DeepSeek Reportedly Reaches $1 Billion Yearly Revenue Pace After API Price Hikes

DeepSeek’s reported revenue pace has passed $1 billion a year, more than double its level a few months ago. The increase followed steep price rises for access to some of its AI models, while demand reportedly remained strong. It gives the company a much larger commercial figure to show investors, but not $1 billion in revenue already earned.

A pace, not a year of sales

CEO Liang Wenfeng reportedly shared the latest figure at a recent investor meeting. The $1 billion number is an annualized run rate: it expresses the company’s current revenue pace as a yearly figure. It is not a tally of sales collected over the past 12 months.

A few months earlier, DeepSeek’s reported annualized pace was below $500 million. Crossing $1 billion therefore marks more than a doubling of that measure, not necessarily a doubling of customer use. The distinction matters here because the company also changed what it charges for some model access.

The price decision behind the rise

DeepSeek raised prices for access to some models through its API, the interface developers use to put the models into their own software. Those prices rose to between 2.3 and 4.5 times their previous levels last month, according to The News International. The increases reportedly contributed to the revenue jump.

Demand reportedly stayed strong despite the higher prices. The News International goes further, saying the customer base remained stable and the increases did not cause customer churn. That is a narrower claim than saying usage grew: customers can stay while changing how much they buy. The accounts do not quantify that usage or separate the revenue gained from higher prices from any revenue gained through greater use.

Revenue rises; training keeps priority

The reported sales pace has not shifted DeepSeek’s stated priority. Liang has reportedly told investors that making money remains secondary to research and development. More than 70% of the company’s computing capacity reportedly goes to training new models; less than 30% goes to inference, the work of running existing models to produce answers.

That split puts the pricing result in perspective. DeepSeek is reportedly earning at a faster pace from its services while reserving most of its computing capacity for models it has yet to finish training. The figures describe its current allocation, not how well its smaller inference share handles customer demand or whether that balance will change.