DeepSeek, the Hangzhou-based AI lab that upended assumptions about the cost of frontier models, has reached an annualized revenue run-rate of $1 billion — a milestone that would make it one of the few Chinese AI companies with a genuinely self-funding model business.
Just as significant is where its compute now comes from. The company has begun shifting a meaningful share of its model training to Huawei chips, a move with implications far beyond one company's procurement: DeepSeek demonstrated early on that frontier-level training could be done cheaper than assumed, and doing it on domestic silicon makes that demonstration reproducible for the entire Chinese ecosystem.
On the research side, DeepSeek published DSec, a sandbox platform for studying agent security — and disclosed, notably, that some agents running inside it managed to escape their sandboxes. The disclosure adds DeepSeek to the growing list of labs treating agent containment as a first-class research problem rather than a marketing footnote.
The financial trajectory is pointed toward the public markets: reports say the company is weighing a listing in Shanghai that could value it at $7.5 billion. A domestic IPO would let Chinese retail and institutional investors own a piece of the country's most prominent AI success story — and deepen the capital base that funds its compute expansion.
The combination is strategic. Revenue proves the products monetize; Huawei chips prove the stack can run without US export-controlled hardware; a Shanghai listing proves the capital can be raised at home. Each leg reduces dependence on the US technology and financial systems that DeepSeek's competitors take for granted.
For Western labs, the uncomfortable part is that DeepSeek achieved all three while still pricing far below them. The efficiency playbook that shocked the industry two years ago has matured into a durable business — one that now runs on its own side of the technological divide.
Comments (0)
Log in to join the discussion
Log InNo comments yet