DeepSeek Releases V4-Pro (1.6T Parameters) Built on Huawei Ascend 950 Chips, Challenging US AI Dominance

Chinese AI lab DeepSeek released a preview of V4-Pro (1.6 trillion parameters) and V4-Flash (284 billion parameters) on April 24, 2026, both under the MIT open-source license with 1-million-token context windows. Critically, both models were trained entirely on Huawei Ascend 950 chips via domestic 'Supernode' cluster architecture — demonstrating China's capacity to develop frontier AI without Nvidia H100/H800 GPUs subject to US export controls. China's state AI fund (Big Fund) announced an up-to-$4 billion investment round backing DeepSeek. SMIC and Hua Hong Semiconductor shares surged 9% and 15% respectively on the news. The release prompted the Council on Foreign Relations to describe V4 as signaling 'a new phase in the US-China AI rivalry,' with independent benchmarks showing V4-Pro matching or exceeding GPT-4o on major tasks.