Chinese AI startup DeepSeek is working on its own AI chip focused mainly on inference tasks, aiming to reduce dependence on existing Nvidia and Huawei chips, according to multiple reports from July 2026 [1, 2, 3]. The project started about a year ago and is still in early stages, involving talks with chip design firms, foundries, and memory manufacturers [1, 2, 3]. DeepSeek has quietly increased recruitment of chip design engineers without publicly posting job openings [1, 2].

DeepSeek rose to international prominence in July 2025 after releasing two highly efficient AI models, surprising Silicon Valley and Washington observers [1, 3]. Initially, it trained its R1 inference model on Nvidia's H800 chips but later switched to Huawei Ascend chips for newer models such as the V4 released in April 2026 [2, 4, 5].

The decision to develop its own inference chip reflects a strategic shift from focusing solely on AI software breakthroughs toward deeper hardware and supply chain control [1, 2]. The startup aims to prioritize cost, energy efficiency, and latency at scale, which are critical for inference workloads, as opposed to the more demanding training phase that requires advanced chip technology [2, 6, 5].

Huawei currently holds about half of China’s $50 billion domestic AI chip market and has supplied DeepSeek so far [1, 2, 7]. However, Huawei's dominance faces growing challenges from Alibaba and Baidu, who are developing their own AI chips, while U.S. export controls limit Nvidia's advanced AI chips from entering China, helping Huawei's market share but also pushing Chinese firms toward domestic alternatives [1, 2, 4]. Analyst Richard Windsor noted, "Nvidia is at zero in China and staying there. DeepSeek has almost no chance of selling silicon outside of China unless it gets access to leading edge manufacturing," underscoring the limits imposed by export restrictions [3].

DeepSeek has kept details of its chip project low-profile and declined requests for comment [1, 2]. Its move into chip design could increase competition and pressure on Huawei in China’s AI chip sector [1, 3]. Meanwhile, in June 2026, OpenAI unveiled its own custom inference chip, Jalapeño, co-developed with Broadcom, and other AI firms like Anthropic are also exploring custom chip development [1, 2, 6].

DeepSeek is expected to continue advancing its chip development throughout 2026 as it aims to secure more hardware independence amid increasing geopolitical and market pressures [1, 3].