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.
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
Just like US usually does for Oil, why doesn't it do it for human resource?
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.
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.
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.
- 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?
What fools!
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 :(.
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.
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.
AndrewKemendo•3mo ago
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.