DeepSeek raises $15B two days after US restricts its models

A ban in one country and a $15 billion cheque in another, in the same week. TechCrunch, citing Bloomberg, reported that Chinese AI lab DeepSeek has raised about $15 billion, around 100 billion yuan, two days after the US ordered federal civilian agencies to remove its technology from their systems over security concerns. CATL, the battery maker, and Tencent are named as potential investors.

The round is one of the largest in China this year and the lab's third in six months. DeepSeek raised more than $7 billion earlier in the year at a $50 billion valuation, and a Bloomberg report later described a further $1.5 billion planned at a $71 billion valuation, with an IPO in China possible next year.

Where the money goes

DeepSeek is "ramping up spending on AI infrastructure and open models for domestic and international markets," according to the Bloomberg report. The company said in April that its V4.1-Flash and R2 inference stacks were part of a push to serve 155 trillion tokens a day through its own hosting, a figure that costs real GPU capacity to add to.

The number also sits against a shrinking Western audience for Chinese AI technology. US federal agencies were told to pull DeepSeek models this week, following earlier moves by the Pentagon and NASA. DeepSeek's own home market and parts of Southeast Asia are where the next trillion tokens of inference will land.

The pattern in Chinese AI funding

The round continues an investor pattern that was running already. Moonshot AI, maker of the Kimi models, closed $2 billion at a $20 billion valuation, TechCrunch reported, and Manus AI raised more than $500 million earlier this month. Zhipu AI, now Z.ai, and MiniMax both went public in Hong Kong earlier this year, with current market capitalisations of roughly $29 billion and $7.2 billion, according to TechCrunch.

Liang Wenfeng, who founded DeepSeek in 2023 after co-founding the quantitative fund High-Flyer, still holds about 81.6 per cent of the Chinese operating entity, according to corporate records that WireScreen analysed for TechCrunch. DeepSeek did not respond to TechCrunch's request for comment.

For a team deciding whose models to build against, the funding tells you a specific thing: the open-weight lane that DeepSeek's R-series, V4.1-Flash and the DeepInfer inference engine occupy has a well-funded backer for the next while. The free developer tier, the open permissive licences and the public kernel code it publishes alongside its models all live or die with how much GPU capacity the company can afford to keep online, and $15 billion buys a lot of H800 time. For a buyer worried about geopolitical exposure, the same round is the reason to put a US-hosted mirror in front of any DeepSeek model you ship into production.

Source

DeepSeek reportedly raises $15B days after US restrictions, Kate Park at TechCrunch, citing a Bloomberg report.