DeepSeek Doubles Its Revenue Run Rate to $1 Billion

DeepSeek is now making money at the rate of $1 billion a year. Founder and CEO Liang Wenfeng gave investors the figure at a recent meeting, The Information reported this week, citing two people with direct knowledge of it. That is more than double the run rate of a few months ago. In late August the company was reported to have taken in about $475 million over the first seven months of 2026, according to PYMNTS, and revenue is up roughly tenfold on 2025.

A price rise did much of the work. In August DeepSeek raised the prices of its models by between 2.3 and 4.5 times, and peak-hour rates for its V4 models went to four times their old level. That sort of rise usually sends business customers looking elsewhere. Liang told investors it had not: demand held up after the change. DeepSeek’s models remain among the cheapest from any major lab even at the new prices, so customers who came for the price had little reason to leave over the increase.

Almost all of the money comes from one place, API access for developers and businesses. DeepSeek’s consumer chatbot is free to use. Liang also told investors that more than 70% of the company’s computing capacity goes to training new models, with under 30% used for inference on the models it has already released. That split is unusual for a lab of its size. The larger Western labs spend a growing share of their compute serving customers, and the balance helps explain how DeepSeek has kept shipping frontier-class open-weight models, most recently DeepSeek-V4.1-Flash.

The number arrives at a useful moment. DeepSeek is finalising its second outside funding round, which aims to raise 50 billion yuan, about $7.5 billion, at a 500 billion yuan valuation, about $75 billion, Investing.com reports. When BitsMinds covered the round at the end of August, signing was expected within days. It has taken longer, and completion is now targeted for the end of October. The round is also the last big step before a planned listing in Shanghai.

Measured against the American labs, $1 billion is modest: OpenAI and Anthropic each report run rates many times larger. The valuation implies a multiple of about 75 times current revenue, rich even by AI standards. But DeepSeek has spent most of its life funded by Liang’s quantitative hedge fund, High-Flyer, and was treated more as a research effort than a business. A lab that can double its revenue within a single quarter while charging customers several times more has given the investors in this round, and those in the listing that follows, a much easier story to buy.