SHANGHAI — Liang Wenfeng told investors this week what the quarterly filings of his American rivals have yet to demonstrate: that a frontier artificial intelligence model can generate genuine money at extraordinary margins, even in a market that spent most of 2026 debating whether any of these companies would ever turn a profit.
DeepSeek, the Hangzhou-based startup that rattled Silicon Valley in January when it released a capable reasoning model at a reported training cost of roughly $6 million, has more than doubled its annualized revenue run rate to approximately $1 billion, according to people familiar with the matter briefed on remarks Wenfeng made at a recent investor meeting. Seeking Alpha reported the figures on Thursday, citing The Information’s earlier account. Revenue comes almost entirely from developer API access. The company’s free consumer chat application generates no advertising or subscription income. The gross margin on that API business reached 82.9 percent through July, a level that rivals what US frontier labs report at far higher revenue bases.
The revenue acceleration comes as DeepSeek moves toward a public listing. The company is targeting a fundraise of 50 billion yuan, equivalent to roughly $7.5 billion, at a valuation of 500 billion yuan, or approximately $75 billion, according to people with knowledge of the plans. CITIC Securities has been selected as lead underwriter. The current plan calls for filing the listing application before year-end and completing the IPO on Shanghai’s STAR Market in 2027.
The deal size would make it one of the largest technology fundraises in China this year. DeepSeek’s previous fundraise, at a $71 billion valuation, closed in July, before the API price hikes and before the revenue acceleration now on offer to investors. At $75 billion, the new target is a conservative markup on a business that has since established a margin profile most frontier AI startups can only project.
The arithmetic behind the $1 billion run rate warrants scrutiny. Wenfeng shared the figure with investors in private conversation, not in a formal financial disclosure. No public revenue or earnings filing exists. Annualized revenue run rates are projections extrapolated from recent monthly billings, not audited results. The 82.9 percent margin figure applies through July, before the September launch of DeepSeek-V4.1-Flash, a new model released at lower per-token prices than its predecessors, even as older model tiers absorbed the 2.3 to 4.5 times price hike that Wenfeng credited for the revenue surge. The combined effect on forward margins is not yet known.
Wenfeng’s trajectory to this position is unusual. He built High-Flyer, one of China’s top quantitative hedge funds, before pivoting almost entirely to AI research. His January model drop demonstrated competitive performance against US frontier labs at a fraction of their claimed training costs, briefly erasing hundreds of billions of dollars from semiconductor stocks globally. Huawei and other Chinese hardware makers have since moved to fill the gap left by US chip export restrictions, giving domestic AI labs including DeepSeek infrastructure alternatives that did not exist at scale two years ago.
The competitive landscape for the IPO window also favors movement. Anthropic has delayed its own public offering to November, waiting on third-quarter financials to underpin its $2 trillion valuation target. The contrast is stark: Anthropic raised at a level more than 26 times DeepSeek’s current ask, on a business that has not disclosed a comparable gross margin. DeepSeek’s approach reflects a deliberate decision to let the numbers do the work that narrative does in San Francisco.
The Shanghai STAR Market listing is also a political statement. China’s domestic technology IPO pipeline has been cautious since Beijing imposed broad restrictions on overseas listings in 2021. A DeepSeek listing on STAR Market would serve as a signal that Beijing is prepared to publicly champion its most prominent frontier AI company, competing with artificial intelligence governance frameworks being shaped largely without Chinese participation at international bodies. How quickly the China Securities Regulatory Commission moves on a listing application of this size will be read as an index of that intent.
Whatever the final terms, the $1 billion milestone changes the terms of the conversation. DeepSeek entered 2026 as a research story. It ends September as a revenue story, with a margin structure that, if it holds through the next model generation, suggests the economics of China’s AI industry may be structurally different from the burn-to-grow model that defines its American counterparts.
