Analysis
DeepSeek, the Hangzhou lab whose R-series models reset global expectations for training efficiency in early 2025, is looking for fresh outside capital and will not be able to lean on High-Flyer Quant, the quantitative hedge fund co-founded by Liang Wenfeng that bankrolled the company for years, CNBC reported. The cost of staying at the frontier has risen far faster than one hedge fund's balance sheet can absorb.
The financing history is unusual. DeepSeek took no external money at all until June 2026, when it raised $7.4 billion at a valuation above $50 billion. Liang personally committed RMB 20 billion of that. Tencent put in RMB 10 billion and battery giant CATL RMB 5 billion, making them the largest outside shareholders. Reporting since has pointed to a follow-on at a valuation as high as $71 billion, ahead of a planned listing on Shanghai's STAR Market -- the city's Nasdaq-style board -- as early as the second quarter of 2027.
Why the self-funding model broke
DeepSeek's original identity was that it did not need venture capital. That was a genuine competitive claim in 2025, when its efficiency work suggested frontier capability could be reached for a fraction of the American labs' spend. Two years of scaling later, the arithmetic has caught up. Export controls raised the effective cost of compute in China, and the labs it is competing with -- Alibaba's Qwen team, Moonshot, Zhipu, ByteDance's Seed -- are all funded by parents or investors with deeper pockets than a quant fund.
The comparison that matters
Set the numbers side by side. OpenAI and Anthropic are heading toward listings measured in the trillions and hundreds of billions respectively. DeepSeek at $71 billion is priced at a fraction of that despite models that repeatedly benchmark close to the frontier. Part of that gap is real -- monetization, enterprise distribution and international trust are all weaker. Part of it is jurisdictional: a Chinese AI lab cannot access US public markets, cannot sell into US government workloads, and cannot buy top-end Nvidia silicon. The STAR Market listing is not a preference, it is the only door.
The read for Western investors
Almost nobody reading this can buy DeepSeek. The reason to track it anyway is that it sets the floor price for open-weight capability. Every enterprise procurement conversation in 2026 has a version of "why not just run DeepSeek weights on our own hardware" in it, and that question caps what OpenAI and Anthropic can charge for mid-tier inference. Nvidia has been bolstering support for Chinese open models even while warning of a White House crackdown -- because those weights drive GPU demand regardless of who trained them.
The date to hold is Q2 2027. A STAR listing at $71 billion-plus would be the first true public comparable for a frontier lab anywhere, and it will arrive before OpenAI's or Anthropic's.
The STAR Market context
Shanghai's STAR Market has been an inconsistent venue. Launched in 2019 as China's answer to Nasdaq, it saw listing approvals slow sharply through 2023 and 2024 as regulators throttled new issuance to support secondary prices, then reopen selectively for strategic technology sectors. AI is squarely in the favored category, which is why DeepSeek, Moonshot and Zhipu are all pointed at it. The tradeoff for issuers is a domestic investor base with limited appetite for pre-profit companies and a regulator that can pause the queue at will. High-Flyer, Liang's quant fund, has itself been positioning around China tech listings -- which is the connective detail in the CNBC piece and a reminder that the founder's two vehicles are exposed to the same market.
