DeepSeek Raised Prices and Doubled Revenue to $1B
2 min readDeepSeek raised its API prices sharply in August. Its customers did not leave, and its revenue run rate doubled. The Chinese lab that started the industry’s price war is now demonstrating that cheap was a choice, not a ceiling.
From Price Leader to Price Setter
DeepSeek built its reputation on undercutting everyone. Its models shipped with open weights and were served through an API priced far below Western frontier labs, which is what made the company a household name in early 2025 and forced rivals to respond.
In August the company reversed course and raised API fees by 2.3 times to 4.5 times, depending on the model. Founder Liang Wenfeng has said the increase did not shrink the customer base.
The DeepSeek Revenue Numbers
According to The Information, DeepSeek’s annualized revenue run rate has reached roughly $1 billion, more than double the pace of under $500 million reported only months earlier. The margin picture is the more striking part. Across the first seven months of 2026 the company’s overall gross margin was 44.6 percent, while its API business alone ran at 82.9 percent.
DeepSeek is also finalizing a second funding round, targeting roughly 50 billion yuan and aiming to close by the end of October at a reported valuation near 500 billion yuan. Its first external round closed in June, with Tencent and CATL among the investors. The company is preparing a listing on the Shanghai Stock Exchange.
Why It Matters
The prevailing story about Chinese AI has been volume at a loss: cheap inference bought with subsidies and sacrificed margin. An 82.9 percent API margin does not fit that story. It suggests DeepSeek’s efficiency work, done partly under export controls that limited its access to top Nvidia parts, produced real unit economics rather than a marketing position.
The direction of travel is also the opposite of the West’s. OpenAI cut API prices roughly in half this week with its GPT-6 tier. DeepSeek raised its own and grew anyway. That divergence matters for anyone building on these models, because the assumption that inference gets cheaper every quarter is no longer safe everywhere. Watch the October funding close and the eventual IPO filing for the first audited look at whether those margins hold.