Comment by xmprt
5 hours ago
Relatives vs absolutes. America will spend $500B and because of leaky pipes that's effectively 100B going directly to what's needed. China gets a lot of bang for their buck so even if they're spending a fraction of the US, they make it worth their money.
That's true for domestic labor and manufacturing, like shipbuilding, but the bleeding edge chips only come from one place and the US labs get the best while Chinese labs do not (unless they smuggle them). China gets creative, sure, but it can't overcome the fundamental issue that the US labs have more magic rocks that do math faster than their magic rocks. And the current state of the art is to just "do more math".
Nvidia GPUs aren't even the absolute SOTA for LLM inference anymore. Some labs are moving to ASICs and Huawei already have their own custom chips running DeepSeek as we speak.
There's enough money and scale on the line that software affinity like CUDA is no longer the deciding factor and there's margin for custom stacks.
Even more so after the USA GPU exports ban which is proving to have backfired by speeding up China's tech growth.