DeepSeek Nears a $7.4 Billion Round at a $74 Billion Valuation — Its Second Mega-Raise in Three Months

Three months after taking outside money for the first time in its history, DeepSeek is reportedly on the verge of doing it again — at nearly 1.5 times the valuation. According to the Wall Street Journal, the South China Morning Post and CNBC, the Chinese AI lab is closing in on a new funding round of roughly 50 billion yuan (about $7.4 billion) that would value the company at around 500 billion yuan — approximately $74 billion — before the new money arrives. The round was expected to close before the end of August, with an official announcement possible in the coming days.

If the terms hold, DeepSeek will have gone from a quant-fund side project valued in the low billions to one of the most valuable private AI companies on Earth in well under a year. The speed of the repricing is startling even by 2026 standards: the lab’s first external round, closed in June, brought in about $7.4 billion at a valuation north of $50 billion. This second raise, at $74 billion pre-money, would value the post-money company above $80 billion — all within a single quarter.

Who is putting in the money

The prospective investor list reads like a reunion of the first round, minus the internet giants. Monolith Management and Shixiang Capital, both existing shareholders, are reportedly committing new capital alongside CATL, the world’s largest battery maker, whose involvement in Chinese AI has deepened steadily over the past year. Funds backed by Chinese local-government vehicles are also expected to participate, according to people familiar with the matter.

Notably absent from the reported lineup are Tencent, JD.com and NetEase, all of which took part in the June round. The second raise appears designed around a different constituency: state-affiliated capital and strategic industrial investors rather than consumer-internet platforms.

The deal structure is changing too. In the first round, most investors were required to route their money through a limited partnership managed by founder Liang Wenfeng himself. The new round reportedly permits more direct investment in DeepSeek — a small technical detail that signals a meaningful shift in governance, away from the founder’s total control and toward the more conventional shareholder structure a public listing demands.

Why the quant fund can no longer carry the load

The deeper story behind the raise is the changing economics of DeepSeek’s original patron. High-Flyer Quant, the hedge fund Liang co-founded, financed the lab from the start and supplied it with a large stockpile of GPUs when export controls began biting. That arrangement made DeepSeek famous as the AI lab that seemingly needed no one’s money.

That era is ending for a practical reason: Rhodium Group analyst Ciel Qi told CNBC that High-Flyer’s revenue has become unstable, even as DeepSeek’s compute appetite keeps compounding. The reported $7.4 billion is earmarked for research and development and a major expansion of computing infrastructure — unrot’s roundup of the deal pegs the ambition at roughly a gigawatt of new capacity. A hedge fund’s trading P&L is a volatile foundation for that kind of buildout; a dedicated capital round is not.

There is also a revenue data point that frames the valuation debate. People familiar with the company’s finances put DeepSeek’s annual recurring revenue at about $500 million. A $74 billion pre-money valuation on $500 million of ARR is a multiple of roughly 148x — a number that would raise eyebrows even in this year’s forgiving market, and one investors are evidently being asked to justify on growth, open-weight ecosystem leverage, and strategic scarcity value rather than current financials.

The IPO runway

The round is widely read as a pre-IPO maneuver. DeepSeek has reportedly hired banks and advisers for a listing on Shanghai’s STAR Market, with a filing possible before the end of 2026 and a public debut targeted for 2027. A listing would be a landmark event: ordinary investors would get their first direct stake in the company whose V-series models detonated the global price war in early 2025 and forced every Western lab to defend its cost structure.

The organization has already been positioning itself inside China’s hardware ecosystem ahead of that debut. High-Flyer affiliates received a combined $26 million allocation in pre-IPO shares of CXMT, the memory-chip maker that recently began domestic HBM3E production. Separately, DeepSeek took 2.31% of Unitree Robotics’ offering and accepted an unusually long 36-month lock-up — three times the 12 months most strategic backers accepted. These are not passive financial positions; they are supply-chain relationships in chip memory and humanoid robotics, the two inputs DeepSeek’s next generation of models and agents will depend on.

What this means for the AI landscape

The competitive logic behind the raise is straightforward. DeepSeek is racing Alibaba’s Qwen team, Tencent’s Hunyuan line, Moonshot and Zhipu for both talent and compute inside China, while OpenAI, Google and Anthropic spend well over $100 billion each on infrastructure outside it. In that race, a $7.4 billion infusion — even at a demanding valuation — buys survival at the frontier, and the involvement of local-government funds signals which way Beijing’s support is flowing.

The round also closes a chapter in the DeepSeek mythology. The lab’s brand was built on ascetic efficiency: world-class models from a shoestring budget, no external capital, no hype cycle. The 2026 version of DeepSeek still ships open weights, but it now raises like a Silicon Valley frontier lab, takes strategic stakes in chipmakers, and courts state capital for a state-side listing. Efficiency got it to the table; scale is what it is now buying.

As ever with DeepSeek, caveats apply. Neither the company nor High-Flyer has confirmed the terms, and the conditional language of “people familiar with the matter” hangs over every figure. The round had not closed as of the latest reporting, and IPO timelines in China’s choppy market have a habit of sliding. But the direction is unambiguous: the most famous bootstrapped AI lab in the world has decided that bootstrapping, like its cheap-model pricing, was a phase.