DeepSeek V4 Integrates Huawei Ascend Chips Amidst Escalating US IP Theft Accusations

DeepSeek, the Chinese AI startup that disrupted global markets last year, released preview versions of its V4 model on Saturday, marking a significant step in the US-China AI rivalry. The new open-source models—V4 Pro and V4 Flash—integrate Huawei's Ascend chips, reducing dependency on US suppliers like Nvidia. DeepSeek claims the V4 Pro Max version delivers 'superior performance' on reasoning benchmarks compared to OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro, while falling 'marginally' short of GPT-5.4 and Gemini 3.1-Pro. For agentic capabilities, the Pro version reportedly outperforms Anthropic's Claude Sonnet 4.5 and approaches Claude Opus 4.5 levels. Both models feature a 1 million token context window, up from 128K in V3, and run more efficiently.

Industry reactions are mixed. Lian Jye Su of Omdia notes V4 appears 'very competitive against US rivals,' while Marina Zhang of UTS calls it a 'pivotal milestone for China's AI industry.' However, Morningstar's Ivan Su cautions it's 'not as big a breakthrough as R1,' emphasizing the need for independent evaluations. The launch follows accusations from Anthropic of 'industrial-scale campaigns' to illicitly extract capabilities from Claude, and OpenAI's similar claims. DeepSeek's free chatbot and open-source approach continue to attract users, especially in developing nations where Huawei phones are prevalent, as the US-China tech decoupling intensifies.