Breaking

DeepSeek Said to Be Building Its Own AI Inference Chip

July 7, 2026 at 09:10 EDT

Infra & Chips
Foundation Models

Chinese AI startup DeepSeek is developing its own AI inference chip, according to a Reuters report published on July 7, 2026. Citing three people familiar with the matter, the report says the chip is aimed at inference rather than training and that development remains at an early stage. The Hangzhou-based company is in talks with chip design firms, foundries and memory suppliers, and has quietly ramped up hiring of chip design engineers in recent months without posting the roles on public platforms. The effort is believed to have started about a year ago. DeepSeek declined to comment.

July 7, 2026 · DeepSeek

DeepSeek Sets Out to Build Its Own AI Inference Chip

The Hangzhou firm is in early talks with chip designers, foundries and memory suppliers to build an in-house inference chip — aiming to cut its dependence on Nvidia and Huawei silicon.

~1 yr

Since the chip effort quietly began

~50%

Huawei's share of China's domestic AI chip market

-1.6%

Nvidia shares in pre-market after the report

What the chip targets

TRAINING GPU

Costly · power-hungry

DEEPSEEK INFERENCE CHIP

Cheaper · efficient

The chip is built for inference only — generating responses, not training models — to attack the operating cost of DeepSeek's own services.

The path to silicon

In talks with designers, foundries & memory suppliers

→

Hiring chip design engineers

→

Mass production uncertain

THE UPSIDE

DeepSeek's low-cost, high-performing models are proven. In-house inference silicon could cut costs and fit Beijing's homegrown-tech push — echoing OpenAI and Anthropic's custom-chip moves.

THE HURDLES

Advanced process nodes and HBM fall under U.S. export controls. No product name, specs or timing disclosed — and interconnect bandwidth, stability and software maturity remain open questions.

Bottom line: The effort is still early stage, and Nvidia's dominance outside China keeps the near-term market impact limited — but a working chip would mark another step toward domestic silicon.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.