DeepSeek V4 Withholding from NVIDIA/AMD (Feb-Mar 2026)
eventAI & Compute · Nations & Geopolitics
Both sides of the US-China AI 'rivalry' started hiding their best models from the other side at exactly the same moment.
Who they are
DeepSeek's decision in early 2026 to withhold its flagship V4 model from US chipmakers NVIDIA and AMD.
What they do
The engine treats it as a mirror-image confirmation of the same gatekeeping seen on the American side.
How it works
Per Reuters reporting, DeepSeek broke the normal practice of sharing new models early with NVIDIA and AMD to optimize them, and instead gave weeks of exclusive early access to China's Huawei and Cambricon chips. At launch, V4 ran best on Huawei hardware and worse on American silicon.
Why it matters
The engine reads this as evidence of a 'false rivalry': both supposedly opposed sides adopted the identical withholding strategy at the same time, pointing to a shared pattern rather than genuine opposition.
The engine's record — word for word
Feb-Mar 2026 Reuters reporting: DeepSeek intentionally withheld early access to flagship V4 model from NVIDIA and AMD, breaking industry norm of pre-release sharing for CUDA/ROCm optimization. Granted exclusive weeks-long early access to Huawei (Ascend architecture) and Cambricon. Mirrors Glasswing flawlessly — adversary-side confirmation of the gating pattern. Per Jiang Framework #4 false dialectic: both sides of the US-China AI 'rivalry' adopted identical withholding postures at the exact same historical moment. Launch-day result: V4 optimal on Huawei hardware, degraded on American silicon — localized optimization race.
Follow the trail
Walk this on the live map →