DeepSeek: $7.5 billion in funding, $1 billion in revenue and an IPO plan

DeepSeek is no longer a research curiosity. The portal IT之家 reports, citing The Information and two people familiar with the matter, that the company's annual revenue has passed $1 billion and that its second funding round has been set at $7.5 billion. A few months earlier, annual revenue was still below $500 million.

Two things drive the growth: higher API prices and rising demand for the model. According to The Information's sources, DeepSeek is running a second round and preparing to list in Shanghai. It intends to close the round in October and raise the equivalent of 50 billion yuan, or about $7.5 billion, at a target valuation of up to 500 billion yuan. Liang Wenfeng is said to have told investors that the price increase did not drive customers away.

How much of it goes to training

The split of computing power is telling. DeepSeek puts more than 70 percent of its capacity into training models and leaves just under 30 percent for serving inference on models already released. That is an aggressive ratio: revenue today comes from the inference layer, while the investment goes into the next models. Liang Wenfeng is also said to have told investors that internal tests show the company's lightweight model runs stably on consumer-grade graphics cards and will handle most of users' everyday tasks.

Cash for the listing and the first round

A stock exchange procedure is running in parallel. Reuters reported earlier that DeepSeek hired CITIC Securities to prepare a listing on Shanghai's science and technology board. The company plans to file an application this year and debut the following year. It wants to raise funds to expand its computing infrastructure, work on models and its own chips, and pay key specialists.

The scale of the first round is known too. In June 2026 DeepSeek closed its first external financing: it raised about $7.4 billion at a post-money valuation above $50 billion. The largest single investor was the founder himself, Liang Wenfeng, with about 20 billion yuan. Tencent is said to have added about 10 billion yuan, and a company linked to battery maker CATL about 5 billion. JD, NetEase, IDG Capital and a state AI fund also took part. The new round has a five-year lock-up period and gives outside investors no voting rights. Even so, as IT之家 reports, demand remains high.

A more cautious side of the boom is visible in the background. The German site heise.de describes an analysis by economist Stijn Van Nieuwerburgh of Columbia Business School. It finds that AI infrastructure investment planned in the USA through 2032 corresponds to 3.63 percent of GDP, nearly twice as much as the record of the railway boom, and that to refinance that spending, AI revenue would have to grow by 80 percent a year.