Chinese AI lab DeepSeek has reached roughly $1 billion in annualized revenue, according to a September 24 report from The Information citing multiple sources, a figure that has nearly doubled in recent months following aggressive price increases on the company's models.
CEO Liang Wenfeng shared the revenue milestone during recent closed-door meetings with investors, the report said. The jump — The Information reported the company nearing $500 million in annualized revenue as recently as July — is the clearest sign yet that the low-cost Chinese lab is converting its technical reputation into a durable business, with major implications for the global AI price war. Anyone following the latest AI developments will recognize the stakes: DeepSeek's pricing decisions have repeatedly forced Western labs to respond. For more context on this story, see our ongoing latest AI developments.
Price Hikes Didn't Scare Users Away
The revenue growth was driven in part by DeepSeek's decision to raise fees for its AI models by roughly two to four times, with some reports describing inference price increases of 2.3x to 4.5x implemented last month. Price increases of that magnitude often trigger customer churn, but Liang told investors that the adjustments have not resulted in customer attrition and that demand remains robust.
Even after the hikes, DeepSeek's pricing reportedly remains below that of its competitors, preserving the cost advantage that has defined the company since its breakout. The episode suggests the lab discovered it had been underpricing its models significantly — and that its user base, anchored by developers, was sticky enough to absorb the correction.
Liang also downplayed the milestone itself. "Generating revenue is not DeepSeek's top priority," he told investors, according to the report — a reminder that the lab, which is funded by quantitative trading firm High-Flyer, continues to measure itself primarily in research terms.
Compute Goes to Training, Not Inference
The company's resource allocation backs that up. DeepSeek currently dedicates more than 70% of its computing resources to training new models, leaving less than 30% for serving inference on existing ones, according to reporting on the investor meetings.
To relieve the inference bottleneck, Liang shared a technical observation with investors: internal testing shows the company's lightweight models can already run stably on gaming graphics cards and handle the vast majority of users' everyday tasks. If DeepSeek can shift more everyday workloads onto consumer-grade hardware, it could serve a large user base without proportionally expanding its data center footprint — a meaningful constraint for a company operating under export controls that limit its access to top-tier chips.
The $7.5 Billion Raise and a STAR Market Listing
The revenue growth arrives as DeepSeek moves to secure one of the largest private fundraises in Chinese tech history. The company plans to complete its second funding round before the end of October, targeting around 50 billion yuan — roughly $7 billion to $7.45 billion, depending on the report — at a valuation of up to 500 billion yuan, or approximately $70 billion to $75 billion.
Reuters reported earlier this month that DeepSeek selected CITIC Securities as lead underwriter for a listing on Shanghai's STAR Market, the tech-focused board of the Shanghai Stock Exchange, with a filing expected as the fundraising concludes. The Information reported the company is aiming to raise about $7.45 billion through the IPO by the end of next month.
For context, the raise would put DeepSeek's war chest in the same conversation as the mega-rounds flowing to US frontier labs this year, though DeepSeek's valuation remains a fraction of the figures attached to OpenAI and Anthropic.
What It Means for the AI Price War
DeepSeek doubling its annualized revenue while raising prices upends the assumption that its business model depends on permanent underpricing. Demand held through a 2x-4x price increase — evidence that its models are winning on value rather than just cost.
That pricing power matters beyond DeepSeek. Western labs have spent 2026 competing on agentic capabilities and enterprise contracts while absorbing enormous training costs. If a Chinese competitor can fund frontier-scale research on a fraction of the revenue, the pressure on US labs to justify premium pricing only grows. The next test arrives with the IPO: public-market investors will scrutinize whether the post-hike revenue curve is a plateau or the start of compounding growth.
Stay Ahead of AIThe business of AI is moving fast. Get the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
