DeepSeek released its V4 model (V4-Pro at 1.6T parameters, V4-Flash at 284B) in April 2026, with Huawei confirming its full Ascend SuperPoD lineup supports the V4 series, a marked departure from DeepSeek's prior reliance on Nvidia hardware. The company claims "drastically reduced compute and memory costs," a 1-million-token context window on par with Gemini, and benchmark performance in "world knowledge" trailing only the latest Gemini model. Structurally, this validates a functioning Chinese AI hardware-software stack and materially weakens the premise that US export controls on advanced chips can constrain Chinese frontier model development. For financial institutions, this raises the credibility of China-domestic AI infrastructure plays (Huawei Ascend supply chain, Chinese hyperscalers) while incrementally pressuring the export-control-as-moat thesis that underpins some US AI equity valuations.
393 other tracked developments touching the same claims.