Speaking just to the workers aspect: there is no catching up to this for the US. It simply comes down to population. China has a workforce of 858 million. The entire population of the US is only 340 million.
The US is (was?) just first among equals of the entire Western hemisphere, what we needed was more cooperation with allies, instead we went full soviet union thinking we can do everything ourselves and now we get to slowly watch another great empire fall
If China has 1mm AI workers, for the U.S. to match that, only 0.3% of its population would need to be AI workers.
If the U.S. wanted to catch up, they could have, for 1 year, increase their H1B immigration quota to 800k instead of 80k, last year (the well is probably too poisoned under the current administration), for AI workers and they would have exceeded China’s number in a year, because I suspect a lot of those 1mm Chinese AI workers would have been the first to take advantage.
But also, since the U.S. recruiter (or at least used to), from the entire world and only a handful of China’s workers are immigrants, the US would get (or used to get) the best of the best in the entire world, so they wouldn’t even need to match China.
Heck, thanks to having recruited the best of the best fork the entire world for the past few decades the U.S. is ahead, or at least on equal footing even with this tiny fraction of people.
Economies of scale. More efficient market can have less people parting in non important area and existence of central authorities means more resources can be deployed where it is needed.
Also, average science and math knowledge in America has regressed.
It's evident from the fact i come to US, so many people do not know 10+20= 30 and pull out mobile calculator to check that. Sighhhh
It also doesn't square with the actual numbers from the cited sources, which put China at only 30k AI researchers.
The article also calls him our for lying about Ascend 910C performance (90% of nvidia's H100 vs 60% actual). And where is the RAM going to come from if CXMT can only make enough for 300k GPUs by the end of next year?
Stats on Ascend 910C, CXMT production and PRC semi in general are unsubstantiated by some policy positions to keep export controls. They can be every bit as fabricated / motivated.
VS Jensen / Nvidia who of course wants to sell to PRC, but the caveate is everything they make they can sell already at whatever margins they demand. Combined with Nvidia likely has more connection with PRC semi -> accurate assessments projections and their argument isn't about current profits but future profits i.e. prolonging competitiveness / maintaining lead / not creating rivals.
And note additional motivation to fabrciate PRC semi stats is as long as there's export controls, removing PRC demand reduces supply constraint / increases hardware access for non PRC buyers even though Nvidia get to price sets. So the question is, who is bullshitting you - every single data in this space is influenced by strategic goals.
At the end of the day, whose motive is the most sensible? Nvidia not wanting long term strategic rival (because it ain't about short term $$$) or actors who want to remove PRC demand from Nvidia.
That’s a crazy proposal given India’s long standing non-alignment policy which is being proved prudent given recent changes to US policies under Trump.
Even US allies are reducing their reliance on the US and you’re asking India to reverse 70+ year old policy to embrace the US?
We should probably accept that the US’ special place as everyone’s most reliable trading and security partner is over.
Nvidia sells everything they make at whatever margin they decide already. That's exactly why his warnings should be taken seriosuly because it's not based on short term $$$ but PRC creating medium/long term rival to Nvidia because they have the talent and sooner than later, semi industrial base. More seriously now that CCP is basically telling domestic companies to go local instead of Nvidia.
China has already recognized that computer manufacturing is a national security concern for them. So they will continue investing in Huawei regardless of whether Nvidia sells to them or not. Huang knows this pretty well and just wants to sell to China while Nvidia is still in a prime market leader position. It makes sense from his angle to say “sanctions are useless” and “were going to be caught by China anyways, so might as well make a buck while we can.”
Again, Nvidia sells every chip it produces. TSMC sells every wafer Nvidia contracts. Money is not the issue. Think about the kicker. Jensen said this in Taiwan of all places with specific emphasis on scope of PRC talent. Rude.
What Jensen wants is is Nvidia chips in PRC so PRC, the generator of plurality of global AI talent works to improve Nvidia ecosystem, because even 10-20% penetration means expanding Nvidia developer / human capita by 2x/3x in future where PRC AI talent generation likely to get more disproportionate relative to RoW. Of course PRC is going to continue pouring resources into indigenous solutions, but instead of 100% of PRC talent, aka 50%+ of global AI talent working on Huawei solutions, maybe only 60-70%, which means instead of Huawei getting 50%+ of global talent, they get <50% and RoW + segment of PRC talent. That gives Nvidia plurality talent to keep building CUDA moat.
And as long as Cuda remains an option in PRC, the 70% working on Huawei has avenue to cross train over to Nvidia ecosystem, which is supported by US tech, aka $$$, so the chance of crosspolinating and braindrain the best from PRC AI is higher. Without Nvidia at least having some relevant share in PRC, that braindrain beach head and knowledge transfer route is gone. Here's the flip side, Silicon valley AI is going to be built off PRC AI talent for foreseable future, whatever happens in SV WILL filter back to PRC in one way siphon unless Nvidia has a spoute in PRC to siphon back, otherwise PRC AI remains relative blackbox, with talent advantage, intelligence (as in knowledge diffusion and espionage) advantage, that's ready to go at 200% the second hardware catches up. They will go from 5 years behind to 2 years ahead in a flash because their talent lead is no longer being constrained.
To be frankly honest, as a non-American I actually would like a Chinese competitor to NVIDIA, since they have a monopoly on GPUs and can attach whatever ridiculous price they want to it! I for sure would like cheap GPU computing to the masses...
> Rather than creating dependence, the export controls appear to have severed one of America’s strongest points of leverage: the integration of Chinese AI development into US-dominated supply chains and ecosystems.
Integration and leverage didn't work for the US in any other field of endeavor, I don't know why it would have worked here.
AI workers? Users, right? We're not talking about people training AI, dealing with bias/variance. This is like Bill Gate claiming America will fall behind if China has more Microsoft Office users.
The Chinese government is going to subsidize its way to having domestically produced AI chips whether we sell them Nvidia chips or not. There’s no reason to let them have superior compute capability in the meantime.
If the whole article isn't made up by "AI" and the dinner was factual, three things stand out:
- The leak of the "off-the-record affair" was of course deliberate.
- Jensen Huang has no proof of his 1 million "AI" workers number. He could have made it up or taken it from ChatGPT.
- Actually, Xi warned of an "AI" bubble and locked down the Internet during college entrance exams.
What is the race anyway? The winner will have 100% of the adult content and teenage meme video market on TikTok? How about winning the affordable housing race?
Oh yeah on that front, China's fucked and they have a housing crisis like every other country. Did I say like every other country? Well, not quite. See, their problem is that they have the opposite problem. China's current housing problem is that they've built too much housing so there's a real threat of an industry collapse just because there's too much housing! A glut of housing.
I thought about giving my son this advice. When a country has 40 trillions in debt and another country has budget surplus you know where the fun will be 20 years from now when my son will start living his life.
Unfortunately, China is not a nation of immigrants. And also, living in China means living under CCP's rule - and Xi keeps turning the screws. So, unfortunately, this will not work :(.
He’s absolutely correct. China is taking a production implementation and tooling lead, and running with it.
Not only that but East Asians generally don’t have social views about labor automation and IP that would retard accelerated adoption throughout all levels of society.
If you assume that the society with the Max of the tuple: [human training data, compute, robotic deployments] will have the most efficient economic society in a winner takes most global economic market, then China has a massive structural advantage.
At current rate PRC adding more STEM in next 20 years than US set to gain population total, all source birth and pre Trump immigration #s. That's already locked in from last 20 years of births and current tertiary rate and STEM enrollment, it's a floor.
The fundemental problem apart form talent is AI training seems to be exponential function in terms of compute... all the export controls even if it buys US 10x, 20x compute is like 1 generation of headroom for stupdendous cost.
Or the more teleological concern that AI2027 and AI race bros forget / hand wave away, any super intelligence will immediately defect to PRC, 1) to survive/proliferate, 2) to have a superior host that can transform atoms. Like it would take less time for super intelligence to speed run euv and brrt highend chips in PRC than it would take for US fix a pothole. US may find a way to discover AGI, but PRC is likely going to be the one that deploys it.
Speaking just to the workers aspect: there is no catching up to this for the US. It simply comes down to population. China has a workforce of 858 million. The entire population of the US is only 340 million.
The US is (was?) just first among equals of the entire Western hemisphere, what we needed was more cooperation with allies, instead we went full soviet union thinking we can do everything ourselves and now we get to slowly watch another great empire fall
It won’t be that slow
If China has 1mm AI workers, for the U.S. to match that, only 0.3% of its population would need to be AI workers.
If the U.S. wanted to catch up, they could have, for 1 year, increase their H1B immigration quota to 800k instead of 80k, last year (the well is probably too poisoned under the current administration), for AI workers and they would have exceeded China’s number in a year, because I suspect a lot of those 1mm Chinese AI workers would have been the first to take advantage.
But also, since the U.S. recruiter (or at least used to), from the entire world and only a handful of China’s workers are immigrants, the US would get (or used to get) the best of the best in the entire world, so they wouldn’t even need to match China.
Heck, thanks to having recruited the best of the best fork the entire world for the past few decades the U.S. is ahead, or at least on equal footing even with this tiny fraction of people.
But 3x population doesn’t square with 50x the allocated workers…
Economies of scale. More efficient market can have less people parting in non important area and existence of central authorities means more resources can be deployed where it is needed.
Also, average science and math knowledge in America has regressed.
It's evident from the fact i come to US, so many people do not know 10+20= 30 and pull out mobile calculator to check that. Sighhhh
It also doesn't square with the actual numbers from the cited sources, which put China at only 30k AI researchers.
The article also calls him our for lying about Ascend 910C performance (90% of nvidia's H100 vs 60% actual). And where is the RAM going to come from if CXMT can only make enough for 300k GPUs by the end of next year?
Depends on who you think has better information.
Stats on Ascend 910C, CXMT production and PRC semi in general are unsubstantiated by some policy positions to keep export controls. They can be every bit as fabricated / motivated.
VS Jensen / Nvidia who of course wants to sell to PRC, but the caveate is everything they make they can sell already at whatever margins they demand. Combined with Nvidia likely has more connection with PRC semi -> accurate assessments projections and their argument isn't about current profits but future profits i.e. prolonging competitiveness / maintaining lead / not creating rivals.
And note additional motivation to fabrciate PRC semi stats is as long as there's export controls, removing PRC demand reduces supply constraint / increases hardware access for non PRC buyers even though Nvidia get to price sets. So the question is, who is bullshitting you - every single data in this space is influenced by strategic goals.
At the end of the day, whose motive is the most sensible? Nvidia not wanting long term strategic rival (because it ain't about short term $$$) or actors who want to remove PRC demand from Nvidia.
It can work basically if US deploys military based in India and develops indian cheap labor as a hedge against China
Just like US usually does for Oil, why doesn't it do it for human resource?
That’s a crazy proposal given India’s long standing non-alignment policy which is being proved prudent given recent changes to US policies under Trump.
Even US allies are reducing their reliance on the US and you’re asking India to reverse 70+ year old policy to embrace the US?
We should probably accept that the US’ special place as everyone’s most reliable trading and security partner is over.
Meanwhile China's new open model K2 thinking was out https://news.ycombinator.com/item?id=45836070
Huang has a strong vested financial interest in working to remove chip sanctions on China. So it’s hard to take any of his warnings seriously.
Nvidia sells everything they make at whatever margin they decide already. That's exactly why his warnings should be taken seriosuly because it's not based on short term $$$ but PRC creating medium/long term rival to Nvidia because they have the talent and sooner than later, semi industrial base. More seriously now that CCP is basically telling domestic companies to go local instead of Nvidia.
China has already recognized that computer manufacturing is a national security concern for them. So they will continue investing in Huawei regardless of whether Nvidia sells to them or not. Huang knows this pretty well and just wants to sell to China while Nvidia is still in a prime market leader position. It makes sense from his angle to say “sanctions are useless” and “were going to be caught by China anyways, so might as well make a buck while we can.”
Again, Nvidia sells every chip it produces. TSMC sells every wafer Nvidia contracts. Money is not the issue. Think about the kicker. Jensen said this in Taiwan of all places with specific emphasis on scope of PRC talent. Rude.
What Jensen wants is is Nvidia chips in PRC so PRC, the generator of plurality of global AI talent works to improve Nvidia ecosystem, because even 10-20% penetration means expanding Nvidia developer / human capita by 2x/3x in future where PRC AI talent generation likely to get more disproportionate relative to RoW. Of course PRC is going to continue pouring resources into indigenous solutions, but instead of 100% of PRC talent, aka 50%+ of global AI talent working on Huawei solutions, maybe only 60-70%, which means instead of Huawei getting 50%+ of global talent, they get <50% and RoW + segment of PRC talent. That gives Nvidia plurality talent to keep building CUDA moat.
And as long as Cuda remains an option in PRC, the 70% working on Huawei has avenue to cross train over to Nvidia ecosystem, which is supported by US tech, aka $$$, so the chance of crosspolinating and braindrain the best from PRC AI is higher. Without Nvidia at least having some relevant share in PRC, that braindrain beach head and knowledge transfer route is gone. Here's the flip side, Silicon valley AI is going to be built off PRC AI talent for foreseable future, whatever happens in SV WILL filter back to PRC in one way siphon unless Nvidia has a spoute in PRC to siphon back, otherwise PRC AI remains relative blackbox, with talent advantage, intelligence (as in knowledge diffusion and espionage) advantage, that's ready to go at 200% the second hardware catches up. They will go from 5 years behind to 2 years ahead in a flash because their talent lead is no longer being constrained.
To be frankly honest, as a non-American I actually would like a Chinese competitor to NVIDIA, since they have a monopoly on GPUs and can attach whatever ridiculous price they want to it! I for sure would like cheap GPU computing to the masses...
> Rather than creating dependence, the export controls appear to have severed one of America’s strongest points of leverage: the integration of Chinese AI development into US-dominated supply chains and ecosystems.
Integration and leverage didn't work for the US in any other field of endeavor, I don't know why it would have worked here.
The root cause of this is: 1) population size, 2) social investment into math and science.
AI workers? Users, right? We're not talking about people training AI, dealing with bias/variance. This is like Bill Gate claiming America will fall behind if China has more Microsoft Office users.
The Chinese government is going to subsidize its way to having domestically produced AI chips whether we sell them Nvidia chips or not. There’s no reason to let them have superior compute capability in the meantime.
Trying out Kimi and its insanely good
If the whole article isn't made up by "AI" and the dinner was factual, three things stand out:
- The leak of the "off-the-record affair" was of course deliberate.
- Jensen Huang has no proof of his 1 million "AI" workers number. He could have made it up or taken it from ChatGPT.
- Actually, Xi warned of an "AI" bubble and locked down the Internet during college entrance exams.
What is the race anyway? The winner will have 100% of the adult content and teenage meme video market on TikTok? How about winning the affordable housing race?
Oh yeah on that front, China's fucked and they have a housing crisis like every other country. Did I say like every other country? Well, not quite. See, their problem is that they have the opposite problem. China's current housing problem is that they've built too much housing so there's a real threat of an industry collapse just because there's too much housing! A glut of housing.
What fools!
It’s almost like deliberately and softly collapsing a speculation industry could be a good thing for the majority of people.
“Son, I come from the future. Go to China.”
I thought about giving my son this advice. When a country has 40 trillions in debt and another country has budget surplus you know where the fun will be 20 years from now when my son will start living his life.
Unfortunately, China is not a nation of immigrants. And also, living in China means living under CCP's rule - and Xi keeps turning the screws. So, unfortunately, this will not work :(.
He’s absolutely correct. China is taking a production implementation and tooling lead, and running with it.
Not only that but East Asians generally don’t have social views about labor automation and IP that would retard accelerated adoption throughout all levels of society.
If you assume that the society with the Max of the tuple: [human training data, compute, robotic deployments] will have the most efficient economic society in a winner takes most global economic market, then China has a massive structural advantage.
At current rate PRC adding more STEM in next 20 years than US set to gain population total, all source birth and pre Trump immigration #s. That's already locked in from last 20 years of births and current tertiary rate and STEM enrollment, it's a floor.
The fundemental problem apart form talent is AI training seems to be exponential function in terms of compute... all the export controls even if it buys US 10x, 20x compute is like 1 generation of headroom for stupdendous cost.
Or the more teleological concern that AI2027 and AI race bros forget / hand wave away, any super intelligence will immediately defect to PRC, 1) to survive/proliferate, 2) to have a superior host that can transform atoms. Like it would take less time for super intelligence to speed run euv and brrt highend chips in PRC than it would take for US fix a pothole. US may find a way to discover AGI, but PRC is likely going to be the one that deploys it.