> My guess is we will want more engineers for a while, not less.
Tech hiring in the US seems frozen or companies are laying off. Can argue about whether this is AI or not, or they are just stealth hiring in lower cost countries, but they are not hiring in the US.
YMMV but just this week a friend of a friend was hired for a junior SDE position in Google in the US. You can have layoffs and hirings at the same time in the same company.
Seems to me that Zvi used a lot of words and missed the point, because the point was unstated. Nvidia must have a silent chain of thought that if China has access to great Nvidia chips, made in Taiwan, then China has less reason to precipitate a military intervention in Taiwan. A military intervention would result in TSMC's production machinery being at least disabled remotely by ASML if not destroyed, unless Trump has already given up Taiwan ahead of time, which maybe he has. Of course Jensen couldn't possibly say that. But how can you miss the backdrop here?
@Andy B, I take your point, but. What Xi can do and what Jensen Huang can do are different matters. Xi has many factors to balance. His economy is stressed with inadequate consumer demand and a debt crisis. He continues purging senior military. And he wants to look reasonable while America alienates almost every country and disarms itself by over-spending its weapon stores. I've seen estimates that an attack on Taiwan would cost China alone more than $1 trillion; never mind Rest Of World. But what Jensen Huang can do is much less. He can sell everything his company produces, and try to postpone what he can't really control. His ability to influence China is very limited, but one thing he can do is work hard to figure out a way to sell his chips to them. He can't say out loud, "China's Taiwan decision is a huge risk to Nvidia." But it is. It is an existential risk, perhaps the only existential risk he faces at the moment and for the foreseeable future.
That's fair. (China wouldn't blink at paying trillions for Taiwan, either, but that's beside the point.) I certainly do agree that Huang is trying to make nice with China because they might foreseeably have the power to destroy his company; I just think his goal is simply to try to be on their good side if they take Taiwan (or otherwise Win), rather than ascribing to him the belief that he might have even the slightly influence over China's Taiwan policy, which would be an Elon level of insanity :).
Many Thanks to both of you for a great discussion!
One thing that has been irking me about the whole analysis of Huang's/Nvidia's comments and actions has been, as Zvi said:
>If Nvidia truly didn’t want to pick winners it would allocate fully via price.
Since, at least in the short term, the default profit-maximizing choice is to charge the market-clearing price, it seems really strange to see Nvidia choosing to charge less than that, and allocate between potential customers.
It takes a really contorted view of future demand (architecture lock-in??? - in a world with Codex and Claude Code??? Really???) for it to make sense as a profit-maximizing decision.
As a please-don't-blow-my-foundry-off-the-face-of-the-Earth, trying to sell to the PRC makes much more sense.
I'm confused why you and Zvi don't believe Jensen when he explains how Nvidia prices GPUs. Or maybe you do believe him but think that he should literally be auctioning them E-bay style?
There are very good reasons why most businesses don't auction goods. Price stability is important for business, especially when deals happen over several months to years.
>Price stability is important for business, especially when deals happen over several months to years.
Is there a way to quantify this? By choosing to not raise their price to the market-clearing number, Nvidia is at least leaving short term money on the table. Is there some way to show that what they are doing is _economically_ rational, rather than _geopolitically_ rational?
I can't be confident this is important because I am politically naive and ignorant.
But in an effort to fix that: what do Zvi or Dwarkesh say to it? What probability does there need to be of China taking Taiwan for it to be very sensible for Jensen and NVidia to be chasing the Chinese market even with such bad arguments?
My best guess is that Huang is trying to play the game Germany was trying to play a decade ago, when they were trying to be on the US's side and on Russia's side at the same time. In particular, Huang wants to make sure that if China ends up Winning, they won't be angry at him. I would be very curious, for example, what he thinks is the probability that China will take over Taiwan within the next decade, and what he thinks is the probability conditional on that that NVIDIA will still be able to buy from TSMC. Of course he can't talk about that, and shouldn't, so I'll never know. But I suspect that that's the worst-case scenario he's trying to hedge against by demonstrating to China that he's happy to do business with them even when it hurts the US.
"Making a deal to not use AI for [X] is much harder than making a deal to ensure AI can’t do [X] in the first place. Making a broad general agreement can actually be much easier than making narrow agreements."
Agreed! In particular, here I think this looks like: 'Making a deal to ensure AI doesn't further advance, particularly in computer-science-y skills, is easier than making (and verifying!) a deal to ensure AI isn't specifically used for cyberwarfare.'
As to the odds for such a broad deal, Metaculus's market
is currently at 4%. Oddly, it hasn't shifted much in the last month, even after Mythos was shown to discover 0-day vulnerabilities in all major OSes.
Separately, I asked Google/Gemini
"In the wake of Mythos, has the PRC increased its LLM development effort?"
and they answered:
"Based on reports following the April 2026 announcement of Anthropic’s "Claude Mythos Preview"—a highly capable, agentic AI model designed for autonomous cybersecurity operations—the People's Republic of China (PRC) has not only sustained but _intensified_ its Large Language Model (LLM) and artificial intelligence development efforts" [emphasis added]
so I think the odds for such a treaty have gone down. Part of the problem may be that prediction markets have some problem reducing the odds of an already-estimated-to-be-unlikely event. Even if the real odds are now 2%, betting 'no' on what was originally a 4% chance isn't very lucrative.
Do we really know what Mythos was trained on? AWS CEO Garman said in an interview that it was trained on Amazon chips, but should we really believe such a statement? Anthopic recently said that they train and run on everything. Maybe they were trained a little on everything, just so that Anthropic and Garman could say that, but interpreting it as mostly or wholly trained on Amazon chips is not justified.
My impression up to now is that Anthropic trains on nvidia chips. There really is an nvidia edge, but it is only worth the markup in training. I am not convinced that there is any update.
"Radiologists were an interesting case, often discussed. Those warning about this were right that AI would be superhuman at analyzing images.
But this caused an increase in demand for radiology, and AI can’t replace many other parts of the job, and because in the longer run radiology is going to be increasingly automated and doctors have 40 year careers, many opted out of radiology."
__
None of this is true and I'm getting a bit of Gell Man Amnesia.
AI is certainly not superhuman an analyzing images, it's useful for extracting a handful of diagnoses that lend themselves well to automation (such as lung nodule detection), but 90% of a radiologist's workflow is unchanged from 8 years ago.
The increase in demand for radiology is driven by ordering physicians and changing practices, not the radiology side.
And finally, the number of physicians matching into radiology has increased monotonically from 2020-2025.
Re: Radiologists: you can add the obvious point that it is illegal to use AI radiology products without employing a human radiologist, thus making it quite hard for them to displace humans.[1]
My person Wild-Ass Guess is that Jensen is trying to do a "Concert of Europe" type thing and keep his customers/rivals equal in power, and thus divided and unable to co-ordinate against him. If NVIDIA ever gets locked into a monopsony / only has 1 buyer, say because OpenAI or Google Deepmind or Anthropic completely runs away from all the other labs and the other labs fold, then the company is in very big trouble. A single all-powerful buyer has a lot of power in negotiations, power it can use to push for lower prices. That's unacceptable. Profits might fall.
But if instead NVIDIA carefully rations out compute to all the major labs, making sure no one pulls too far ahead of any of the others, he can keep them weak and divided. He can keep them constantly competing with each other for NVIDIA's chips, rather than working together to demand lower prices. Even if that means selling chips for lower prices today, it's still worth it -- if Jensen just sold all his chips to the highest bidder, there's a risk that Google or whatever would buy them all up, use the compute advantage to crush the other labs, and come back to him in 5 years demanding lower prices. If Jensen instead comes up with excuses to evenly spread the chips around, he makes less money today, but can keep making money tomorrow.
There's just one catch. This "balance of power" type thing only works if the US Government doesn't just seize and nationalize all the major labs tomorrow. (Or if the AI Safety movement gets its way, and the major labs voluntarily agree to merge together to tackle the challenge of alignment together.) If that happens, Jensen is going to be negotiating with a single all-powerful customer anyways.
Now, what's the easiest way to prevent all that? Make sure at least one of the major labs isn't American. Artificially pump up Deepseek or Alibaba or whoever, so worst come to worst, you can still play your customers against each other. Even if that means *losing* money, jumping through all sorts of hoops to evade export controls (e.g. producing export-compliant but straight up worse chips, or letting scalpers claim some of the profit on your best chips as they resell 'em to China), it's still worth it in the end. Better to lose a little money today, than a lot of money tomorrow.
Thus, Jensen's observed behavior, where he's tying himself into knots trying to justify his actions. All the arguments against selling to China, are actually arguments FOR selling to China, from his perspective. E.g.
1) The American labs are compute constrained and really want the chips? Great, denying them that will help cut them down to size.
2) The Chinese labs have a lot of talent and electrical power, and would make full use of those chips? Great, then this investment will actually pay off (unlike attempts to boost, say, Mistral or whatever).
3) Mythos proves AI is very important and extremely valuable? Great, then that means it's even more important to make sure NVIDIA gets its share of the pie, rather than say xAI winning and grabbing the entire pie.
4) The Chinese Government seems to be angry at your attempt to strangle Huawei in the cradle and is trying to ban your imports? Great, that's a sign our plan is actually working, let's double down on strangling Huawei.
But the trouble is, Jensen can't just go out and say any of this, so he has to tie himself into knots saying the exact opposite of what he really believes. Our customers don't need our chips, China doesn't even want our chips, in fact no one really wants our chips because this whole AI thing is overblown, and Huawei is so so strong, perhaps even stronger than we are, so there's no way we could ever strangle them in the cradle. (But it's also really, really important that I get to sell my worthless chips anyways, I don't want to allow another repeat of Anthropic. If there's going to be rivals to the major labs, they better be running on NVIDIA.)
I think that's an underrated aspect of this all... Jensen let Anthropic slip through his hands. He's not going to let that happen again. And if the next Anthropic is in China, he's *going* to sell to China, whether Washington approves of that or not. He's not a man weak of will, Jensen Huang.
In conclusion: I don't know how to end this. But hopefully this was helpful.
People who are either much dumber than they are, or are indifferent to telling the truth save when it serves their own personal interests, and will prioritise those interests over everyone else's, in the most literal sense, _and_ who have really important positions:
This one was painful to listen to. Huang's arguments make no sense and one gets the sense he's not used to people pushing back on him. Dwarkesh Patel did well here.
Huang is also just a bit of a dick here and keeps interrupting Dwarkesh.
This is the opposite of the Satya Nadella interview, where Nadella dealt with challenging questions with humility and grace.
Another commenter already made the same point I was going to: Nvidia wants to be a monopoly and avoid any customers having pricing power. That's it. Everything else is window dressing on this position.
on radiology: surprisingly, the reason there are still human radiologists is because AI radiology models are bad! yes, even though in principle 2026 AI is amazing at image classification. Actually-existing radiology models are based on 2010s machine learning and they are not good enough to replace doctors in real life. There's still a regulatory story here, but it's a *different* story -- the bar for getting a diagnostic-assistance model FDA approved is so *low* that lousy models clear it.
Some of the comments here have speculated on Huang’s motivations and I found those perspectives insightful. What doesn’t make sense to me is the idea that he’s just trying to boost the share price.
If:
• Nvidia doesn’t increase prices to meet demand; and
• Nvidia is not able to produce enough chips to meet U.S. demand
Then, how would selling to China increase profits? Nvidia wouldn’t be selling more chips, nor selling more expensive chips, so it shouldn’t make a difference to revenue. Huang is not a dumb guy. Thus there has to be some other reason, which for whatever reason he does not feel able to say out loud, even quietly.
«• Nvidia doesn’t increase prices to meet demand; and
• Nvidia is not able to produce enough chips to meet U.S. demand
Then, how would selling to China increase profits?»
The reason is not consistent with the "you are only as good as your latest trade/quarter" NYC/London/anglo-american mindset: repeat business.
Trying to screw your customers as hard as you can and running away with the loot does not lead to repeat business and people who have a longer term mindset and/or have lots of capital immobilized for the long term want repeat business from broad customer base; it is true that a new sucker is born every minute, but that does not scale much.
To get repeat business at scale it is a good idea to avoid creating what is called in the business jargon a "price umbrella" that both restricts the customer base and attracts competitors.
Huang has been around for decades and has been through several periods of IT "winters" where having a broad customers base for repeat business is valuable.
Note: many IT businesses in the past have sought to shrink and screw their customer base as hard as they could (IBM , DEC, ...) once they had achieved lock-in of enough suckers and that is why few IT businesses last much longer than the retirement of their founder cohort.
I think you have the impression that I am arguing Nvidia should raise prices; but that is not my argument at all. Rather it is the observation that, given they have chosen not to raise prices, there is no obvious additional profit to be found in selling to china. So why is this such a significant priority for Huang?
«I think you have the impression that I am arguing Nvidia should raise prices»
I did not have at all that impression. I had the impression that it was a genuine question as to why allocations rather than pricing seems to be Huang's policy.
«why is this such a significant priority for Huang?»
I think I answered that: "people who have a longer term mindset and/or have lots of capital immobilized for the long term want repeat business from broad customer base".
Actually this may not be *NVIDIA*'s policy: it may be TSMC's policy and NVIDIA is really dependent on TSMC, because what really matters is TSMC's silicon wafer allocations.
If their main customers are all in the USA and are just OpenAI, xAI, Amazon, Azure and he charges them whatever the market will bear they make a huge profit in the short term only and then unleashes long term counterforces because of that "price umbrella". Export controls are in effect a very tall "price umbrella".
But that is incomprehensible for those who believe that "you are only as good as your latest trade/quarter" for example the executives of IBM, DEC, etc. etc. etc.
«What doesn’t make sense to me is the idea that he’s just trying to boost the share price.»
The real question is: why haven't the owners and CEOs of TSMC and NVIDIA taken advantage of market conditions to screw their customers really hard as quickly as possible to cash in and retire early in immense wealth, and leave badly asset-stripped and damaged businesses to their successors, as so many owners and CEOs of USA/UK/... businesses have done?
I'm having trouble parsing "makes predicts markets then price at essentially 0%"
he makes predictions. markets then price the chance of those predictions coming to pass at essentially 0%
Ah, I kept trying to parse "predicts" as a verb rather than a noun.
> My guess is we will want more engineers for a while, not less.
Tech hiring in the US seems frozen or companies are laying off. Can argue about whether this is AI or not, or they are just stealth hiring in lower cost countries, but they are not hiring in the US.
YMMV but just this week a friend of a friend was hired for a junior SDE position in Google in the US. You can have layoffs and hirings at the same time in the same company.
Seems to me that Zvi used a lot of words and missed the point, because the point was unstated. Nvidia must have a silent chain of thought that if China has access to great Nvidia chips, made in Taiwan, then China has less reason to precipitate a military intervention in Taiwan. A military intervention would result in TSMC's production machinery being at least disabled remotely by ASML if not destroyed, unless Trump has already given up Taiwan ahead of time, which maybe he has. Of course Jensen couldn't possibly say that. But how can you miss the backdrop here?
If destroying TSMC (even permanently) were the price China had to pay to reclaim Taiwan, they wouldn't even blink.
@Andy B, I take your point, but. What Xi can do and what Jensen Huang can do are different matters. Xi has many factors to balance. His economy is stressed with inadequate consumer demand and a debt crisis. He continues purging senior military. And he wants to look reasonable while America alienates almost every country and disarms itself by over-spending its weapon stores. I've seen estimates that an attack on Taiwan would cost China alone more than $1 trillion; never mind Rest Of World. But what Jensen Huang can do is much less. He can sell everything his company produces, and try to postpone what he can't really control. His ability to influence China is very limited, but one thing he can do is work hard to figure out a way to sell his chips to them. He can't say out loud, "China's Taiwan decision is a huge risk to Nvidia." But it is. It is an existential risk, perhaps the only existential risk he faces at the moment and for the foreseeable future.
That's fair. (China wouldn't blink at paying trillions for Taiwan, either, but that's beside the point.) I certainly do agree that Huang is trying to make nice with China because they might foreseeably have the power to destroy his company; I just think his goal is simply to try to be on their good side if they take Taiwan (or otherwise Win), rather than ascribing to him the belief that he might have even the slightly influence over China's Taiwan policy, which would be an Elon level of insanity :).
Many Thanks to both of you for a great discussion!
One thing that has been irking me about the whole analysis of Huang's/Nvidia's comments and actions has been, as Zvi said:
>If Nvidia truly didn’t want to pick winners it would allocate fully via price.
Since, at least in the short term, the default profit-maximizing choice is to charge the market-clearing price, it seems really strange to see Nvidia choosing to charge less than that, and allocate between potential customers.
It takes a really contorted view of future demand (architecture lock-in??? - in a world with Codex and Claude Code??? Really???) for it to make sense as a profit-maximizing decision.
As a please-don't-blow-my-foundry-off-the-face-of-the-Earth, trying to sell to the PRC makes much more sense.
Yes, selling below market-clearing makes sense from this geopolitical PoV.
As with football and concert tickets, you get a secondary market (scalpers), giving Huang plausible deniability when the chips end up in China anyway.
Many Thanks! And good point about the plausible deniability for Huang from the secondary market.
I'm confused why you and Zvi don't believe Jensen when he explains how Nvidia prices GPUs. Or maybe you do believe him but think that he should literally be auctioning them E-bay style?
There are very good reasons why most businesses don't auction goods. Price stability is important for business, especially when deals happen over several months to years.
Many Thanks!
>Price stability is important for business, especially when deals happen over several months to years.
Is there a way to quantify this? By choosing to not raise their price to the market-clearing number, Nvidia is at least leaving short term money on the table. Is there some way to show that what they are doing is _economically_ rational, rather than _geopolitically_ rational?
I can't be confident this is important because I am politically naive and ignorant.
But in an effort to fix that: what do Zvi or Dwarkesh say to it? What probability does there need to be of China taking Taiwan for it to be very sensible for Jensen and NVidia to be chasing the Chinese market even with such bad arguments?
My best guess is that Huang is trying to play the game Germany was trying to play a decade ago, when they were trying to be on the US's side and on Russia's side at the same time. In particular, Huang wants to make sure that if China ends up Winning, they won't be angry at him. I would be very curious, for example, what he thinks is the probability that China will take over Taiwan within the next decade, and what he thinks is the probability conditional on that that NVIDIA will still be able to buy from TSMC. Of course he can't talk about that, and shouldn't, so I'll never know. But I suspect that that's the worst-case scenario he's trying to hedge against by demonstrating to China that he's happy to do business with them even when it hurts the US.
What is Tritium in this context? Do you mean Trainium?
Dwarkesh should interview Zvi!
"Making a deal to not use AI for [X] is much harder than making a deal to ensure AI can’t do [X] in the first place. Making a broad general agreement can actually be much easier than making narrow agreements."
Agreed! In particular, here I think this looks like: 'Making a deal to ensure AI doesn't further advance, particularly in computer-science-y skills, is easier than making (and verifying!) a deal to ensure AI isn't specifically used for cyberwarfare.'
As to the odds for such a broad deal, Metaculus's market
https://www.metaculus.com/questions/39149/us-china-agi-limiting-treaty-signed-before-2030/
is currently at 4%. Oddly, it hasn't shifted much in the last month, even after Mythos was shown to discover 0-day vulnerabilities in all major OSes.
Separately, I asked Google/Gemini
"In the wake of Mythos, has the PRC increased its LLM development effort?"
and they answered:
"Based on reports following the April 2026 announcement of Anthropic’s "Claude Mythos Preview"—a highly capable, agentic AI model designed for autonomous cybersecurity operations—the People's Republic of China (PRC) has not only sustained but _intensified_ its Large Language Model (LLM) and artificial intelligence development efforts" [emphasis added]
so I think the odds for such a treaty have gone down. Part of the problem may be that prediction markets have some problem reducing the odds of an already-estimated-to-be-unlikely event. Even if the real odds are now 2%, betting 'no' on what was originally a 4% chance isn't very lucrative.
No evidence, but I do wonder if the CCP has threatened Jensen. Like if demand is saturated, there’s no reason to sell to China for now
Do we really know what Mythos was trained on? AWS CEO Garman said in an interview that it was trained on Amazon chips, but should we really believe such a statement? Anthopic recently said that they train and run on everything. Maybe they were trained a little on everything, just so that Anthropic and Garman could say that, but interpreting it as mostly or wholly trained on Amazon chips is not justified.
My impression up to now is that Anthropic trains on nvidia chips. There really is an nvidia edge, but it is only worth the markup in training. I am not convinced that there is any update.
"Radiologists were an interesting case, often discussed. Those warning about this were right that AI would be superhuman at analyzing images.
But this caused an increase in demand for radiology, and AI can’t replace many other parts of the job, and because in the longer run radiology is going to be increasingly automated and doctors have 40 year careers, many opted out of radiology."
__
None of this is true and I'm getting a bit of Gell Man Amnesia.
AI is certainly not superhuman an analyzing images, it's useful for extracting a handful of diagnoses that lend themselves well to automation (such as lung nodule detection), but 90% of a radiologist's workflow is unchanged from 8 years ago.
The increase in demand for radiology is driven by ordering physicians and changing practices, not the radiology side.
And finally, the number of physicians matching into radiology has increased monotonically from 2020-2025.
I am a radiologist. I’d say 99% of our workflow is unchanged from 8 years ago.
There are some AI tools out there but they are marginal and rads debate if they improve efficiency at all.
I like Jensen and NVDIA but I am surprised he keeps saying this.
Re: Radiologists: you can add the obvious point that it is illegal to use AI radiology products without employing a human radiologist, thus making it quite hard for them to displace humans.[1]
1. https://forum.effectivealtruism.org/posts/LPgBtsnXtYyWdDurP/radiology-automation-does-not-generalize-to-other-jobs
My person Wild-Ass Guess is that Jensen is trying to do a "Concert of Europe" type thing and keep his customers/rivals equal in power, and thus divided and unable to co-ordinate against him. If NVIDIA ever gets locked into a monopsony / only has 1 buyer, say because OpenAI or Google Deepmind or Anthropic completely runs away from all the other labs and the other labs fold, then the company is in very big trouble. A single all-powerful buyer has a lot of power in negotiations, power it can use to push for lower prices. That's unacceptable. Profits might fall.
But if instead NVIDIA carefully rations out compute to all the major labs, making sure no one pulls too far ahead of any of the others, he can keep them weak and divided. He can keep them constantly competing with each other for NVIDIA's chips, rather than working together to demand lower prices. Even if that means selling chips for lower prices today, it's still worth it -- if Jensen just sold all his chips to the highest bidder, there's a risk that Google or whatever would buy them all up, use the compute advantage to crush the other labs, and come back to him in 5 years demanding lower prices. If Jensen instead comes up with excuses to evenly spread the chips around, he makes less money today, but can keep making money tomorrow.
There's just one catch. This "balance of power" type thing only works if the US Government doesn't just seize and nationalize all the major labs tomorrow. (Or if the AI Safety movement gets its way, and the major labs voluntarily agree to merge together to tackle the challenge of alignment together.) If that happens, Jensen is going to be negotiating with a single all-powerful customer anyways.
Now, what's the easiest way to prevent all that? Make sure at least one of the major labs isn't American. Artificially pump up Deepseek or Alibaba or whoever, so worst come to worst, you can still play your customers against each other. Even if that means *losing* money, jumping through all sorts of hoops to evade export controls (e.g. producing export-compliant but straight up worse chips, or letting scalpers claim some of the profit on your best chips as they resell 'em to China), it's still worth it in the end. Better to lose a little money today, than a lot of money tomorrow.
Thus, Jensen's observed behavior, where he's tying himself into knots trying to justify his actions. All the arguments against selling to China, are actually arguments FOR selling to China, from his perspective. E.g.
1) The American labs are compute constrained and really want the chips? Great, denying them that will help cut them down to size.
2) The Chinese labs have a lot of talent and electrical power, and would make full use of those chips? Great, then this investment will actually pay off (unlike attempts to boost, say, Mistral or whatever).
3) Mythos proves AI is very important and extremely valuable? Great, then that means it's even more important to make sure NVIDIA gets its share of the pie, rather than say xAI winning and grabbing the entire pie.
4) The Chinese Government seems to be angry at your attempt to strangle Huawei in the cradle and is trying to ban your imports? Great, that's a sign our plan is actually working, let's double down on strangling Huawei.
But the trouble is, Jensen can't just go out and say any of this, so he has to tie himself into knots saying the exact opposite of what he really believes. Our customers don't need our chips, China doesn't even want our chips, in fact no one really wants our chips because this whole AI thing is overblown, and Huawei is so so strong, perhaps even stronger than we are, so there's no way we could ever strangle them in the cradle. (But it's also really, really important that I get to sell my worthless chips anyways, I don't want to allow another repeat of Anthropic. If there's going to be rivals to the major labs, they better be running on NVIDIA.)
I think that's an underrated aspect of this all... Jensen let Anthropic slip through his hands. He's not going to let that happen again. And if the next Anthropic is in China, he's *going* to sell to China, whether Washington approves of that or not. He's not a man weak of will, Jensen Huang.
In conclusion: I don't know how to end this. But hopefully this was helpful.
This is precisely it. Huang wants Nvidia to be a monopoly and to avoid any of its customers approach monopsony.
His arguments make no sense otherwise.
People who are either much dumber than they are, or are indifferent to telling the truth save when it serves their own personal interests, and will prioritise those interests over everyone else's, in the most literal sense, _and_ who have really important positions:
1. Sam Altmam
2. Jensen Huang
This one was painful to listen to. Huang's arguments make no sense and one gets the sense he's not used to people pushing back on him. Dwarkesh Patel did well here.
Huang is also just a bit of a dick here and keeps interrupting Dwarkesh.
This is the opposite of the Satya Nadella interview, where Nadella dealt with challenging questions with humility and grace.
Another commenter already made the same point I was going to: Nvidia wants to be a monopoly and avoid any customers having pricing power. That's it. Everything else is window dressing on this position.
on radiology: surprisingly, the reason there are still human radiologists is because AI radiology models are bad! yes, even though in principle 2026 AI is amazing at image classification. Actually-existing radiology models are based on 2010s machine learning and they are not good enough to replace doctors in real life. There's still a regulatory story here, but it's a *different* story -- the bar for getting a diagnostic-assistance model FDA approved is so *low* that lousy models clear it.
https://x.com/s_r_constantin/status/1971193406455157085
Some of the comments here have speculated on Huang’s motivations and I found those perspectives insightful. What doesn’t make sense to me is the idea that he’s just trying to boost the share price.
If:
• Nvidia doesn’t increase prices to meet demand; and
• Nvidia is not able to produce enough chips to meet U.S. demand
Then, how would selling to China increase profits? Nvidia wouldn’t be selling more chips, nor selling more expensive chips, so it shouldn’t make a difference to revenue. Huang is not a dumb guy. Thus there has to be some other reason, which for whatever reason he does not feel able to say out loud, even quietly.
«• Nvidia doesn’t increase prices to meet demand; and
• Nvidia is not able to produce enough chips to meet U.S. demand
Then, how would selling to China increase profits?»
The reason is not consistent with the "you are only as good as your latest trade/quarter" NYC/London/anglo-american mindset: repeat business.
Trying to screw your customers as hard as you can and running away with the loot does not lead to repeat business and people who have a longer term mindset and/or have lots of capital immobilized for the long term want repeat business from broad customer base; it is true that a new sucker is born every minute, but that does not scale much.
To get repeat business at scale it is a good idea to avoid creating what is called in the business jargon a "price umbrella" that both restricts the customer base and attracts competitors.
Huang has been around for decades and has been through several periods of IT "winters" where having a broad customers base for repeat business is valuable.
Note: many IT businesses in the past have sought to shrink and screw their customer base as hard as they could (IBM , DEC, ...) once they had achieved lock-in of enough suckers and that is why few IT businesses last much longer than the retirement of their founder cohort.
I think you have the impression that I am arguing Nvidia should raise prices; but that is not my argument at all. Rather it is the observation that, given they have chosen not to raise prices, there is no obvious additional profit to be found in selling to china. So why is this such a significant priority for Huang?
«I think you have the impression that I am arguing Nvidia should raise prices»
I did not have at all that impression. I had the impression that it was a genuine question as to why allocations rather than pricing seems to be Huang's policy.
«why is this such a significant priority for Huang?»
I think I answered that: "people who have a longer term mindset and/or have lots of capital immobilized for the long term want repeat business from broad customer base".
Actually this may not be *NVIDIA*'s policy: it may be TSMC's policy and NVIDIA is really dependent on TSMC, because what really matters is TSMC's silicon wafer allocations.
If their main customers are all in the USA and are just OpenAI, xAI, Amazon, Azure and he charges them whatever the market will bear they make a huge profit in the short term only and then unleashes long term counterforces because of that "price umbrella". Export controls are in effect a very tall "price umbrella".
But that is incomprehensible for those who believe that "you are only as good as your latest trade/quarter" for example the executives of IBM, DEC, etc. etc. etc.
«What doesn’t make sense to me is the idea that he’s just trying to boost the share price.»
The real question is: why haven't the owners and CEOs of TSMC and NVIDIA taken advantage of market conditions to screw their customers really hard as quickly as possible to cash in and retire early in immense wealth, and leave badly asset-stripped and damaged businesses to their successors, as so many owners and CEOs of USA/UK/... businesses have done?