<aside> <img src="/icons/circle-dashed_purple.svg" alt="/icons/circle-dashed_purple.svg" width="40px" />

"They must have broken the rules to do this"

</aside>

Reality

• Explanation: Funnily enough, DeepSeek is getting so much attention because of how they innovated within export control constraints.
  • What that means is because they had to use the less powerful H800 GPUs, they had to come up with creatives ways to optimize their model architecture. Most analysts agree that had DeepSeek been able to access the more powerful H100 chips, they would have done things differently.
  • Note: The chip ban prohibits sales of H100s to Chinese companies, but sales of the nerfed H800s were still allowed back then. Also, both H100s and H800s are "Hopper generation" GPUs from NVIDIA, so DeepSeek having 50,000 Hopper GPUs makes sense.
    • Oversimplified table
• Analogy: Like how Samsung legally uses slightly slower processors in some regions due to licensing agreements. They didn't break rules - they just optimized their design around the constraints.