The big question then is how Deepseek continues to improve its LLMs over the next few years without access to the latest generation of Nvidia GPUs.
Nonetheless, Deepseek R1 is a tremendous accomplishment by an incredible team, and they deserve to be applauded for open sourcing such an incredible LLM.
H800 GPUs were not subject to US export controls to China until Oct 2023. Tencent built up a huge stockpile of H800s (https://www.businessinsider.com/tencent-has-big-nvidia-h800-ai-us-chip-stockpile-2023-11).
Presumably, Deepseek did as well, and these are the GPUs that allowed it to train R1 (https://stratechery.com/2025/deepseek-faq/).
Deepseek CEO Liang Wenfeng has stated: "Money has never been the problem for us; bans on shipments of advanced chips are the problem." (https://www.chinatalk.media/p/deepseek-ceo-interview-with-chinas)
The big question then is how Deepseek continues to improve its LLMs over the next few years without access to the latest generation of Nvidia GPUs.
Nonetheless, Deepseek R1 is a tremendous accomplishment by an incredible team, and they deserve to be applauded for open sourcing such an incredible LLM.