DeepSeek, the Hangzhou-based AI lab that upended US tech markets in January 2025 with the release of its R1 reasoning model, has hit a $1 billion annualised revenue run rate. The milestone, first reported on the company's financials this week, transforms DeepSeek from an open-weight disruptor into a commercial operator with real recurring revenue.
The number puts DeepSeek in rarefied company. Reaching a $1 billion run rate is the same threshold that Anthropic crossed in late 2024 and that OpenAI passed in mid-2023. Both American labs got there on the back of multi-billion-dollar venture rounds and deep partnerships with hyperscalers. DeepSeek got there while releasing model weights for free and operating from outside the US venture ecosystem.
That distinction is the story. DeepSeek's public identity has been shaped by open-weight releases — R1 in January, V3 before it, and a steady cadence of updates since — that let developers download and run the models locally at no cost. The commercial revenue, by contrast, comes from paid API access, enterprise deployments inside China, and inference services priced well below what US labs charge for comparable capability.
Key facts
- 01DeepSeek's annualised revenue run rate has reached $1 billion, according to reporting on the Hangzhou-based lab's financials.
- 02The milestone puts DeepSeek in the same commercial tier as several Western AI startups that raised billions in venture funding to reach similar run rates.
- 03DeepSeek rattled US tech stocks in January 2025 with the release of R1, an open-weight reasoning model competitive with frontier US systems at a fraction of the training cost.
- 04The company has continued to release open-weight models throughout 2025 and 2026, monetising through API access and enterprise deployments in China.
The January R1 release wiped roughly $1 trillion in market value off US tech stocks in a single session, with Nvidia alone losing close to $600 billion in market cap on the day. The trigger was the model's benchmark performance, competitive with OpenAI's o1, paired with training-cost disclosures that suggested DeepSeek had built the system for a small fraction of what Western labs were spending. Investors read it as a repudiation of the compute-scaling thesis underpinning much of the AI trade.
That thesis has since recovered, but DeepSeek's revenue trajectory suggests the company is doing more than one-off benchmark stunts. A $1 billion run rate implies sustained paid usage — developers and enterprises repeatedly choosing DeepSeek's hosted inference over free-tier alternatives, or over rival Chinese offerings from Alibaba, Baidu, and Moonshot.
Pricing is central to the pitch. DeepSeek's API rates have consistently undercut US competitors, sometimes by an order of magnitude, and the company has passed through cost reductions as inference efficiency improved. For Chinese enterprises operating under export controls that limit access to Nvidia's top-end chips, DeepSeek's ability to deliver frontier-class capability on domestically available hardware is a practical procurement advantage, not just a nationalist talking point.
The company's structure remains unusual. DeepSeek was founded in 2023 by Liang Wenfeng, who also runs the quantitative hedge fund High-Flyer, and it operates without the sprawling headcount typical of US frontier labs. Reporting through 2025 pegged DeepSeek's research team at roughly 150 people, a fraction of what OpenAI, Anthropic, or Google DeepMind field. The revenue-per-employee math, if the $1 billion figure holds, would be extraordinary.
US policymakers have taken notice. DeepSeek has been named repeatedly in congressional hearings and Commerce Department briefings on Chinese AI capability, and several US states and federal agencies have restricted the app on government devices. The Trump administration has continued the Biden-era policy of tightening chip export controls, with recent measures targeting Shenzhen-based buyers, though DeepSeek itself has not been individually sanctioned.
The $1 billion figure is a run rate, not audited annual revenue, and DeepSeek has not confirmed the number publicly. Chinese AI companies operate under disclosure regimes that make independent verification difficult, and the reporting relies on unnamed sources with visibility into the company's financials. A revenue milestone is also not the same as profitability — inference at DeepSeek's price points may still run at a loss when fully loaded with training and infrastructure costs.
Still, the direction is clear. DeepSeek has moved from a curiosity that spooked markets for a day to a commercial AI business with revenue in the same range as the Western labs it was supposed to be undercutting. That reframes the competitive picture for OpenAI, Anthropic, and Google, whose pricing power in inference has long assumed that the credible frontier-class alternatives were also expensive. If a Chinese lab with 150 researchers and open weights can carry a $1 billion book, the moat around Western hosted inference is narrower than the valuations imply — and the next round of API price cuts is coming from Hangzhou, not San Francisco.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
