DeepSeek, the Hangzhou-based AI startup, is developing its own chip for inference workloads, three people familiar with the plan told Reuters. The news sent Nvidia shares down 1.6 per cent in premarket trading, though London-based analyst Richard Windsor of Radio Free Mobile noted the US chipmaker “is at zero in China and staying there,” limiting the direct commercial impact. The chip is designed for the stage of AI computing where a trained model generates responses, rather than for training new models, and would give DeepSeek greater control over its hardware stack.

Viewed from Washington, the effort carries a clear strategic dimension. US export controls bar Chinese firms from buying Nvidia’s most advanced processors, and Beijing has been pressing its technology champions to build domestic alternatives. DeepSeek founder Liang Wenfeng acknowledged in a rare 2024 interview that chip restrictions posed a challenge. The company previously trained its R1 reasoning model—whose low-cost performance triggered a rout in US tech stocks in January 2025—on Nvidia’s H800, a chip designed for the Chinese market that Washington banned in late 2023. Since then, DeepSeek has leaned increasingly on Huawei’s Ascend chips, with its V4 model optimised for that hardware.

DeepSeek’s move mirrors a broader industry trend. OpenAI last month unveiled its first custom inference chip, developed with Broadcom, while Anthropic is weighing a similar step. In China, Alibaba and Baidu are also developing proprietary AI chips, eroding Huawei’s near-50 per cent share of the domestic AI chip market. For DeepSeek, the in-house chip would reduce dependence on both Nvidia and Huawei, though the project remains at an early stage. The company has been quietly hiring chip-design engineers and has held discussions with external design, foundry and memory partners over the past year.

Significant hurdles remain. Designing a competitive AI chip typically takes years and substantial capital. Manufacturing is constrained by US bans on Chinese access to advanced overseas foundries and high-bandwidth memory, a critical component for inference chips. Meanwhile, US congressional committees are examining ways to limit the use of Chinese AI models by American businesses, citing cybersecurity concerns, according to CNBC. Some US government agencies have already banned DeepSeek, even as tech executives openly promote Chinese models to cut costs. The next factual milestone will be any public confirmation from DeepSeek of a tape-out or manufacturing partnership, though the company has not responded to requests for comment.