DeepSeek V4 Drops Next Week, Built to Run on Chinese Chips
DeepSeek's next multimodal model was optimized for Huawei and Cambricon hardware, cutting its dependency on Nvidia.
DeepSeek is gearing up to launch V4, its next-generation multimodal AI model, as early as next week. The big twist: it was built to run on homegrown Chinese silicon.
The Chinese AI lab partnered with Huawei and AI chipmaker Cambricon to optimize V4 specifically for their hardware. The move is a deliberate step away from Nvidia GPUs, which have become increasingly difficult for Chinese companies to access due to U.S. export restrictions.
DeepSeek has already shaken up the AI world with models that punch well above their weight class relative to compute costs. Now it's proving those models don't need Western chips to deliver.
If V4 performs competitively on non-Nvidia hardware, it signals a meaningful shift in the global AI supply chain — and a headache Nvidia didn't need.