China’s leading AI startup DeepSeek has withheld its latest advanced AI model from US chipmakers including Nvidia, citing national security concerns amid escalating US-China tech tensions.
Strategic Withholding Amid Chip Wars
DeepSeek’s decision prevents American semiconductor giants from optimizing their hardware—like Nvidia’s H100 GPUs—for the new model, potentially slowing US AI benchmarking and deployment speeds. Sources reveal DeepSeek notified partners that export controls and supply chain risks prompted the move, aligning with Beijing’s push for AI self-reliance after US restrictions on high-end chips since 2022.
The model reportedly rivals GPT-4 performance on key benchmarks while running efficiently on domestic hardware like Huawei Ascend chips.
DeepSeek’s Rapid Rise Challenges West
Founded in 2023, DeepSeek disrupted global AI rankings with R1 (December 2025), matching OpenAI’s o1 at 1/50th cost using just 2,000 Nvidia H800s versus millions for competitors. Its open-weight strategy fueled adoption, but this pivot signals hardening lines—limiting US firms’ access to training data and inference optimizations.
China’s “civil-military fusion” policy likely influenced the block, protecting dual-use tech amid Trump’s 100% semiconductor tariffs.
Global AI Race Implications
Nvidia faces revenue hits as DeepSeek shifts to SMIC/Huawei ecosystems; US developers lose cutting-edge Chinese benchmarks. This mirrors Huawei’s 5G playbook, pressuring Biden-era export controls.
DeepSeek’s move accelerates China’s “indigenous innovation” drive, potentially leapfrogging sanctioned hardware via efficient algorithms.
