[HarHarVeryFunny]

But it actually had the exact opposite effect.

First the US blocked China from buying NVIDIA's H100, but allowed NVIDIA to sell them a China-special nerfed H100, the H800

Then the US blocked the H800

Then the US realized that China was indeed accelerating their US independence, so does a U-turn and has now approved the H200 (more powerful than both the H100 and H800) for sale to China, on a case-by-case basis

However - and here is the real kicker - China themselves are now blocking H200 purchases since they want the acceleration towards Chinese homegrown solutions to continue, and now we have Chinese models being served on Huawei Ascend chips, with next generation Ascend 750 chips (using CXMT made memory) targetting training currently in testing.

Now we have Apple asking the US government for permission to buy memory from CXMT given the global shortage!

[aspenmartin]

The Jensen argument that "oh we get them hooked on our technology and that will actually be better" is bullshit -- they would do the exact same thing they're doing now of building their own supply chains and compute power but be able to accelerate their progress in the meantime with US chips. It is worse in every way to sell them SOTA chips (or even previous generation chips).

Of course they have ways around this -- you can get black market GPUs and also API costs are SUPER cheap there -- they hack the subscription model, bundle a bunch of user accounts, and route API requests through them.

And yes they are getting to parity with US technology and will get there in a few years, they have decent chips but still not the quality of NVIDIA.

It's really a very complex situation

[HarHarVeryFunny]

I don't know what you mean by the "quality" of NVIDIA. All the western AI accelerators, from NVIDIA, AMD, Google (TPU) and Amazon (Trainium) are made by TSMC, and their speed/density is only possible due to TSMC using EUV machines made by ASML (a dutch company).

Without access to ASML EUV machines, the Chinese will be stuck on older less-dense chip manufacturing nodes, but in terms of building a cluster this is just a cost/efficiency issue - it means you need more chips, more electricity!

[willy_k]

There is a whole lot more to chip and GPU design than the silicon manufacturing itself

[HarHarVeryFunny]

Sure, but the Chinese are plenty competent at processor design - it's the EUV manufacturing part they are lacking