Article
DeepSeek built its reputation by showing that capable AI doesn't have to cost a fortune. Now it has a fortune of its own. According to Bloomberg and Reuters, the Chinese AI lab has raised about 100 billion yuan, roughly $15 billion, in one of the largest funding rounds China has ever seen.
What happened
Bloomberg and Reuters reported the raise on October 22, citing people familiar with the matter. DeepSeek lifted its target at least twice during talks, from 50 billion yuan to 80 billion and finally to about 100 billion. Reuters put the pre-money valuation at roughly $74 billion. Battery maker CATL and Tencent were reported among the largest investors, with many of the new backers coming from industrial capital firms.
The timing is striking. The round surfaced two days after the US government ordered federal civilian agencies to remove DeepSeek's AI technology from their systems over security concerns.
What isn't settled
In the latest reports, the round hadn't formally closed, and the final terms and investor list could still change. Coverage also differs on which investors led the deal. DeepSeek hasn't publicly laid out how it will spend the money, though reports point to computing infrastructure, research and a possible stock market listing in China.
Why the price war matters to you
DeepSeek's models are open-weight and aggressively priced, and its low-cost releases have kept pressure on the prices Western labs can charge. A lab with $15 billion more to spend can train larger models, buy more compute and hold prices down for longer. For any company paying for AI, that pressure is good news. Cheaper capable models make more products viable and give buyers leverage in negotiations.
The complication
There's a second, harder story. The US government now treats DeepSeek's technology as a security risk inside its own systems. For startups, that creates a split decision: the cheapest capable model may be off-limits if your customers are in government, defense, finance or healthcare, or simply if an enterprise security team asks where your AI comes from.
Open weights aren't the same as the app
One nuance matters here. Running an open-weight model on your own servers is very different from sending data to a company's hosted service. In the first case your data stays inside your own infrastructure. In the second, it travels to the provider. Procurement and security reviews don't always draw that distinction, though, so even self-hosted use can trigger tough questions from sensitive customers. Knowing exactly how you use a model is part of being able to defend that choice.
Price your model choice in trust, not just tokens
Founders should enjoy the price pressure without letting it make their decisions for them. Before adopting any model, think through who your customers are and what their security and procurement teams will accept, then document where each model in your stack comes from and where its data goes. If you sell into regulated or government markets, plan a model lineup that passes those reviews even when it costs more per token. And if your margins are thin, use the competition to negotiate with your current providers, because every major lab is watching what a well-funded, low-cost rival does to the market.
This article is information, not investment advice.
Leave a comment