What Happened
DeepSeek's annualized revenue run rate has crossed $1 billion, according to a report published September 23 and first detailed by The Information, roughly double the under-$500 million pace the Chinese AI lab was posting just a few months earlier. The jump comes almost entirely from a round of API price increases the company rolled out after announcing them in August, with hikes ranging from 2.3 to 4.5 times the previous rates depending on the model, and peak-hour pricing on the flagship V4 model climbing by a factor of four.
The scale of the shift is easier to see in the raw numbers. DeepSeek generated roughly 475 million yuan, or about $70.7 million, in actual revenue across the first seven months of 2026 alone. That is already ten times what the company brought in during all of 2025, when full-year revenue came to an estimated $7 million. Founder Liang Wenfeng has told investors that despite the size of the increases, the great majority of developers kept paying rather than routing their traffic elsewhere.
Notably, none of this touches DeepSeek's consumer-facing chatbot, which remains free to use. Every dollar of the new run rate is coming from the API business that developers and enterprises tap to build products on top of DeepSeek's models, a detail that matters for how the company is now being valued.
Why It Matters
The headline number is the run rate, but the figure that has caught investors' attention is the margin behind it: DeepSeek's API business posted an 82.9 percent gross margin through July. For a lab that trains its own frontier-scale models from scratch, a margin in that range is unusual, and it is the clearest evidence yet that a Chinese open-weight developer can turn heavy compute spending into durable, repeatable profit rather than running at a loss to chase usage.
What makes the margin more striking is that DeepSeek's prices, even after the increases, are still below what OpenAI and other major labs charge for comparable capability. The company has effectively found room to raise prices sharply while remaining the cheaper option in the market, a position that reflects how much of its cost base sits in infrastructure rather than headcount. By its own account, DeepSeek still directs more than 70 percent of its compute to training new models rather than serving existing traffic, meaning the current margin is being achieved even while the bulk of its GPU fleet is tied up building the next generation rather than monetizing the current one, a detail first reported by Crypto Briefing.
That combination of pricing power and margin discipline is exactly what a company needs to demonstrate before going public, and it lines up with DeepSeek's stated target of listing on Shanghai's STAR Market, the exchange's board for technology companies, sometime in 2027. Regulators and underwriters evaluating that listing will be looking for precisely this kind of evidence that the business behind the model releases can support a market valuation on its own.
Reaction
The clearest reaction so far has come from DeepSeek's own customer base. Liang has described the response to the price increases as muted in the way that matters most for a business: developers grumbled but largely stayed, rather than migrating their workloads to rival APIs. For a company whose entire new revenue base rests on developers choosing to keep paying, that retention is the single most important signal to come out of the past month.
Investors have responded in kind. The willingness of underwriters to finalize a funding round at roughly double DeepSeek's prior valuation, only a few months after its last raise closed, suggests the revenue and margin figures are being read in Shanghai's financial circles as confirmation that the company's commercial model works, not just its research output, a reading echoed in AI Weekly's coverage of the raise.
What's Next
CITIC Securities is leading a new funding round targeting $7.5 billion, or about 50 billion yuan, at a valuation of roughly $75 billion, or 500 billion yuan, according to PYMNTS' reporting. The round is expected to close by the end of October, and would value DeepSeek at nearly 50 percent above the roughly $52 billion valuation it carried when its first major institutional round, worth about $7.4 billion, closed this past June.
From there, the path leads toward the Shanghai STAR Market listing DeepSeek has signaled for 2027, which would make it one of the first Chinese frontier AI labs to test whether public markets are ready to value a model developer the way they value an established software company. Any public offering that far out will also depend on whether DeepSeek can keep its margin steady as it rolls out its next generation of models and whatever pricing structure comes with them.
Closing Thoughts
Liang Wenfeng built DeepSeek after years running the quantitative hedge fund High-Flyer, and the discipline that background implies is starting to show up in the company's numbers as much as in its model releases. A lab that spent its first two years being judged almost entirely on benchmark scores is now also being judged on gross margin, revenue growth, and its readiness for a stock exchange listing, which is a different kind of scrutiny entirely.
That shift says something about where the broader AI industry is heading. The labs that matter most over the next few years may not be the ones with the single best model on a given week, but the ones that can prove their compute spending converts into a business investors can put a price on. DeepSeek's climb to a billion-dollar run rate, built on price increases its customers largely absorbed, is a small but concrete data point in that direction, for a Chinese AI sector that Western labs will now be watching as closely on financial terms as on technical ones.
한글 요약
딥시크(DeepSeek)의 연환산 매출이 지난 9월 23일 보도를 기점으로 10억 달러를 돌파했다. 불과 몇 달 전까지 5억 달러에도 못 미쳤던 수치가 두 배 넘게 뛴 배경에는 지난 8월 발표된 API 가격 인상이 있다. 모델별로 2.3배에서 4.5배까지 가격이 올랐고, 성수기 시간대 플래그십 모델 V4 요금은 네 배로 뛰었다. 실제로 2026년 상반기 7개월간 실적은 약 4억 7,500만 위안(약 7,070만 달러)으로, 2025년 한 해 전체 매출(약 700만 달러)의 열 배에 달한다. 이용자에게 무료로 제공되는 챗봇과 달리, 이 매출은 전부 개발자·기업 대상 API 사업에서 나왔다.
주목할 점은 매출 규모보다 수익성이다. 7월까지 API 사업의 매출총이익률은 82.9퍼센트에 달했는데, 자체적으로 프런티어급 모델을 처음부터 훈련하는 랩으로서는 이례적인 수치다. 가격을 큰 폭으로 올렸음에도 여전히 오픈AI 등 경쟁사보다 저렴한 가격대를 유지하고 있으며, 전체 컴퓨팅 자원의 70퍼센트 이상을 신규 모델 훈련에 투입하면서도 이 정도 마진을 냈다는 점이 특히 두드러진다. 창업자 량원펑은 가격 인상에도 대다수 개발자가 이탈하지 않고 남아 있다고 전했다.
이러한 실적을 바탕으로 중신증권(CITIC Securities) 주관 하에 약 75억 달러(500억 위안), 기업가치 약 750억 달러(5,000억 위안) 규모의 신규 투자 유치가 10월 말 마무리를 앞두고 있다. 이는 지난 6월 마감된 1차 라운드(약 74억 달러, 기업가치 약 520억 달러) 대비 크게 높아진 밸류에이션이다. 딥시크는 2027년 상하이증권거래소 커촹반(STAR Market) 상장을 목표로 하고 있어, 이번 매출·수익성 지표가 향후 기업공개 심사에서 핵심 근거가 될 전망이다.