DeepSeek Closes a Second Round of About 105.5 Billion Yuan, Nearly $14.8 Billion, at a Roughly $86 Billion Valuation

DeepSeek, the Chinese AI lab behind the V4 model family and the open-source DeepInfer inference stack, has closed a second external funding round of about 105.5 billion yuan - roughly $14.8 billion - according to foreign media reports relayed on October 30 by Sina Technology and Wallstreetcn. The round values the Hangzhou-based lab at more than 615.5 billion yuan, or approximately $86 billion. Neither DeepSeek nor its backer High-Flyer has officially confirmed the figures; the report relies on people familiar with the matter.

Nearly 30 investors participated. Tencent invested another 10 billion yuan in this round, matching its first-round check and bringing its cumulative commitment to 20 billion yuan, and co-led the round alongside battery giant CATL, which is making a continued bet from the earlier round. The investor list reads like a cross-section of China's tech supply chain: new entrants include memory maker ChangXin Memory (CXMT), foundry SMIC, delivery platform Meituan, smartphone vendor vivo and social commerce platform Xiaohongshu - spanning chip manufacturing, consumer electronics and internet services.

One structural detail stands out. State-owned capital now accounts for about 27 percent of the company's shareholding after this round, up from the previous level, according to the reports. Combined with the presence of China's two leading chipmakers among the investors, the cap table increasingly looks like a coordinated build-out of a domestically controlled AI stack rather than a conventional venture round.

The round's size reportedly doubled from earlier plans. A report about three and a half weeks ago said DeepSeek was set to raise at least $12 billion backed by Tencent and CATL; the final figure of about 105.5 billion yuan exceeded that, and Chinese relays attributed the expansion to rising chip prices and a deliberate effort to prevent speculative bubbles by broadening the investor base. These explanations come from unnamed sources and should be treated accordingly.

The funding lands while DeepSeek's technical momentum is measurable. Its V4 family commands the largest traffic on OpenRouter, with 33.6 trillion weekly tokens as of mid-October, and the lab has open-sourced DeepInfer, the inference framework behind its models, along with search APIs and domestic chip compatibility testing reported by Chinese brokerages. Unlike its American frontier peers, DeepSeek has kept its models open-weight and its prices aggressively low, monetizing indirectly through its parent hedge fund's trading operations until now.

The shareholder composition is the real signal. The chipmakers who invest in DeepSeek also sell the silicon it runs on; the consumer platforms who invest also distribute the models its weights power. Chinese capital markets have seen this pattern before in semiconductors and EVs - a strategic supply chain underwriting its own demand. If the reported numbers hold, DeepSeek now has a war chest on the scale of a frontier lab, without having abandoned open weights. The next test is official confirmation, which neither DeepSeek nor any investor has yet provided.