DeepSeek is building its own AI inference chips to cut reliance on Nvidia and Huawei

What happened

Chinese AI startup DeepSeek is developing its own AI chip, according to a Reuters report surfaced via Engadget. Three people familiar with the matter said the chip is designed specifically for inference — the stage of AI computing where a trained model generates responses for users — rather than for training new models.

The push could reduce DeepSeek's reliance on both Nvidia and Huawei, the two chip suppliers it has depended on to train and run its globally popular models. DeepSeek has also been increasing its hiring of engineers to support the effort.

This is a notable strategic shift for a company that has long been known for emphasizing AI model breakthroughs rather than commercializing its technology or building hardware. DeepSeek rose to global fame more than a year ago after releasing two highly efficient AI models that went viral worldwide, surprising many in Silicon Valley and Washington.

Why it matters

DeepSeek's expansion into semiconductor development carries weight on several levels.

First, it signals that China's AI champion is serious about controlling its full hardware stack, not just model software. This aligns with a global trend set by firms like OpenAI and Anthropic, which are also reportedly exploring custom silicon. For DeepSeek, the motivation is sharpened by U.S. export controls that restrict Chinese companies from accessing Nvidia's most advanced chips.

Second, the move could add pressure on Huawei. Despite U.S. sanctions since 2019, Huawei has captured roughly half of China's estimated $50 billion domestic AI chip market, supplying DeepSeek and other leading players. But Huawei's grip is already weakening as rivals like Alibaba and Baidu develop their own AI chips. A DeepSeek-branded chip would further fragment that market.

Third, the focus on inference rather than training is telling. Inference is where the bulk of operational compute costs live — it's what happens every time a user queries a model. A custom inference chip could meaningfully lower DeepSeek's per-query costs, reinforcing the company's reputation for doing more with less.

Context from earlier in 2026 is relevant: in April, DeepSeek released its V4 model in two versions — DeepSeek-V4-Pro and DeepSeek-V4-Flash — and adapted it to run on Huawei's Ascend chips. That adaptation was itself seen as a breakthrough for China's AI industry, marking a departure from DeepSeek's past reliance on Nvidia silicon. Now the company appears to be taking the next logical step: designing its own hardware.

Public reaction

No strong public signal was available from Reddit or other discussion forums at the time of this report. The story is developing, and community reaction will likely depend on whether DeepSeek discloses chip specifications, manufacturing partners, or timelines.

What to watch

• Manufacturing partner: DeepSeek will need a foundry to fabricate its chip. Whether it turns to SMIC, Huawei's HiSilicon, or another partner will reveal a lot about the chip's performance ceiling.
• Timeline to silicon: The sources did not indicate when a chip might be ready. Chip design cycles typically run 18–24 months from specification to tape-out.
• Huawei's response: If DeepSeek shifts inference workloads off Ascend, Huawei loses a marquee customer and reference account.
• U.S. policy reaction: A Chinese AI champion building custom silicon could draw further scrutiny from Washington, potentially tightening export controls on semiconductor manufacturing equipment.

Sources

• Reuters: DeepSeek is developing its own AI chips — Engadget
• China's DeepSeek developing its own AI chip, increase hiring of engineers — The Telegraph Online via Dailyhunt
• DeepSeek's Bold Move: Developing In-House AI Chips — Devdiscourse
• China's DeepSeek says new V4 AI model can run on Huawei chips — Asia Financial