DeepSeek's annualised revenue run rate hits $1 billion

Chinese AI startup DeepSeek's annualised revenue run rate has reached $1 billion, more than double the level of a few months ago, The Information reported on Thursday, citing two people with direct knowledge of the matter. Reuters said it could not immediately verify the report, and DeepSeek did not respond to a request for comment.

A run rate shows how much a company would earn over a year if it kept performing at its current pace, so the figure reflects momentum rather than money already collected. Founder and CEO Liang Wenfeng shared the number at a recent meeting with investors, according to The Information.

What fuelled the growth

Part of the increase came from higher prices: DeepSeek raised its model pricing by 2.3 to 4.5 times, depending on the model, and Liang told investors that customers stayed, The Information reported. In August the company warned developers that a sharp rise in API prices was coming.

Earlier reporting by The Information put DeepSeek's revenue for the first seven months of 2026 at about 475 million yuan ($70.7 million), roughly ten times its revenue for all of 2025. Most of that money comes from developers who pay to use DeepSeek's models through its API, while the chatbot itself is free.

Funding round and listing plans

DeepSeek is finalising a second funding round that targets 50 billion yuan ($7.45 billion) at a valuation of 500 billion yuan by the end of October, The Information said, and is preparing for a potential listing on the Shanghai Stock Exchange. Bloomberg has put the round's valuation at about 500 billion yuan, or $74 billion.

Reuters reported this month that DeepSeek had hired CITIC Securities to prepare for a possible IPO on Shanghai's STAR Market, with the timing, valuation and size of the fundraising still undecided. The startup's first outside round, of roughly $7 billion, closed in June.

Where the resources go

Liang told investors that DeepSeek still devotes the bulk of its resources to new models: more than 70% of its computing capacity goes to training, while less than 30% is used for inference, or running existing models. Earlier in September the company released DeepSeek-V4.1-Flash, which it says is built for greater capability, faster inference, higher throughput and scaling to larger models.