DeepSeek V4: China's Trillion-Parameter AI Model Ditches Nvidia for Huawei Chips

DeepSeek V4 Confirmed for Late April Launch

DeepSeek founder Liang Wenfeng has confirmed that the company's next-generation flagship model, DeepSeek V4, will officially launch by the end of April 2026. The announcement follows months of delays and intensifying speculation about the model's release timeline, which had slipped from earlier targets of February and March.

A Trillion Parameters on Chinese Silicon

The V4 model features a trillion-parameter Mixture-of-Experts architecture with a context window reaching up to one million tokens. Despite its massive scale, only roughly thirty-seven billion parameters activate per inference pass, a design intended to deliver frontier-level reasoning at a fraction of the computational cost of dense models. What makes V4 truly significant is its deep compatibility with domestic Chinese chips, specifically Huawei's Ascend 950PR processor, which delivers up to one petaflop of FP8 compute performance.

Breaking Free From Nvidia

DeepSeek has spent months collaborating with Huawei and chip designer Cambricon Technologies to rewrite portions of V4's foundational code for Huawei's Ascend architecture. In a pointed strategic move, DeepSeek withheld early access to V4 from US chipmakers including Nvidia, granting optimisation privileges exclusively to domestic suppliers. If successful, this would represent the first frontier AI model capable of operating entirely without Nvidia hardware.

Chinese Tech Giants Prepare for Deployment

In anticipation of the launch, major Chinese technology companies including Alibaba, ByteDance, and Tencent have placed bulk orders for hundreds of thousands of Huawei's next-generation AI chips. The surge in demand has driven chip prices up by roughly twenty percent in recent weeks. Huawei plans to produce roughly six hundred thousand Ascend chips in 2026, double its 2025 output, as China accelerates its push toward AI self-sufficiency.