Remix.run Logo
▲ fakedang 5 hours ago

Well for starters, the US will need to rip up the CLOUD Act, get on its knees to win back Western goodwill and restart resource extraction and onshore manufacturing. Oh, and restart research spending across all domains.

On the other hand, China could just replicate GPU technology or adapt its models to its own chips (already happened), and then push that onto the rest of the world for cheap. Of late, they too have been very aggressively building out their data centers in Europe and the Middle East. China also controls the resource chain for most essential minerals (thanks to decades of US offshoring).

▲achierius 4 hours ago | parent [-]

Already happened? Where's the Huawei B100 equivalent? Why are the latest Chinese models trained on Nvidia hardware and not Ascends?

▲bostik 3 hours ago | parent | next [-]

A recent episode of Odd Lots had a segment about this question. Their guest, a senior analyst who has been tracking the GPU hardware space (+ politics around it) for years had quite revealing observations.

Huawei's hardware is apparently good enough for inference and the CCP are actively promoting their domestic AI gear ... among other things by banning high-end chip imports. The export bans from the US and import bans from China both have same technical targets but different political goals. The analyst commented that Huawei's gear is not yet good enough for model training.

The situation is going to change once China can - to quote the analyst - "complete a training run on Huawei chips". When that happens, China will start to really crack down on AI chip imports to ensure that their domestic technology is forced to catch up.

▲phatfish 3 hours ago | parent | prev [-]

Surely this won't be a problem with super intelligent LLMs?