DeepSeek Nears $7.4B Raise at a $74B Valuation

DeepSeek is close to signing a 50 billion yuan raise — about $7.4 billion — at a pre-money valuation of roughly 500 billion yuan, or $74 billion, which would put the company at about $81 billion post-money. Signing was expected by the end of August, and the round is the clearest signal yet that the Hangzhou lab is being groomed for a public listing rather than run as a research shop.

The pricing is a steep markup. The new valuation sits roughly 43 percent above where DeepSeek's first outside round landed just two months earlier — an unusual re-rating over a single quarter, and one that says more about Chinese investor appetite for a domestic frontier lab than about any single product launch. Existing backers CATL, Monolith and Shixiang Capital are participating again, while CPE, Legend Capital, Stony Creek Capital, funds tied to GigaDevice and state investment vehicles from Hefei have been in discussions, according to reporting from Tech Startups and Bloomberg. The mix of battery manufacturer, growth funds and municipal state capital is characteristic of how large Chinese AI rounds now get filled.

An IPO is the point. DeepSeek could file as soon as the end of 2026, with a debut targeted for 2027 on Shanghai's STAR Market. That timeline explains the capital-structure work: for most of its life DeepSeek was funded out of High-Flyer Quant, the quantitative hedge fund founder Liang Wenfeng also runs, an arrangement that produced remarkable models on a shoestring but is not what an exchange wants to see on a prospectus. This round completes the shift from hedge-fund side project to independently capitalised company.

The money has an obvious destination. DeepSeek is planning to add roughly a gigawatt of computing capacity — a reminder that efficiency at the model level does not exempt you from the capital demands of serving at scale. The lab built its reputation on getting frontier-adjacent results for a fraction of Western training budgets, and it has kept shipping on that basis: it pushed V4-Pro to general availability with no announcement at all earlier this month, after gaining ten index points on V4-Flash through post-training alone in July. Cheap models still need expensive inference once a few hundred million people are using them.

At $74 billion pre-money, DeepSeek would rank among the most valuable private AI companies outside the United States — a striking number for a lab that was, eighteen months ago, best known for undercutting everyone else on price. It is also a number that only makes sense with a domestic listing behind it, since the capital and the eventual buyers are both Chinese, and Beijing's $295 billion plan to build out a national AI grid is the demand curve a STAR Market listing would be priced against.

What the round does not settle is revenue. DeepSeek's commercial traction has never been the story its valuation rests on, and a STAR Market prospectus will eventually require one. The distance between a $74 billion price tag and a business built largely on being the cheapest option in the market is the thing to watch between now and a filing.