DeepSeek is making real money now, according to a new report — and it is raising prices to make even more. The Information reported on Thursday that the Chinese AI startup's annualized revenue run rate has crossed $1 billion, more than double where it stood a few months ago, citing two people with direct knowledge of the matter. Reuters, which syndicated the report, said it could not independently verify the figure, and DeepSeek could not be reached for comment.

A run rate is not the same as revenue booked: it extrapolates what the company would earn in a year if it kept performing at its current pace. Still, as a milestone, the reported figure marks a turning point for the startup that shocked the tech world last year when its models challenged US assumptions about China's AI capabilities. DeepSeek had been famous for undercutting Western labs on price; now, per the report, it is flexing pricing power.

Price hikes drove the jump#

According to the report, the revenue surge was partly driven by raising model pricing by 2.3 to 4.5 times. That fits with what developers have been experiencing firsthand: earlier this month, DeepSeek released DeepSeek-V4.1-Flash, pitched for greater capability and faster inference, and moved API customers through a rapid price and model transition — one that sparked public backlash from production teams with tuned pipelines when notice was short.

The timing is no accident. Demand for inference across the Chinese market is rising fast, and DeepSeek — which gives its models away as open weights while charging for its hosted API — sits at the center of it. If the reported figures hold, the company has proven something no other Chinese lab has yet: that frontier models can be a billion-dollar business even under US chip export controls.

The DeepSeek Android app interface, the company's consumer chatbot
DeepSeek Android app screenshot, CC0, via Wikimedia Commons.

A fundraise — and a STAR Market listing#

The report adds detail to DeepSeek's capital plans. CEO Liang Wenfeng reportedly shared the revenue figure at a recent meeting with investors, as the company pushes ahead with a second funding round targeting 50 billion yuan ($7.45 billion) at a valuation of 500 billion yuan, to close by the end of October. DeepSeek is also preparing for a potential listing on the Shanghai Stock Exchange's STAR Market, and has hired CITIC Securities to prepare for it, Reuters reported earlier this month — though timing, valuation, and fundraising size remain undecided.

It would be DeepSeek's first major outside capital. The company has been funded almost entirely by High-Flyer, the quantitative hedge fund Liang founded in 2016 that spun DeepSeek out in 2023. A successful raise at that valuation — roughly $75 billion at current exchange rates — would make DeepSeek one of the most valuable AI startups on earth, within striking distance of the Western frontier labs.

Diagram of the DeepSeek-V4 series model architecture
DeepSeek-V4 series architecture diagram, MIT license, via Wikimedia Commons.

Most compute still goes to training#

The most revealing detail in the report is where the money is not going. Liang told investors that DeepSeek allocates more than 70% of its computing capacity to training new models, reserving less than 30% for inference — actually running the models that generate revenue.

That ratio says a lot about the company's priorities. A billion-dollar run rate built on under a third of compute means inference margins are strong; but the bet is clearly on the next generation of models, not on squeezing the current one. It is the classic frontier-lab playbook — research first, monetization second — running at a scale that even many Western labs would envy.

What to watch#

  • Independent confirmation. DeepSeek discloses nothing as a private company; the $1 billion figure rests on two anonymous sources via The Information. Watch whether the fundraising process produces harder numbers.
  • The funding round. A 50 billion yuan raise closing by end-October would be the first external validation of DeepSeek's valuation. Any delay or repricing would say as much as a success.
  • STAR Market timing. A Shanghai listing would give Chinese public-market investors their first direct exposure to a frontier AI lab — and subject DeepSeek to its first real disclosure regime.
  • Price elasticity. DeepSeek built its name on cheap inference; the 2.3–-4.5x hikes test whether developers stay when the bargain ends.

Run rates get revised, fundraisers slip, and private-company leaks serve the leaker. But the direction of travel is hard to miss: the lab that made AI cheap is now demonstrating that AI can be lucrative — and investors are lining up to price it.