The number that moved on October 22 was not a benchmark. Bloomberg reported that DeepSeek had raised roughly 100 billion yuan, about $15 billion. For a lab that gives models away, that is a lot of cash to explain.
What is confirmed, and what is repeated
The figure people are quoting comes from Bloomberg, via Techmeme: about $15 billion. Later summaries added color, including a claim that CATL and Tencent led and that a June round had already been large. Treat the extra color as reported, not as a filing you can download from DeepSeek.
Until the company or an investor states the terms, the honest sentence is the short one. A major newsroom says the round is about $15 billion. The cap table is not on the lab’s blog.
What the money is for
DeepSeek’s models are priced and licensed in a way that does not, by itself, pay for a frontier training run. A round of this size is about electricity, memory, and the next cluster. It landed in the same few days as DeepInfer, the open serving stack, which is a hint about the strategy: publish the model, publish the engine, and fund the machines.
US agencies, separately, have been telling civilian offices to get DeepSeek off federal systems. Capital and policy are pulling in opposite directions. Both can be true in the same week.