The H200 is a more powerful H100, and the H100 is far from obsolete - for example it is what Musk's Colossus-1 data center, currently being rented to Anthropic, uses.

The only difference between using a slower chip such as H100 (or Huawei's Ascend 750) vs NVIDIA's newer Blackwell chips (B200 etc) is that you need more of the slower chips to achieve the same total FLOPs in your cluster. It has zero effect on what models you can run on it.

It's hard to understand: Do you think that having to use chips that were 20% less performant would lock China out of anything? Are you not aware that with the low costs they have, they can just stack ten times or more datacenters and run workloads in parallel to make up for that performance difference - even if there was actually one that high?

No? They are actively in the race, what are you talking about