DeepSeek Valuation Hits $74B as Funding Round Tops $12 Billion

What happened

DeepSeek, the Hangzhou AI lab that for years ran largely on its founder's hedge-fund money, is closing one of the largest private rounds in Chinese tech history. Reports say the company has secured at least 80 billion yuan, about $12 billion. Signed term sheets suggest the total could approach 100 billion yuan, roughly $14.9 billion 45. That is well above the original target of around 50 billion yuan, or $7.4–7.5 billion 35.

The price is the headline. The round reportedly values DeepSeek at about 500 billion yuan before the new money, roughly $74–75 billion 24. One account puts its valuation at just over $5 billion as recently as June, which would make this roughly a fourteenfold markup in a few months 3. Tencent and battery giant CATL are leading, and Tencent had already put in 10 billion yuan earlier 45. Existing backers Monolith Management and Shixiang are staying in, along with several funds tied to Chinese local-government financing platforms 3.

The money is aimed at compute, model training and IPO preparation. DeepSeek has reportedly hired banks and advisors for a 2027 listing in Shanghai, likely on the STAR Market 23. Two reports place the listing in early 2027 45. On the infrastructure side, DeepSeek is expanding capacity with 160,000 Huawei chips 4.

The revenue question

The numbers diverge most on revenue. One report says DeepSeek hit a $1 billion annualized run rate as of September 2026, more than double its earlier figures. The same report cites API gross margins of 82.9% 2. Another puts annualized recurring revenue at about $500 million and describes the company as still burning cash while it expands 3.

The gap matters. Against $500 million, a $74 billion valuation is about 148 times revenue. Against $1 billion, it is closer to 74 times. Either multiple is aggressive, but the lower one is easier to defend for a lab whose investors are underwriting long-horizon research. The two figures may also measure different things, such as recurring revenue versus a total run rate, or different points in time. They are not necessarily contradictory. Still, the range shows how loosely private AI companies are being priced.

Why V4.1-Flash changed the conversation

Several reports tie the surge in investor demand to the September release of the V4.1-Flash model 45. One describes it as cutting operating costs relative to proprietary models from OpenAI and Anthropic 4.

Current published rates for V4.1-Flash are:

Off-peak rate	Per million tokens
Input, cache miss	$0.15
Input, cache hit	$0.003
Output	$0.60

Peak-hour prices double. The flagship V4-Pro lists at $0.66 per million input tokens 1.

The price history shows a quieter shift. Until September 9, Flash output cost $0.28 per million tokens. It now costs $0.60 off-peak, more than double 1. Input prices rose only slightly. Peak pricing also lets DeepSeek charge more for scarce capacity. Taken together, these look like a company tuning prices for margin rather than pure market share, which fits the 82.9% gross-margin figure 2. DeepSeek is still very cheap by Western standards: a team pushing 100 million tokens a day can pay around $610 a month off-peak 1. But the era of near-free pricing as a growth strategy appears to be giving way to monetization ahead of an IPO.

A structural shift in ownership

The round also changes how investors hold DeepSeek. Previously, most backers held stakes indirectly through a limited partnership managed by founder Liang Wenfeng. The new round opens the door to more direct investment 3.

That is close to a prerequisite for a public listing, and it signals a step back from the closely held, research-first structure that defined DeepSeek's early years. The arrival of local-government-backed funds 3 and the Huawei chip deployment 4 also point to DeepSeek's growing role as a strategic national asset, not just a startup.

Our reading

The valuation is less a bet on today's revenue than on DeepSeek remaining China's cost-efficient frontier lab while building a domestic compute stack. The oversubscription suggests investors see the strong margins and recent price increases as proof that cheap models can still be profitable.

The risks are real. Revenue figures are inconsistent across reports. The company is still described as cash-burning 3. And raising Flash output prices could slow the developer adoption that powered the story in the first place. Some round details may still change in final negotiations 5. If DeepSeek can show a credible $1 billion revenue base by the time it files for its IPO, the multiple will look ambitious but defensible. If the lower figure holds, public-market investors will be asked to pay a very steep price for an AGI-first lab.