*China may welcome the new H200s initially—but if past is precedent, Huawei or another national champion will rapidly release a new product eerily similar to the H200, leading to a precipitous drop in Nvidia’s sales as China captures that $50 billion domestic market for itself. We have seen this play many times. Telecommunications company Cisco, for example, dominated China’s early internet infrastructure until Huawei successfully cloned its source code, allowing Beijing to systematically purge the U.S. giant in favor of its own sovereign enterprise.*
I think Chinese open models enabled by Nvidia GPUs have been a massive boon for US small businesses and startups who cannot afford OpenAI et al’s exorbitant margins.
Western firms like Cursor and Windsurf with their own Chinese derived models (Composer and SWE-1) is a testament to how an open competitive market helps all firms US ones included.
OpenAI bears the original sin for closing off most open industrial AI research and publications by defecting first. Google was once a lot more open with their research (including the Transformers paper) but they have been forced to be more selective.
I think AI is just another relatively “normal” technology and it would not be existential. Even without restrictions I think the US with all of their advantages would “win” the race as is. Hopper is approximately 3 years old and I don’t think it will cause a crisis if Chinese companies had access to them.
For the sake of non-incumbents and the AI research community as a whole it seems like the best thing to happen is to allow Chinese firms and researchers to import 3 year old technology by paying substantial amount of revenue to a US firm.
2 Comments
*China may welcome the new H200s initially—but if past is precedent, Huawei or another national champion will rapidly release a new product eerily similar to the H200, leading to a precipitous drop in Nvidia’s sales as China captures that $50 billion domestic market for itself. We have seen this play many times. Telecommunications company Cisco, for example, dominated China’s early internet infrastructure until Huawei successfully cloned its source code, allowing Beijing to systematically purge the U.S. giant in favor of its own sovereign enterprise.*
https://archive.ph/edeIa
I think Chinese open models enabled by Nvidia GPUs have been a massive boon for US small businesses and startups who cannot afford OpenAI et al’s exorbitant margins.
Western firms like Cursor and Windsurf with their own Chinese derived models (Composer and SWE-1) is a testament to how an open competitive market helps all firms US ones included.
OpenAI bears the original sin for closing off most open industrial AI research and publications by defecting first. Google was once a lot more open with their research (including the Transformers paper) but they have been forced to be more selective.
I think AI is just another relatively “normal” technology and it would not be existential. Even without restrictions I think the US with all of their advantages would “win” the race as is. Hopper is approximately 3 years old and I don’t think it will cause a crisis if Chinese companies had access to them.
For the sake of non-incumbents and the AI research community as a whole it seems like the best thing to happen is to allow Chinese firms and researchers to import 3 year old technology by paying substantial amount of revenue to a US firm.