DeepSeek Moves to Build In-House AI Inference Chip Amid U.S. Export Controls

DeepSeek, the Hangzhou-based AI startup behind models R1 and V4, has begun designing a custom chip dedicated to inference workloads. The project, launched about a year ago, seeks to replace Nvidia and Huawei silicon powering its services.

DeepSeek Develops Custom AI Chip as Nvidia (NVDA) Stock Falls 2%

U.S. export bans have blocked Chinese access to Nvidia’s latest GPUs, including the H800 chip used for training R1. The restriction has elevated Huawei’s Ascend processors to roughly half of China’s $50 billion AI-chip market, though Alibaba and Baidu are introducing competing designs.

Reuters

Founder Liang Wenfeng leads DeepSeek’s effort, which is courting chip-design firms, foundries and memory suppliers. The company is raising $7 billion in an external round that would value it at $52-$59 billion. Huawei, Alibaba, Baidu and SMIC are mentioned as potential collaborators.

Channel News Asia - CNA

The planned silicon targets inference—the stage where a trained model generates responses—rather than model training. DeepSeek has hired chip-design engineers and is in early talks with partners; no prototype has been disclosed.

China's DeepSeek Begins Developing Proprietary Inference AI Chip, Aiming to Reduce Reliance on Nvidia and Huawei — BigGo Finance

• China AI-chip market: ? $50 billion.

Channel News Asia - CNA

• Huawei’s share: ~ 50 %.

Reuters

• DeepSeek’s target valuation: $52-$59 billion.

China's DeepSeek Begins Developing Proprietary Inference AI Chip, Aiming to Reduce Reliance on Nvidia and Huawei — BigGo Finance

• Nvidia’s pre-market share in China: effectively zero, according to analysts.

Benzinga

Huawei confirmed that its Ascend processors were used in part of the training of DeepSeek’s V4-Flash model and reported a surge in orders for Ascend 950 chips after the V4 launch. DeepSeek did not comment.

China's DeepSeek Begins Developing Proprietary Inference AI Chip, Aiming to Reduce Reliance on Nvidia and Huawei — BigGo Finance

Analyst Richard Windsor warned that DeepSeek faces near-zero chances of selling silicon outside China without access to leading-edge manufacturing, and noted that Nvidia’s market share in China has collapsed to effectively zero, highlighting the difficulty of competing internationally.

Reuters

All sources agree the chip effort is in its infancy, but DeepSeek has not publicly verified milestones. Design specifications, expected tape-out dates and the chosen foundry remain undisclosed.

DeepSeek develops proprietary AI inference chip

• “Nvidia is at zero in China and staying there.” — Richard Windsor, Analyst

Reuters

• “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,” — Richard Windsor

Reuters

• “DeepSeek founder Liang Wenfeng said in a rare 2024 interview with a Chinese media outlet that chip export controls were a challenge for the company.” — Liang Wenfeng, Founder, DeepSeek

Reuters

• “The chip is designed ?for inference — the stage of AI computing in which a trained model generates responses for users — rather than for training new models, the sources said.” — Reuters

Reuters

The $7 billion funding round is slated to close later in 2026, providing capital for chip design and a tape-out. If production materializes, DeepSeek could lower inference costs, pressure Huawei’s market share and add further strain to Nvidia’s position in China.

DeepSeek develops proprietary AI inference chip