News · 2026-07-12
DeepSeek is designing its own AI chip -- and raising outside money for the first time
DeepSeek, the Chinese AI startup, is developing its own chip designed to run trained models for users -- an inference chip -- and is simultaneously raising its first-ever round of outside capital, roughly $7 billion at a $52-59 billion valuation. Reuters reported the chip effort citing three people familiar with the matter, marking DeepSeek's move from a pure research lab toward a full-stack AI company that controls its own hardware.
Key facts
- DeepSeek's chip targets inference (running trained models), the fastest-growing segment of AI compute demand, not training.
- The effort began about a year ago and is at an early stage -- discussions with chip-design, foundry, and memory companies, plus quiet hiring of chip-design engineers.
- DeepSeek is raising ~$7 billion at a $52-59 billion valuation, reversing years of refusing outside investment.
- Reported by Reuters, July 7, 2026, from three sources; corroborated by multiple wires. Reuters.
The hook is the word "inference." There are two phases in an AI model's life: training, the enormously expensive one-time process of building the model, and inference, running it to answer every user query afterward. Inference is where the ongoing costs -- and the ongoing chip demand -- pile up, because it happens billions of times. By aiming its chip at inference rather than training, DeepSeek is targeting the part of the market that is growing fastest and where a cheaper domestic chip could make the biggest dent.
Background for the non-expert: most of the world's AI runs on Nvidia chips, and US export controls restrict which Nvidia hardware China can buy. Inside China, Huawei supplies about half of the roughly $50 billion domestic AI-chip market, but its grip is loosening as Alibaba and Baidu build their own silicon. DeepSeek joining that race is part of a clear trend -- AI developers everywhere are trying to reduce dependence on a single chip supplier. It echoes OpenAI unveiling its own custom inference chip last month and reports that Anthropic has weighed building its own.
Think of it like a popular restaurant chain that has always bought its ovens from one manufacturer, now designing its own ovens -- not the industrial ones used to develop new recipes, but the everyday ones that cook the dishes customers actually order. The recipe development stays hard; the day-to-day cooking is where owning your equipment pays off at scale.
The hard part is manufacturing, and this is the honest caveat. Designing a chip is one thing; fabricating a competitive one is another, and US rules bar Chinese designers from the most advanced overseas foundries while separate controls have cut China's access to the high-bandwidth memory that inference chips need. Radio Free Mobile analyst Richard Windsor was blunt: "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." So the realistic frame is a domestic play -- a chip to serve DeepSeek's own models inside China, not a global Nvidia competitor.
Why it matters: this is the hardware front of the day's larger story. If DeepSeek can ship a cheap inference chip inside China, the cost gap between local/open models and frontier/proprietary ones narrows further -- exactly the commodification dynamic George Hotz and Satya Nadella were describing the same week. And the $7 billion raise at a $52-59 billion valuation is its own signal: a lab once famous for rejecting outside money is now capitalizing to go vertical -- models, and the chips to run them. It sits alongside the day's other China-hardware stories (a domestic 2D-semiconductor pilot line) as evidence that the AI supply chain is fracturing along geopolitical lines, with each major player racing toward hardware independence.
Key questions
What kind of chip is DeepSeek building?
Why does DeepSeek want its own chip?
Is DeepSeek raising money?
Cite this
APA
Ground Truth. (2026, July 12). DeepSeek is designing its own AI chip -- and raising outside money for the first time. Ground Truth. https://groundtruth.day/news/deepseek-building-its-own-inference-chip.html
BibTeX
@misc{groundtruth:deepseek-building-its-own-inference-chip,
title = {DeepSeek is designing its own AI chip -- and raising outside money for the first time},
author = {{Ground Truth}},
year = {2026},
month = {jul},
url = {https://groundtruth.day/news/deepseek-building-its-own-inference-chip.html}
}
Comments are replies to this story on Bluesky — reply with any Bluesky account to join in.