DeepSeek is closing in on a financing round of roughly 50 billion yuan, about $7.4 billion, that would value the Chinese AI lab near $74 billion before the new money lands, according to Tech Startups, which cited reporting from South China Morning Post and the Wall Street Journal. The raise has not closed and neither DeepSeek nor its investors have confirmed it publicly.
The deal matters because of what it funds. DeepSeek built its reputation on doing more with less: its models challenged the idea that competing with frontier American labs required massive compute budgets. A multibillion dollar round earmarked substantially for research and compute infrastructure marks a reversal of that positioning, not an extension of it.
South China Morning Post, quoted in the Tech Startups piece, described DeepSeek as “nearing the completion of a new funding round valuing the company at about 500 billion yuan (US$74 billion) before investment,” attributing the figures to people familiar with the talks. One source put the closing inside August, the outlet reported. Tech Startups ran the arithmetic: take the full $7.4 billion against a pre-money mark of $74 billion and the post-money number lands near $81 billion, a figure worked out in the reporting rather than confirmed by anyone involved.
The financing is being read as preparation for a share listing. A filing to list on the STAR Market in Shanghai could arrive as soon as late 2026, with the debut aimed at 2027, a timeline Tech Startups attributes to people who know the plans. Nothing about the IPO timeline is fixed. Filing dates slip, and Chinese regulators have final say over STAR Market listings.
Existing backers reportedly participating include Monolith, Shixiang Capital and battery maker Contemporary Amperex Technology Limited (CATL). New investors said to be in discussions include CPE, Legend Capital and Stony Creek Capital, a semiconductor-focused private equity firm, along with money connected to the chipmaker GigaDevice and state-backed funds out of Hefei.
The bigger shift is structural. Founder Liang Wenfeng bankrolled DeepSeek mostly out of High-Flyer Quant, a hedge fund he led before the lab existed, drawing on its money and its machines instead of raising conventional venture rounds. That arrangement is straining as the company scales: it is reportedly planning to add close to a gigawatt of computing capacity to train larger models and compete for scarce AI researchers. A gigawatt of capacity is data-center-scale infrastructure, the kind of buildout normally associated with the U.S. labs DeepSeek’s models were designed to undercut on cost.
“DeepSeek’s founding team, including Liang, are still traders at heart and inclined to chase maximum upside,” said Ke Zong, who runs a portfolio at a hedge fund based in Shanghai, speaking to Tech Startups.
The company has also raised its API prices this year, a move Tech Startups tied to a broader repricing across China’s AI market as Alibaba, Tencent, Baidu and Zhipu AI pull back from aggressive discounting. Cheaper inference has not translated into cheaper AI companies to build. Chips, training runs and researcher salaries still draw on the same capital-intensive playbook DeepSeek was supposed to be an exception to.
Liang’s High-Flyer affiliates have also picked up pre-IPO allocations in other Chinese listings drawing attention, among them the memory maker CXMT and the humanoid robotics company Unitree Robotics, where DeepSeek itself holds a 2.31 percent stake under a 36-month lock-up. Hutong Research analyst Sigrid Wang, cited by Tech Startups, called the distinction between quant-fund returns and DeepSeek’s use of its balance sheet to build “strategic relationships around the future AI stack.”
If the round closes as reported, DeepSeek moves further from its origins as a hedge-fund side project and closer to a conventionally capitalized AI company answering to outside shareholders and, eventually, public markets. Operators benchmarking DeepSeek’s API pricing against frontier labs should expect that gap to keep narrowing as its own compute bill grows.
Reporting from Tech Startups, published August 28, 2026.
