DeepSeek Hits $1B Run Rate, Faces a Regulatory Shadow

DeepSeek CEO Liang Wenfeng told investors this week that the company's annualized revenue run rate has crossed $1 billion, more than double the under-$500 million pace reported just months earlier. The number arrives as DeepSeek finalizes a roughly $7.5 billion funding round ahead of a planned Shanghai Stock Exchange listing. It also arrives while Chinese regulators are actively investigating DeepSeek over a separate matter: whether the company improperly routed sensitive user data to Anthropic's Claude models. Both stories are true at once, and neither cancels the other out.

Table of contents

• The headline number, and what it actually means
• Where the growth came from
• The IPO math
• The compute split
• The open regulatory question
• What determines whether $1 billion holds
• Frequently asked questions

The headline number, and what it actually means

A run rate is not a bank balance. RuntimeWire's explanation is the clearest available: the figure is a snapshot revenue pace projected across a year, not a statement that DeepSeek has already collected $1 billion in sales. It sits alongside a separate, earlier disclosure that DeepSeek generated roughly 475 million yuan, about $70.7 million, in the first seven months of the year. The two numbers describe different things, a recent pace versus money already booked, and are not necessarily inconsistent with each other.

Liang shared the $1 billion figure in an investor meeting, according to Dealroom's reporting, which also puts DeepSeek's gross margin at 82.9% through July. Every dollar of that revenue comes from developers and enterprises using DeepSeek's API, per CryptoBriefing's account. The consumer chatbot that made DeepSeek a household name in early 2025 remains completely free and contributes nothing to the figure directly.

Where the growth came from

The doubling is not simply more customers buying more. PYMNTS reports DeepSeek raised its model prices by 2.3 to 4.5 times last month, and Liang told investors the customer base held steady through the increase. AI Weekly's analysis flags the real question this raises for anyone pricing DeepSeek's coming IPO: revenue doubled partly from one-off price resets, not purely from demand growth, which makes the run rate harder to treat as a stable baseline. If growth mostly reflects durable demand at the new prices, the number supports a higher valuation. If a meaningful share came from the price change itself rather than more usage, the same number tells a shakier story once the reset year rolls off the comparison.

Techstartups' framing softens that concern somewhat: even after the increase, DeepSeek's prices remain among the lowest for major AI models, which is one reason customers may not have walked away. Whether that pricing gap survives the IPO process, when investors typically want margin expansion, is unresolved.

The IPO math

DeepSeek is targeting a raise of 50 billion yuan, about $7.5 billion, at a valuation of 500 billion yuan, roughly $75 billion, according to multiple outlets tracking the round. Eastern Herald reports CITIC Securities has been selected as lead underwriter, with a plan to file the listing application before year-end and complete the Shanghai STAR Market IPO in 2027. RuntimeWire does the simple math worth sitting with: a $75 billion valuation against a $1 billion run rate is roughly 74 times revenue, a comparison that only holds if the current pace is sustained rather than a temporary spike from the price change.

The listing itself carries meaning beyond DeepSeek. Eastern Herald notes China's domestic technology IPO pipeline has been cautious since Beijing imposed broad restrictions on overseas listings in 2021, making a Shanghai-based AI listing something of a political statement about where Chinese tech capital formation happens next, separate from the commercial case for DeepSeek specifically.

The compute split

Whalesbook's reporting adds a detail that says something about DeepSeek's own priorities: the company is directing 70% of its total computing power toward training new models, with only 30% going to inference, the process of actually running the product for paying customers. That is a heavier tilt toward research than a company primarily optimizing for near-term revenue would typically choose, and it suggests DeepSeek is still betting its next model generation matters more than squeezing more margin from the current one. The company released DeepSeek-V4.1-Flash earlier this month, aimed at faster inference and throughput, which we covered in our report on the V4.1-Flash release.

The open regulatory question

Running alongside the revenue story is a dispute that has nothing to do with pricing. On September 10, Anthropic published a 154-page threat intelligence report alleging that DeepSeek routed more than 12.1 million exchanges to Claude over a 14-day window in July, and that Moonshot routed more than 23 million exchanges between May and July, including close to 300,000 customer requests through accounts Anthropic considered fraudulent. According to the South China Morning Post's account, Anthropic said some of the routed material included Chinese police case data and surveillance footage.

China's Cyberspace Administration summoned all seven labs named in Anthropic's report before narrowing its focus to DeepSeek and Moonshot specifically, per AI Weekly and The Next Web. The regulator is examining whether sensitive Chinese police, military and state-linked data reached a US AI system through the alleged routing. China's Foreign Ministry has publicly rejected Anthropic's characterization, with a spokesperson accusing the US company of distorting facts and stating that China supports AI for good, according to Roic.ai's reporting. No penalty has been decided, and DeepSeek has not publicly detailed its response to the investigation.

Neither the revenue milestone nor the regulatory probe overrides the other. DeepSeek can be genuinely doubling its API business while also facing an open, unresolved question from its own government about how that business handles data, and investors weighing the coming raise will have to price both facts rather than pick the more convenient one.

What determines whether $1 billion holds

Whether the run rate survives contact with a full year depends on questions nobody outside DeepSeek can currently answer with confidence: whether rival Chinese labs undercut the new, higher prices and pull customers back, whether the CAC investigation results in restrictions that affect DeepSeek's API business, and whether the 74-times revenue multiple implied by the IPO target looks reasonable once a full year of post-price-hike data exists rather than a few months of it.

Frequently asked questions