A $1B annualized revenue figure now sits on DeepSeek’s books, according to The Information, a pace more than twice what the outlet reported only months ago. Chief executive Liang Wenfeng gave the number at an investor meeting.
Pricing did a lot of the work: the company lifted what it charges for model access by roughly 2.3 to 4.5 times, and buyers kept coming.
A run rate is not a year of bookings. It is shorthand for the revenue a company would book over 12 months if current levels hold. For a lab that spent last year arguing frontier AI did not require frontier budgets, the trajectory is the point.
Training, not serving, still takes the larger share of DeepSeek’s machines. Liang put it at more than 70% for new models, against less than 30% for inference.
On capital, a second round is in the works at roughly 50 billion yuan, about $7.5B, at a reported valuation near 500 billion yuan, with a target of closing by the end of October. Word of the hiring came from Reuters earlier this month; a Chinese brokerage has been lined up to work on a stock market debut, and the name attached to that preparation is CITIC Securities. The venue under discussion is Shanghai’s STAR Market.
The next test is commercial. DeepSeek released V4.1-Flash this month, promising faster inference and support for scaling to larger models, and a completed round would put it among China’s biggest new AI businesses.
