China's DeepSeek Developing Its Own AI Chip

Chinese AI startup DeepSeek is developing its own AI chip, according to a Reuters report citing three people familiar with the effort. The chip is reportedly designed for inference, the stage where a trained model generates outputs for users, rather than for training new models. The project remains at an early stage, with DeepSeek said to be in discussions with chip-design, foundry, and memory partners rather than committed to a finalized product.

The move comes as U.S. export controls continue to bar Chinese companies from buying Nvidia's most advanced AI accelerators, while Beijing has been pressing its domestic tech champions to build homegrown alternatives. DeepSeek's push could also reduce its reliance on Huawei, China's other major domestic chip supplier, suggesting the country's AI leaders want supply-chain options beyond a single vendor. The effort echoes a broader pattern among leading AI labs: OpenAI unveiled its first custom inference chip, codenamed Jalapeno, built with Broadcom, in June 2026, and Anthropic has reportedly been weighing a similar in-house chip program of its own.

For NVDA, the development adds to a growing list of China-specific pressures. Beijing has already certified nine domestically developed AI processors for state procurement as of May 2026, and Chinese buyers have been shifting toward Huawei's Ascend chips as export restrictions and self-sufficiency mandates take hold. A homegrown chip effort from DeepSeek, even one still limited to early partner discussions and focused on inference rather than training, signals that even China's most prominent AI lab sees strategic value in reducing dependence on foreign silicon. Investors should watch whether DeepSeek's chip moves beyond discussions into an actual shipping product, and how much further it could narrow Nvidia's addressable market in China over time.