DeepSeek Reportedly Hits $1B Annualized Revenue
On a recent Sunday, Liang Wenfeng disclosed the figure to the investors himself. The Information reported what he shared: “DeepSeek’s annualized revenue run rate has hit $1 billion, more than double the rate of a few months ago, buoyed by a recent price hike as well as continued popularity of its models.”
A few months earlier that same run rate had still been under $500 million. The billion was not cash already collected over a full year. It was the current pace of earning, projected across twelve months—what the company would take in if the rate held.
The climb rested on customers who kept using the models and a recent lift in what those customers paid. In mid-August DeepSeek had rewritten the fees for access to its models, and the higher charges were already inside the figure on the table. Peak hours and off-peak hours now set the rates. The company had moved API access—the paid link that lets outside developers call its models from their own software—onto a two-tier schedule that charged more when traffic was heavy and less when it was light. Developers paid according to how much text the models read and wrote, and under the new structure those fees climbed by factors of 2.3 to 4.5 times what they had been. The multiples were steep. They applied across the model lineup.
Liang told the investors the customer base did not shrink. Paying users stayed at the new levels. The systems remained among the cheapest of the major large language models—the software that turns an ordinary written prompt into generated text, code, or answers—and traffic kept arriving after the hike. Dana Whitfield stated the demand signal in one sentence: “Doubling revenue after raising API fees that sharply shows real demand for their models despite the higher costs.”
Nearly all of what the company collected moved through those paid connections. The free chatbot app generates no revenue. Outside software teams bought the right to call DeepSeek’s models from their own applications and paid by the volume of text the systems read and wrote. That single stream funded the company. The chatbot remained open to any user at no charge and returned nothing.
The Information had previously measured the cash that arrived in the first seven months of 2026. The total came to about 475 million yuan, or roughly $70.7 million—already about ten times DeepSeek’s revenue for all of 2025. Across the same stretch the gross margin on the API work reached 82.9 percent. Gross margin was the portion of each payment left after the direct cost of delivering the request. More than four-fifths of every dollar that came through the paid interface stayed with DeepSeek once those costs were paid.
Liang told the investors where the processors actually went. More than 70 percent of DeepSeek’s computing capacity ran to training—the long work of teaching new models on large stores of data until they could produce coherent text and code. Under 30 percent stayed with inference, the everyday task of answering the prompts that arrived through the free app and the paid interface. Even as the commercial side expanded, the company remained mostly a training lab.
DeepSeek had been founded in July 2023 by Liang in Hangzhou. It was owned and funded by High-Flyer, the Chinese hedge fund he had helped build years earlier. From the beginning the firm released open-weights large language models, systems whose finished parameters anyone could download and run while the training data stayed closed. In January 2025 it launched its chatbot together with the DeepSeek-R1 model, and the release broke through worldwide. That path—capable models shared openly at low cost—was what later made the paid traffic possible.
The first outside funding round closed in June 2026. It brought in roughly $7 billion, and Liang retained a large share through the structure of the deal. The capital and the machines stayed pointed at the next generation of models. In July 2026 DeepSeek paused its second fundraising effort after remarks by Liang leaked and circulated on Chinese social media. In August the company restarted. Investors were now being asked for roughly $7.45 billion to $7.5 billion, about 50 billion yuan. The valuation ran from roughly $68.8 billion to $75 billion; Bloomberg placed the figure near 500 billion yuan. DeepSeek aimed to close the financing by the end of October 2026. None of the money had locked yet.
Reuters reported that DeepSeek had selected CITIC Securities as lead underwriter for a planned listing on the Shanghai Stock Exchange’s STAR Market, the board built for technology companies. An initial public offering is the process in which a privately held firm sells shares to institutional investors and usually also to retail investors; one or more investment banks underwrite the sale, arrange the listing, help set the share price, and open a public market where the stock can trade freely afterward. Capital from newly issued shares goes to the company. The lead underwriter manages the largest share of that work and coordinates any syndicate. CITIC Securities was the bank named to lead it. Shanghai’s STAR Market was the board chosen for the debut.
