two goals: maximize America’s AI lead and control Chinese access. A blanket ban on ownership, deployment and rentals, he argues, could trigger retaliation against components embedded in NVIDIA, Google and Amazon servers. He cites NVIDIA’s largest PCB supplier being Chinese and Washington’s repeated unwillingness to absorb supply-chain pain: “We clearly don’t have the stomach to disentangle the supply chain.”
- His preferred equilibrium is the least Chinese compute compatible with avoiding a supply-chain rupture, under maximum American oversight. U.S.-controlled hyperscalers could rent GPUs abroad, keep data outside China and cut access during escalation: “You can turn it all off.” Closing that route would increase incentives to smuggle and build domestic alternatives, while removing Beijing’s reason to preserve the existing arrangement.
- Schneider’s pushback: Chinese leaders may welcome RASA as leverage over firms reluctant to abandon NVIDIA, and China’s “escalatory dominance” should not be assumed. He cites the absence of comparable retaliation after controls in 2022, 2023 and 2024. Patel replies that those rules contained loopholes and later trade fights ended in compromise because Washington would not stomach the pain; he says the pair are “generally pretty close.”
3. China’s rental bill can subsidize America’s AI ecosystem
- Patel puts the economics per megawatt at roughly $7 million of NVIDIA profit plus another $7 million–$10 million for the cloud provider at current prices. Requiring Chinese companies to rent rather than buy preserves that margin stack for America. In his ByteDance–Meta example, Meta could potentially have one-quarter of ByteDance’s compute cost because the American competitor avoids some of those markups.
- Schneider asks whether the better use of Malaysia-bound gigawatts is simply giving OpenAI, Anthropic and DeepMind more capacity to improve models and earn money. His counterexample to commercial accommodation is TikTok spending billions on Instagram ads: receiving a rival’s money does not settle whether letting that rival compete is strategically wise.
- Patel explicitly casts his response as devil’s advocacy: American application companies using open models generate “well north of $10 billion” in aggregate ARR. He says DeepSeek earns less than U.S. serving companies such as Baseten and Fireworks. “America’s making so much money off of the Chinese models,” allowing American businesses to accrue revenue and invest more.
- Schneider adds the engineering relationships: China-based NVIDIA and AMD teams bring up GPUs and work on low-level hardware and software, while Chinese-born researchers contribute at U.S. frontier labs. Escalation could sever relationships that are difficult to restore: “I think it’s a one-way street.” Schneider also says he learns from people in China. The panel discusses whether the U.S. should accept more restrictions in order to widen its lead.
4. MATCH targets the equipment China needs to scale
- Schneider describes MATCH as extending restrictions to advanced DUV lithography alongside EUV, which he says has not been able to get into China. He wants Japanese, Dutch and Korean suppliers prevented from filling gaps left by American equipment controls. Here the panel converges; Patel’s closing instruction is blunt: “Pass the fucking MATCH Act.” The shared objective is constraining China’s ability to manufacture advanced chips at scale.
- Patel’s analogy is a jigsaw puzzle: “You remove one piece, and they’ll engineer that piece. If you blow up the board, it’s harder.” His prescription is to block lower-value production inputs and force China to purchase increasingly expensive outputs: equipment gives way to chips, chip ownership to rentals, and rentals to purchased AI. That raises China’s cost disadvantage in a scale-driven competition.
- Schneider says substitution is happening regardless of RASA. In February, GLM supported seven domestic accelerators: Huawei, Moore Threads, Hygon, Cambricon, Kunlun, MetaX and Enflame. DeepSeek published a megakernel with Huawei Ascend for V4; Meituan claims LongCat was trained entirely on Ascend. His comparison is pointed: “People don’t even train on AMD,” while Chinese developers support a broad domestic stack.
- Schneider judges the model-capability gap wider but the chip gap “as close as it’s ever been,” comparing Ascend with Hopper and claiming Huawei’s Kirin phone chips are as good as or better than Apple’s. He returns to volume: making 1,000 chips is different from making 100,000. Useful economic deployment requires megawatts of advanced hardware; equipment denial constrains that scaling.
5. A few billion dollars cannot guarantee an open-model champion
- Schneider asks why an American player has not spent a few billion dollars to beat Kimi or DeepSeek and open-source the result. Patel’s answer: “It’s not just money.” Talent is concentrated at Anthropic, OpenAI, DeepMind and potentially Meta, with the remainder scattered across a hundred startups. He rates Moonshot “probably just as good as the U.S. labs” skill-wise, with deficits in compute and deployment; Schneider adds data.
- Schneider points out that NVIDIA already tried: its spending on Nemotron did not produce a model better than the Chinese open-source alternatives. “You need compute, people, and data.” Patel’s conclusion is that engineering and research quality at Chinese labs is extremely high; spending more does not automatically reproduce that capability.
6. NVIDIA’s moat can endure while its execution slows
- Schneider rejects the leap from Huang’s exit to NVIDIA going to zero. Near-term orders require execution against an established roadmap for Rubin and Rubin Ultra; supply-chain control, sales and support, neocloud ties, lab investments and balance-sheet backing remain. “It’s still NVIDIA’s to lose.” He sees a “very high chance” the company is still doing fine two, three or four years out.
- His one-year test is retention. NVIDIA has had few high-profile executive departures; he singles out Rene Haas’s move to Arm. Now venture capitalists can offer billions for new chip companies or labs. A year of departures would make him “really concerned”; a team still moving together under Colette Kress or an emerging technical leader such as Ian Buck would make him more confident.
- Patel’s pushback is immediate execution risk: Rubin manufacturing problems and yield ramps could take longer to resolve as decisions move through committees. SemiAnalysis itself has needed to email Huang to get open-source software issues fixed. “Things get solved more slowly. Things get solved by committees. Things don’t get escalated up.” Slower adaptation to DeepSeek and Moonshot software, delayed Rubin Ultra changes, or weaker supplier allocations could each erode shipment volumes and performance.
- His best example is Volta: Tensor Cores arrived three months before tape-out because Huang demanded a change engineers said required six months. Multiply such interventions across a hundred decisions, Patel argues, and the next architecture, Feynman, could arrive later and less aggressively. In 2029, NVIDIA might lack both the best chips and a majority of the supply chain. His categorical conclusion: “Hardware progresses more slowly without Jensen at the helm of NVIDIA, period.”
7. Succession leaves an unresolved fight over autonomy
- Schneider stresses that 2026 has more capable competitors than 2022 and asks whether departing executives’ new bets could benefit the Western ecosystem. Nanos invokes Apple after Steve Jobs to show that losing a founder need not destroy shareholder returns. Patel favors Charlie Kawwas’s technical-business balance; Schneider says he favors Colette Kress. Patel warns of possible “civil war” among finance, technology and sales, with finance now in charge.
- Schneider says he cannot fathom Kress being unaware of alleged chip diversion and predicts congressional investigations. Patel acknowledges, “We don’t know how deep this web goes.” Nanos counters that immediately appointing a fully complicit insider would be “really strange” after intelligence briefings to the board. The exchange offers conflicting inferences about executive knowledge; it does not establish Kress’s involvement.
- Schneider compares the intervention with Jack Ma disappearing and argues they are watching NVIDIA’s “nationalization.” Patel and Nanos reject that characterization outright, framing it as export-law enforcement. Patel says Huang could sell H20 or H800 within the rules; Schneider says lawful advocacy remained available. The disagreement survives the closing banter: “There’s so much between ‘don’t break the law’ and ‘nationalized.’”
Verification Notes
The transcript uses anonymous “Speaker 0” and “Speaker 1” labels in several stretches. Attributions in those stretches are softened where the speaker is not explicit.
Full transcript
All right. Hey, Dylan. Hey, Jordan. Thanks for joining. Did you guys hear? The CEO of the world’s largest and most important company just resigned.
Nuh-uh. Are you serious? Are you serious?
Wait, what? What happened?
What could possibly have led to that?
I mean, there are allegations of some pretty serious conduct violations here. Apparently, you can’t sleep with your subordinates and Chinese intelligence assets—
Oh, what’s the point of having subordinates?
—and keep your job. Yeah.
Wait, Jordan. We’ve got Elon, we’ve got Sergey. Everyone in Silicon Valley—once you get to a trillion dollars, I think you’re legislated to start sleeping with your subordinates. There must be something else here.
Yeah. I think the trillion-dollar class might be fine, but once you get to $5 trillion, then you have problems with this sort of stuff.
Then woke 2.0 really starts coming after you. Oh, my God.
3.0, maybe.
—
Yeah, we’re rotating back.
So, yeah, this is truly a momentous day, a momentous moment for the future of America, China, and artificial intelligence, where Dylan somehow looks bored after all of this, or just drained. I don’t know. How are you processing it?
Dude, I don’t know. You look up to some guy, this guy, and then he does these things. It’s not fun.
Yeah.
Look, there are a lot of things going on and a lot to contemplate. Given we’re China Talk, we shouldn’t even talk about the future of NVIDIA. We should talk about the future of open source and AI, and China–US relations.
I’ve obviously been a hawk, or a fucking bald eagle, as we said in a prior podcast, for quite some time. At the same time, I’m worried this is going to be used as cover to pass really stupid laws. Obviously, NVIDIA without Jensen is a much weaker company. Whether you think NVIDIA’s interests are the same as America’s national interests is something most people would probably say yes to, though some people in the audience would probably say no.
It’s pretty clear NVIDIA’s chips were ending up in China, knowingly or not. It was probably knowingly. Jensen and NVIDIA were the biggest defenders against not shipping chips to China, and now they’re gone and out of play. So what regulations pass? I hope this means the MATCH Act passes, right? That’s far more important for US national interest than the RASA Act. RASA is a pretty dumb bill. Biden-style diffusion controls should totally be done, but if the US government passes RASA, that’s going to be a big issue. It’s probably going to be really bad for American interests in the short, medium, and long term.
Those are the key aspects of it. The other undertone is that you can’t just ban everything, because China can and will use the fact that its components are in every single NVIDIA server, every single Google server, every single Amazon server, et cetera. You can’t get materials or build these things without them. There’s a level of tit for tat here. I don’t know. I think those are the main things for this audience.
You can add a—well, I think this is a crossover one, right? We should talk about the future of AI a little bit too. But we opened with MATCH and RASA, so let’s start there. The MATCH Act, for those who haven’t been tuning in to my MATCH Act shilling over the past few months, is this idea that advanced DUV lithography, alongside EUV, which already hasn’t been able to get into China, shouldn’t be able to go into China.
Allegedly.
Allegedly, shouldn’t be able to go into China. RASA is slightly different. It’s aiming to close the “cloud” access loophole, by which Alibaba, Tencent, and ByteDance can legally access GPUs that are banned from being imported into China by setting up cutouts in Malaysia, which you guys have done such a great job of untangling.
Jordan, do you share Dylan’s skepticism about RASA and its potential fallout? Or maybe, Dylan, why don’t you expand on why you think it’s such a bad thing?
What is the US national interest? Is it to maintain as large a lead on AI as possible, and to control China’s access to AI as much as possible? If your goal is to maintain the largest lead for as long as possible and maintain control over China, you don’t prevent China from having any compute, because that actually means the US will have no compute either.
If you full-scale ban anything, I believe China will just start enacting economic war. I don’t know if you remember: in 2022, I stayed up all night and wrote the October 7 chip restrictions regulation. I said China and the USA were officially at economic war. Or I said it was a new Cold War, an economic war. In there, I found about 20 loopholes that I posted about that very night. I’m just one dude, not an army of lawyers—
—and was hoping they’d get patched. Many did not get patched, and they still aren’t patched.
One of the key aspects of the original restriction, which I think is mostly true, is that there’s civil-military fusion within China. The Chinese military and state will and can apprehend any resource for military and intelligence purposes, even if it’s a completely commercial thing. We’ve seen it with things like Douyin and TikTok already, and that’s as consumer-facing as you can get. So clearly they do utilize it.
At the same time, they’ll also enact and respond in certain ways. We’ve seen this with the trade wars, with some of the auto tariffs, with our auto plants almost shutting down, and with threats around rare earth minerals. China has threatened or taken some actions and then pulled back to show its fangs.
If you ban chips completely from China, including rental and deployment, there’s no reason for China not to blow the whole thing up. You have to have some tit for tat. The optimum thing for American policy is to figure out the minimum amount of compute China can have with the most American oversight, so they won’t blow up the supply chain. We clearly don’t have the stomach to disentangle the supply chain. We just don’t, over and over again.
Given that, you can’t have China have no chips, period, either internally or from abroad. That will blow the whole thing up. But to minimize the amount of compute and maximize control, you can have Biden-style diffusion laws for certain companies. Yes, it’s basically a bill written for hyperscalers, but hyperscalers can rent to Chinese companies, with America having the power to say, “Hey, turn it off at any moment.” You want all of their compute to be like that.
If you allow Alibaba, Oracle, ByteDance, Moonshot, or whoever to rent compute from American companies outside of China, that’s supremely awesome if it’s their main access to compute. China won’t want to send its data externally in many cases, and if it does, it’ll probably be stuff the US intelligence apparatus can handle. And if shit goes tits up, you can turn it all off.
Whereas if you turn off remote access, there’s even more incentive to smuggle, ramp up your own domestic industry—which they’re going to do anyway, by the way, which is why you need the MATCH Act—and really prevent any reason for them to continue the status quo. A lot of Chinese industry is dedicated to making American compute, at least in the electronic space, for American companies. The largest PCB supplier to NVIDIA is a Chinese company, for example. It’s not just NVIDIA, of course.
I think you have to approach it from a game theory perspective, and the best thing for America to do is allow remote access to GPUs, because the alternatives are much worse.
Yeah. A few pushbacks on that. I think there’s a piece of the Chinese ecosystem, Xi Jinping likely included, that’s deeply embarrassed by the fact that these companies really want to import the GPUs. Part of what we’ve learned over the course of the Trump administration so far is that even when Trump is offering them, the Chinese government is like, “I’m not entirely sure we even want these companies to buy them.”
So, yes, we have some equilibrium right now, but I definitely see a world in which chunks of the Chinese government see something like RASA and say, “All right, great. These companies thought they could get cute and not really invest in the domestic ecosystem, and now they’re going to really have to play ball.”
I think there are also other things going on with Xi that would make him less likely to want to declare economic warfare on the US by doing something as dramatic as shutting off all global manufacturing of chip production, even if we end up figuring out how to do PCB—
Yeah, but we already shut them off.
Why wouldn’t they retaliate?
Because then you get into a tit for tat. And, by the way, this economy isn’t growing, and unemployment’s at 20%. Look, we’ve—
Every single time we’ve gotten into that tit for tat, it’s led to a compromise, and there was no tit for tat, right? Tariffs got reduced. There were some restrictions that went up, but mostly China didn’t just take it. They did retaliate, and the US economy and administration wouldn’t be willing to stomach the pain.
Yeah. They didn’t in 2022. They didn’t in—
Well, in 2022, the loophole—
They didn’t—
The rules had so many loopholes.
Yeah. They didn’t in 2023, and they didn’t in 2024. I think Trump kept doing the war in Iran even though he lost 10 Senate seats. I think there’s an aspect of pain, especially in a second-term lame duck, when all the cabinet is leaking against Lutnick because they’re pissed. There was an Axios report that said China is already breaking the trade deal and doing economic warfare against the US. I don’t think escalatory dominance is something we should completely assume on the part of the Chinese government.
The other argument I’d make is that in RASA, if China has access to compute outside of China, they’re now paying humongous margin dollars to the US ecosystem, right? If you take a megawatt of compute outside of China, there are multiple layers of margin stacking. First of all, Nvidia’s making $7 million of profit per megawatt that gets deployed. In addition, the cloud provider is making another $7 million to $10 million of profit if it’s selling at today’s pricing.
There’s a tremendous amount of profit we get to capture. Companies like Anthropic, OpenAI, Google, Amazon, and Meta either do less of that because they don’t have the cloud, or they’re just flowing their own chips. 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? Our cost structure is better, and our economies of scale are higher.
Allowing them to ship all this money out—if ByteDance wants to compete with Meta in the global market on video recommendation algorithms, Meta potentially has a fourth of the cost of compute. If you don’t let ByteDance directly buy the chips, but you do let them rent them—
Yeah. But here’s the thing. We want to compare ByteDance and Meta, but Zuck was totally fine letting TikTok spend billions of dollars on Instagram ads, right? If you just don’t let them play the game, then you kind of win by default.
And also, Dylan, isn’t there enormous demand, even in our shaky AI times over the past week? Do you really think it wouldn’t be better for the US to get ahead by having however many gigawatts are currently flowing through Malaysia into the Chinese ecosystem help OpenAI, Anthropic, and DeepMind make better models?
I think there’s also a reflexive nature to this. I’m just going to devil’s advocate this for a while—
Okay.
—I think we’re generally pretty close, but it’s fun to devil’s advocate so people hear. This is sort of what Jensen’s argument was, to some extent: “AI diffusion across the US is largely aided by Chinese models.” If we look at total AI ARR in the AI ecosystem, and add up all the application-layer companies running on open models, it’s well north of $10 billion. It’s not some tiny amount of open-model revenue going to American companies.
DeepSeek makes less revenue than Baseten, Fireworks, and all these guys serving the Chinese models. America is making so much money off the Chinese models, and it’s amazing. It lets us accrue and invest more and more.
Yes, ByteDance bought a bunch of ads on TikTok, and Zuck didn’t really see the threat immediately. Of course, everyone’s short-sighted; everyone makes mistakes. Jensen got a little horny. It was his short-sighted mistake at the end of the day. But, come on.
He did more than that, right? It’s clear what he did—
He had many relationships over a long period of time, yes.
It’s not just the women. It’s doing a little more than turning a blind eye to the fact that his suppliers were not only doing the Malaysia thing, which is kind of legal even today, but bringing the chips into China—
No, it’s fully legal. I don’t think it’s just kind of legal.
Yeah, no. I mean, these are ethical hoops. This is unethical.
I think it’s about the letter of the law versus the spirit of the law. Jensen is in these closed-door meetings with a lot of US leaders, who clearly have the spirit of wanting to slow down China’s advance. He has decisions he can make, and they didn’t necessarily point to the exact—
But I don’t think that’s why he’s losing his job. I think he lost his job because there’s some evidence that they knew the chips were going into China—
One thousand percent. That has nothing to do with the affair.
Yeah. And look, circumstantially around Jensen, his Taiwan sales manager is indicted for smuggling chips out of Taiwan. His closest OEM partner, Supermicro, had its COO and co-founder—or head of sales, whatever, Wally—arrested. One of their sales managers is currently a fugitive in Taiwan. So it’s not out of the blue that they’re cracking down on this, and that people around him are being caught with clear evidence. The picture of the Supermicro Taiwan guy with a hairdryer in his hand, moving the stickers from the fake servers onto the real ones—or from the real ones—
Unbelievable.
—onto the fake servers that passed an audit in September of last year is pretty damning evidence that they’re aware of this and actively working—
And there was no honeypot required to get Jensen to do that with Supermicro.
Well, I think the evidence that needs to pass the test would be Jensen actually knowing about this, being complicit in it, and doing something. But either way—
If I had to guess—
—the public opinion that terminated him—
—I bet the board has some evidence here.
—right now, right?
If I had to guess, the board has evidence on this.
I mean—
Yeah.
How do you explain no longer reporting Singapore revenue?
Yeah. I mean—
True. Yeah.
The circumstantial evidence is there. I think this is the big thing: even in the case of his affairs, people 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 the US should have with China and the technology they’re working on there. And it’s a convenient time, right around the midterms, when he gets whacked.
The cover is gone, right? The cabinet members basically took the blue wave as, “Okay, it’s fine. We can now blow the cover, and it’s over for Lutnick and Jensen.”
Yeah, basically, because—
It’s pretty clear Lutnick always—and Trump—wanted to make deals, and they used export restrictions as a deal-making tool. But in another sense, it’s the law. You can’t negotiate with the law. You make a law, and you enforce it.
No, I think enforcement has been a big question so far. If you look at the Biden-era diffusion framework, it was passed and then just not enforced. It’s still on the books. Trump just directed people not to enforce that framework.
Really? The diffusion rules weren’t removed? I didn’t realize. 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, it’s guidance to the BIS, basically.
Yes. Kessler never published real guidance saying, “We are no longer doing this.” He just made a footnote somewhere that said, “By the way, guys, don’t worry about this. We’re going to bring a new thing for you soon.” That was in May 2025. The level of indecision and incompetence we’ve seen, and the cognitive dissonance in this policy process, is really breathtaking.
But anyway, I want to come to Jordan. Jordan, what’s your RASA take? Where do you want to put the temperature on just how many chips we should let China access remotely?
I think what Dylan said about having the US be able to monitor what China is doing with the chips, as well as having the option to cut off its access and then start to use the chips for what it wants in a future where the intelligence wars are playing out and there’s an escalation, is important optionality for them.
The other thing is that it’s so clear how much American companies benefit from a mutually beneficial relationship with China, Chinese companies, Chinese researchers, Chinese software, and Chinese people. Lots of Chinese citizens work for the frontier labs today, or people who were born in mainland China. It’s really complicated for me to understand. We work with engineers based in China sometimes who work for Nvidia and AMD, have access to early products, bring up the new GPUs, and work on the software and low-level hardware. As tensions escalate, all of this relationship breaks down, and everything gets harder for everybody.
Insofar as anybody is pro-AI, this is a bad day for AI progress. As tensions escalate, it seems like a bad day for AI progress. That would be my high-level take.
1. China’s Chipmaking Catch Up
The one thing about RASA that makes it clear to me is, like you said, China is trying to get off GPUs. In our tracking in February, GLM was released with support for 7 domestic accelerators, not just Huawei, but also Moore Threads, Hygon, Cambricon, Kunlun, MetaX, and Enflame. DeepSeek published a megakernel with Huawei Ascend for its V4 model. Meituan claims it trained LongCat entirely on Ascend. People don’t even train on AMD. In the US, it’s only Nvidia or TPU that you can train on.
That’s happening with or without the remote access laws being passed. It’s more about motivation. I think it’s a lot about inference as well. Are you going to let ByteDance compete globally with inference capacity from these providers on Nvidia GPUs? Or are you just not going to let them into the US, let them have a business in Southeast Asia or anywhere else, or make them serve it from within China? That’s an open question.
The final thing is that there are so many ways in which I personally, and many people I know who work at the American labs, learn from people in China. If you cut off those direct working relationships, I think it’s a one-way street. It’s really hard to understand how you roll this stuff back.
But then, to Dylan’s point, I’d love to get your take on this: I think you can do all of the meaningful restrictions just by focusing on ASML and the MATCH Act. Expand it beyond EUV and stop them getting access to EUV tools, too. They need a lot of those tools if they’re going to produce any of these chips that go into data centers. Make them build lithography from scratch. That’s kind of the easiest way to stop them making progress here.
The main thing people don’t understand is that they think if you remove one item at the top of the supply chain, they won’t do everything. I’ve used this analogy a hundred times, but it’s the jigsaw puzzle: you remove one piece, and they’ll engineer that piece. If you blow up the board, it’s harder.
At the end of the day, this is an economic war. If you think you can get away with something without much blowback, without huge ramifications across supply chains, you’re not going to be able to solve the issue. You have to 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 goods, or goods with low total market cap value, and make them buy the expensive stuff. Don’t let them buy ASML equipment and make chips and models. Don’t let them buy 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. Make them transfer up the value chain, where the value chain gets larger and larger. If you do that, China has a massive economic disadvantage. This is a scale game, and therefore the West will win.
I think that’s the key reason. 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 from underneath them. Don’t just attack 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.
I think it’s also a timing thing. We’re watching China on the rise, and I think that in response to a lot of these actions, they don’t want to escalate because the longer they wait in these intelligence wars, the more time they have to catch up. There’s a lot of discourse about whether the gap is wider or smaller. I would say the gap in model capabilities is wider at this point, but the gap in chip capabilities is as close as it’s ever been. Ascend chips are pretty comparable to Hopper chips. Just look at Huawei’s phone chips right now. If you look at Kirin and compare it to what Apple is shipping for phone SoCs, it’s as good or better.
That’s the thing to focus on, not the models, which I think are much more clearly behind the American frontier at this point.
There’s a lot here. To start, the lodestar from a national interest China perspective is relative capability. Even if this is bad for progress, if our ecosystem is 20% cut off but the gap between the US and China is getting tighter, then that may be worse than if the ecosystem is 40% or 50% cut off at different levels, however you want to slice it. The delta is increasing over time.
You mentioned the Kirin chips versus iPhone chips, and where Huawei is versus Nvidia chips. But the question isn’t just capability; it’s also scale. That’s where it comes back to equipment. Yes, it sucks, or it’s not ideal, that China has been able to break through the 14-nanometer line that we tried to draw in the sand in 2022. But again, it’s a scale game. It’s not just that you can make 1,000 chips. You need to make 100,000. You need to have megawatts upon megawatts of advanced hardware to train and deliver this AI in a way that makes a difference to your economy.
As long as you get the equipment side right, it makes it so much harder for the Chinese ecosystem to compete with the American one if they can’t rely on the Japanese and the Dutch. They already can’t buy the stuff from the Americans, so it’s not like we’re screwing over US companies on the semiconductor manufacturing equipment side. It’s making sure that the Japanese, Dutch, and Koreans can’t make up for the fact that we’re not able to sell to them.
I hear you, Jordan, on the interpersonal stuff. There is an aspect of this that is deeply tragic. AI has the potential to deliver incredible things for humanity. There is a side of this that’s a beautiful, humanity-scale experiment to make the future now.
But the problem is that the US and China have a lot of issues with each other, and China happens to be run by an authoritarian leader who runs a Leninist system. It’s not necessarily Liang Wenfeng’s fault that that’s the case. It absolutely isn’t. There are old blog posts of his talking about how he’s not interested. You can read in his writing a vision of the world that’s much more unified and holistic than what you see from Xi Jinping, his speeches, and his framing of the world as a struggle for commanding heights.
But Liang Wenfeng doesn’t run the country. Xi does. He’s been doing it for 15 years, and he’s going to be around for a long time to come. That’s the reality people need to operate under.
Jensen clearly didn’t. He didn’t buy into this framework. He didn’t think China advancing at the same clip as the U.S. was a problem. There’s enough of a consensus among America’s elected officials that that’s wrong that we’ve come to where we are now.
2. Why Open Models Lag
Dylan, any closing policy thoughts? I want your take on how bad it is that Jensen’s now off the table for the future of American innovation. We should talk about open models. He was the big champion, right? Dylan, you were saying earlier that you still don’t understand why someone didn’t spend a few billion dollars, make something better than Kimi or DeepSeek, and open-source it.
Because it’s not just money.
I guess that was supposed to be—
It’s talent, man. All of the most talented people primarily work at the major labs: Anthropic, OpenAI, DeepMind, and Meta TBD. Those who don’t are split across a hundred different startups. There’s no startup—I mean, yes, Thinking Machines has a crack team, but is Thinking Machines as good as Moonshot? No. Moonshot is probably just as good as the U.S. labs skill-wise. They just have a big deficit in compute and deployment.
If we don’t have a compute scale advantage, then you don’t win. These smaller labs have spent a lot of money on models, but DeepMind, Groq, and Meta have regularly fallen behind the frontier of Chinese labs, even though they’re spending more. It’s not just, “Spend a few billion dollars and you can make a better model.” The innovation and the quality of the engineering and researchers at the Chinese labs are very high.
Data.
If we don’t have a compute scale advantage, then you don’t win. These smaller labs have spent a lot of money on models, but DeepMind, Groq, and Meta have regularly fallen behind the frontier of Chinese labs, even though they’re spending more. It’s not just, “Spend a few billion dollars and you can make a better model.” The innovation and the quality of the engineering and researchers at the Chinese labs are very high.
Wait a second. NVIDIA did spend this money and try to do it, and still didn’t produce anything with Nemotron that was better than the open-source models from the Chinese labs. So even though Jensen is a big supporter of open source, you’re not guaranteed to be effective just because you pour money and time into it. It’s hard. You need compute, people, and data.
So we just have to get Alex Wang on the case? Is that the only zero-to-one we’ve had that’s been successful in the past 3 years?
They hired a ton of great researchers, have a ton of compute, and spent a bunch of money on high-quality data. That’s part of it. I don’t know. I think there are 2 things the discourse is shifting toward that I’m not completely against, but I want to approach with more nuance.
One is ramping up export controls in the form of RASA and MATCH and the diffusion framework, as if that’s obviously good and the U.S. should do it immediately. I’m not sure people are thinking through all the implications, which is what we tried to discuss here.
3. NVIDIA After Jensen
The second is people saying, “Jensen leaves NVIDIA, and the company is guaranteed to go to zero. It could never run without him, and it has such a unique organizational structure.” All of that is true, but the more nuanced take is that NVIDIA is fine in the short term. They need to deliver on existing commitments to customers and build a bunch of GPUs with the supply chain they have. Nobody set them up for success as well as Jensen could.
Now they need to execute on what he set up for Rubin and Rubin Ultra. A lot of people know what to do. They know their marching orders, their responsibilities, and the mission. The really big test for NVIDIA is that Jensen has been incredibly good at retaining talent, especially on his executive team. There have been very few high-profile departures. A lot of people retire, and basically only Rene Haas has gone to Arm and become a CEO elsewhere. There aren’t many EVPs, SVPs, or VPs who leave NVIDIA.
This is when Jensen and the mission aren’t at the helm. Somebody needs to keep giving people compensation and direction, and retain them. This is probably when the sharks are in the water. VCs are going to come in and offer people billions of dollars to start their own chip companies, their own neocloud labs, and say, “Hey, you should come join us over here.”
We’ll see in the medium to long term whether NVIDIA can operate without Jensen. If a year from now we’ve seen a bunch of departures, I’m going to be really concerned about their future. If a year from now everybody is still rowing in the same direction, Colette is driving things, or somebody else—especially a technical leader—has emerged, like Ian Buck stepping up, or maybe they transition Michael Kagan out of the CTO role, then I’d come back and say I’m much more confident in NVIDIA.
I really think people are underestimating NVIDIA’s dominance in the supply chain, its sales and support, its relationships with the neoclouds, and its investments in the labs and startups. It can use its balance sheet to backstop everybody and make them partners. It can invest in everybody and acquire these companies as soon as they’re successful. It’s still NVIDIA’s to lose, and I think there’s a very high chance they’ll still be doing just fine 2, 3, or 4 years from now.
Another open question is: if some people get poached, raise big rounds, and start doing new things, is that really the worst thing? Are we sure that’s the worst thing for the Western ecosystem—to have different people trying new bets?
The other question is that we’re going from a founder to someone who isn’t the founder. Aside from executing on the roadmap, there must be some big, out-of-consensus things Jensen would have done over the next 5 or 10 years that might not have paid off for another 5 or 10 years. To what extent does that happen half as much? To what extent does it happen outside NVIDIA, in some EVP’s new startup? That’s the question I have, going from the Jobs era to the Tim Cook era of NVIDIA.
I’m certainly not sitting here saying this is a really good day for NVIDIA.
I don’t understand why NVIDIA isn’t down a lot more.
I think it’s really bad, for sure. But I’m talking to people, and they’re saying, “NVIDIA’s at zero.” You used the Apple versus Steve Jobs analogy. If you bought stock when Steve Jobs went on leave and then passed away, you’d have made a ton of money on Apple stock since then.
Yeah, but didn’t he come back?
No, he was gone. Gone 2.0. I think this is Jobs leaving in 2015, not leaving in ’99.
Sorry, okay. I thought you meant the original gone.
Apple in 2015, looking back, is a buy.
I think Steve Jobs wasn’t a maniacal guy who ran every single part of his company. He cared about every aspect of his firm, but at the end of the day, I don’t think he exerted as much control over Apple as Jensen does over NVIDIA or Elon does over his companies.
Really?
I don’t know Steve Jobs or Apple as well. I’ve only read some books, but it doesn’t seem like he was that important to Apple. I just don’t think it’s the same level of work.
So how do you think they’ll falter? If they start faltering, on what dimensions will that happen?
I think if there are any manufacturing issues with Rubin, people are distracted and Jensen isn’t there. Things get solved more slowly. Things get solved by committees. Things don’t get escalated up. How many times at SemiAnalysis have we emailed Jensen to get something solved? Why are we emailing Jensen to fix something in open-source software? And yet that’s the only way we get it solved, and we’re irrelevant. Compare that to actual companies in the space. How often is Jensen going out there and solving a problem? From what I’ve heard from everyone around the ecosystem, it’s a lot.
How many times has Elon told Jensen something, Jensen gone and told his team, and the team fixed it? The team was trying to work with the team.
Yeah. All of the—
And it doesn't work out.
All of these examples of the reasons why NVIDIA would technically have problems going forward were actually already present with Jensen at the helm. They have a big problem with all the custom chips coming up that are—
Yeah, exactly.
Yeah.
But I'm saying Rubin could slip. The yield curve could look worse. Feynman could slip. Feynman could be less aggressive. One of the popular stories I love telling is that Volta didn't have Tensor Cores until 3 months before tape-out. Jensen was the one who decided this, right? There are little people bickering, and then Jensen's like, “No, AI is the way. Let's do fucking Tensor Cores.” It's like, fuck everything else. I don't care that we're 3 months away from tape-out and you say it'll take 6 months to incorporate the change. You're going to do it, and you're going to do it now.
Stack on 100 of these decisions: Feynman could be much worse than it is, and it could come later. Rubin's yield ramp could be slower. Rubin Ultra could be slower. The de-spec on Rubin Ultra could take longer to do, or might not even get done, and therefore the volumes NVIDIA can ship are lower. There are a bajillion things you could decide. XYZ things on software. DeepSeek open-sourcing all this software, and DPP, and all this 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 Jensen driving things, and the supply chain giving him 70% of their supply instead of much less because it's Jensen—all of these things are exacerbating. In 2029, the best chips might not be NVIDIA's, and they might not have the majority of the supply chain.
Yeah. Well, I mean—
Yeah.
That's relevant to the stock, right? But I think this is 2026, not 2022. There are lots of other competent teams running at this in a way that there kind of was only one before.
NVIDIA is still the fair watermark. NVIDIA's 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.
Yeah, it's a question mark maybe, depending on your situation for inference, and then the phone, all of computing—
No matter what, the industry is progressing more slowly. Hardware progresses more slowly without Jensen at the helm of NVIDIA, period. Yes, there are competitors, but without NVIDIA in the lead, people are running less fast. Even if they're running as fast as they were before, because NVIDIA was in the lead and they're no longer as relevant, progress has slowed. This is a real slowdown in progress.
Man, this is pretty morbid for all of us here. It's sad. This is very Shakespearean.
Dude, I was so sad. I've been a Jensen fanboy since I was a teenager. I don't think people understand how much I look up to the man and love him. It's crazy.
And it's sad, the circumstance. A lot of us were always impressed by NVIDIA, and to see— I think a lot of people are trying to excuse conduct that's just so bad to read about this. I see all the memes about it. I don't know.
Dude, the memes are fire. These past few days, the memes are up there with Republican primaries in 2014 and 2015. The memes are that fucking good.
Yeah. I laugh and then I'm sad. That's what the memes are doing to me right now.
Eliot Spitzer, as a little high schooler into politics, was my hero. He was crusading against corruption, governor of New York State, and then it comes out that he'd passed an anti-prostitution law, and then it comes out that he had $2,000-an-hour escorts and had to leave the governorship. That was my wake-up moment, realizing that these men with egos aren't necessarily all they're cut out to be. It's not surprising, but it is depressing, and it reminds you that humans are frail. Don't ever meet your heroes. But we could still look up to Lisa Su, right? She's still there for us. Dylan, who's your dream new NVIDIA CEO? Who are we putting in if we can snap our fingers?
If I could snap my fingers, Lisa Su, but I don't think she'll do it. Charlie Kawwas. Johny Srouji. Charlie Kawwas. Charlie Kawwas would be the best.
How about you, Jordan?
Well, if Charlie's on the table, I'll just say Hock Tan, man. Bring the guy in and tear it to pieces—
I think Hock is too good of a businessman. I think Charlie is a bit more technical and less of a businessman than Hock, even though Hock is a really good businessman. I think you actually want Charlie for that dynamic.
Yeah, my personal opinion is a lady named Colette Kress. I really look forward to working with her, and I am—
Well, we do love Colette. She's great.
—very impressed, yeah.
There is no—I know you guys are buds. There is no way the CFO of NVIDIA isn't implicated in whatever has been happening with these chips. It's just not possible. I could not fathom a world—
We don't know how deep this web goes, but I don't see how it's possible that Jensen and Colette aren't both involved. I don't know. It's hard. Colette's not the one who's in Asia making the agreement, so maybe she could pretend, “Oh, I didn't know. I signed a lot of pieces of paper.” I don't know.
But she—
The web goes deep.
Look, she's going to have to testify in front of Congress about just how much she didn't know, and there are going to be congressional investigations into this. It may be true, but that's why you want a technical CEO or someone from outside, because I just don't understand how someone deeply involved with the operations and interacting with the clients—
Yeah, the new CEO isn't going to be called to testify for days and days and days. They get to focus on the company.
Yeah. Look, I think the fact that this originates from the intelligence community kind of rules out the person who is immediately installed as CEO having in-depth knowledge of it. The board has clearly been briefed on something initially that Rubio knew back in May, before they did the visit to China and stuff. I think that would be really strange. But—
It's not the Chinese spy. It's all the facts with Supermicro. It's not the affair anymore. It's Supermicro, it's Megaspeed, it's chip smuggling into China, and the sort of willful covering of your eyes—
Yeah. Willful ignorance, yeah.
I think anyone who's complicit in that is going to have a really hard time staying out of the line of fire.
Yeah. I'm watching the Supermicro indictment really closely to see how they actually play out prosecuting this stuff, because they flipped the guy who was doing the hairdryer stuff, taping and moving the stickers, who deleted his encrypted messaging app the day he got arrested. Apparently he's working with the prosecutors now. We'll see what happens there. There's not really an obvious way to hide this stuff and be willfully ignorant in the technology-connected internet world we live in right now, when you're trying to do things in Taiwan from the U.S. We'll see. I would find it really strange if they immediately installed somebody who was completely complicit in all the same stuff as Jensen, which is motivating the change.
Maybe. I don't know. I hope not. I hope she does a great job. I hope she delivers all our new chips online and brings me AGI, but we'll see. Any closing thoughts? Sorry.
I think the closing thought is just that there's potential for NVIDIA to go into civil war here over this, because she's interim. She's not full-time. I think there's a constant clash in any company between finance, technology, and sales—the three big areas. Sales wants to do deals. Technology wants to make big bets. Finance needs to figure out how to pay for it. Finance is in charge now. Well, we know what's going to happen to the NVIDIA business going forward.
It’s probably going to be very profitable for a little while, and then medium to long term it’s probably going to look a lot different. My closing thought on the political side is that I really hope the people who are—
Pass the MATCH Act.
Yeah, I really hope people are—
Pass the fucking MATCH Act.
—thinking this through. Think this stuff through, guys. Please.
I think you can be a hawk and have a debate on RASA, but it’s hard to have a debate on—
Yeah. It seems so obvious. Anyway, we’ll be watching closely, guys.
All right. We’ll—yeah. Tune in for more. I tweeted on Monday, “I did my emergency pod.” Or we should just end it. Let’s end it.
Say it. Say it. What’s up?
Oh, no. I think I did my emergency pod on—
Saturday.
—Friday. I was like, “Oh, man, how many days is it going to take before I have to do another one?” Six days. Okay.
Yeah, yeah.
So, tight turnaround.
A daily emergency pod. That’s what we were saying internally. Every single one of our pods has to be named “Emergency” right now. What do you think’s going to happen next week, Dylan? How can we ramp up from here?
All right.
We thought the Anthropic investment was a greater thing through the capital markets, but now the CEO of the biggest, most valuable, most important company in the world goes away. What are we doing here? What can we—
Look, AI 2027 said China wakes up to AI in November. So we’ve got a big bear awakening.
Next November, or right now?
No, no. We’ve got to—
This November.
—we’ve got to nationalize Anthropic and SpaceX. And then Dario gets thrown in jail.
All secret subscriptions gone.
Oh, no, no, no. I’m just saying, for season two, we’ve got to up the stakes, right? For our next emergency pod, if they want it to go crazier—
Hold on. Okay, to—
—if they want it to go crazier.
Just to be clear, how do you think it functionally plays out to nationalize a company of that size?
I have no idea. No, it’s not actually going to happen. Sorry. We’ve got to end this podcast in like 3 minutes.
We have to kind of force the CEO out and install somebody new who’ll be happy to work this over.
We lost the thread.
Okay, we can cut all this, but—
I mean, I know—
I mean—
I know Jordan’s a socialist, and he loves and admires China so much, so—
Don’t you call this Canadian. Wait, what?
No, no, Schneider, Schneider, Schneider.
Oh, Schneider.
You need to pass some more powerful AI, AGI—you know, there needs to be some legislation around it. I don’t think the executive branch can just nationalize something like that.
Okay, cut this part out, but what I’m getting at is that I think we are watching the nationalization of NVIDIA play out right now. This is the equivalent of Jack Ma disappearing.
Leave it in.
I don’t think so at all.
I disagree completely.
I don’t think so at all.
I disagree completely, Jordan.
All he had to do was not sell chips to people smuggling them to China, and he would’ve been fine. That’s all he had to do.
Yeah, so he does something against the national interest, and the national intelligence community comes in, takes him out, and installs somebody they have more influence over.
I mean, yeah, when you break the law, there are consequences.
It’s not influence. It’s just don’t break the fucking law.
Yeah, just don’t break the law.
Okay, sorry, guys. Sorry for the conspiracy theory.
He can advocate all he wants. He can scream to the rafters and tweet.
He could’ve done H20 all he wanted to, and then we would—he didn’t break the law. He did H800. He didn’t break the law.
Wait, all you’re saying right now is he deserved to be nationalized, which, fair enough.
No.
He did.
No.
I’m saying that—
There’s so much between “don’t break the law” and “nationalized.” This is a fucking massive chasm.
Are you saying that this is not them nationalizing?
They are not nationalizing NVIDIA right now. That’s crazy.
No chance. Absolutely no chance. Absolutely no chance.
What is the difference? What would they have to do for it to look like nationalization to you?
I don’t know. AOC becomes president.
What does nationalization even mean to you?
I’m stunned into silence, guys.
Sorry, what does nationalization mean to you?
Where is my point?
Does it mean a CEO who—
Yeah, they have to remove the CEO—
—who doesn’t break the law?
No, you—
I don’t know. The people we’ve said for CEO are not nationalizing the company. They’re just logical CEOs.
Right, so let’s see who it is. Jared Kushner for CEO of NVIDIA.
When Donald Jr. becomes CEO, then we’ll know—
No, no—
—who was really behind this from the beginning. Oh my God. All right.
I don’t know. Maybe this is too far gone for you guys, but I think that’s exactly what’s happening right now. Anyway, cut this part out.
Well, it’s aligning a country—
No, leave it in. Leave it in.
—it’s aligning a country with the laws on the books.
I don’t think so.
Like—
Leave it in, Jordan.
All right. We’ll leave it in. Okay.
All right, you guys got on camera pushing back at my wild conspiracy theory that I came up with—
Press the button. Press the button. Vance, end this.
—and I didn’t rip everything we’ve said today on this show. It’s not financial advice. Don’t come after me, please.
Wait, who’s going to end the—
Vance isn’t even here. Where is he at?
Somebody press the button. Dylan, can you press the button? Is it on your computer?
I don’t know. We’ve got people recording this.
Dude, I can’t press the button.