DeepSeek nears $7.4B round at $74B valuation

DeepSeek is reportedly closing a funding round of roughly 50 billion yuan, about $7.4 billion, that would value the Chinese AI lab at $74 billion before the new money goes in. Reports put a close by the end of August, ahead of a planned Shanghai STAR Market IPO filing as soon as late 2026, with a public debut targeted for 2027.

If the numbers hold, the valuation more than doubles the roughly $50 billion mark DeepSeek set in its previous round this past June. Existing backers Monolith Management, Shixiang, and battery maker CATL are expected to continue investing, alongside funds tied to Chinese local-government financing platforms. Proceeds are earmarked for model research and, notably, expanding computing infrastructure.

Why the compute line matters

DeepSeek built its reputation on getting more out of less hardware, its DSpark and V4-Flash releases both leaned on inference efficiency rather than raw scale. A $7.4 billion raise aimed partly at buying more compute is a tell: efficiency gains alone are no longer enough to keep pace with frontier labs that are still scaling raw training and serving capacity. It follows DeepSeek’s own price hikes earlier this month, another sign the “cheap AI” era it helped create has real costs behind it, and it lands in the same week Nvidia committed $3.5 billion to MediaTek to keep its own infrastructure central to exactly this kind of buildout.