claim C-0056 Partly supported unnamed source
Unnamed senior administration official: DeepSeek’s upcoming model was trained on Blackwell, likely in Inner Mongolia
Unnamed-source claim. The official also said DeepSeek could remove technical indicators of US-chip use, and did not say how the chips were obtained. Partly supported: DePue's unreplicated forensics reads DeepSeek-V4 (April 2026) as Blackwell-initialized. The forensics cannot speak to location, so the Inner Mongolia detail remains unsupported.
Sources
Related
- C-0044 DePue report: six models from four labs were initialized on Blackwell-class hardware (148/152 SMs); earlier models read as Hopper
- F-0047 Reuters: a senior Trump administration official says DeepSeek trained its upcoming model on Blackwell
Referenced by (4)
- fact F-0047 Reuters: a senior Trump administration official says DeepSeek trained its upcoming model on Blackwell
- claim C-0044 DePue report: six models from four labs were initialized on Blackwell-class hardware (148/152 SMs); earlier models read as Hopper
- claim C-0055 Nvidia (Dec. 2025): "We haven’t seen any substantiation" of DeepSeek using smuggled Blackwells
- theory T-0002 Administration inaction / policy capture