AI Agent UpdateDeepSeek is closing a $15B war chest — the open-weight assault is going commercial
CATL and Tencent are anchoring a round that doubled its target, one day after the US ordered DeepSeek off federal systems. The research lab is becoming an empire.
The News
DeepSeek, the Chinese AI lab that rattled global markets with its low-cost models, is closing one of the largest private funding rounds in AI history: roughly $15 billion, at a pre-money valuation of 500 billion yuan — about $70 billion — according to Bloomberg and Reuters reporting on October 21–22. “Pre-money” means the $70 billion is what investors agreed the company was worth before their new cash goes in. The round is anchored by CATL, the world's biggest EV battery maker, and Tencent, China's biggest social and gaming platform, and the full list runs past 15 investors: Midea, Meituan, Geely, GigaDevice, Boyu Capital, CPE and Legend Capital among them. DeepSeek has doubled its original fundraising target of 50 billion yuan.
The timing is pointed. The round closes days after the US government ordered DeepSeek removed from federal systems. It also follows DeepSeek open-sourcing DeepInfer, its serving stack — the software layer that runs models for users — and Reuters reported in September that the company has hired CITIC Securities to prepare an initial public offering on Shanghai's STAR Market, China's tech-focused stock exchange. An IPO prep plus a $15 billion private raise is not a research lab's financial plan; it is a commercial empire's.
One clarification on the ticker: DeepSeek is a private company, so there is no DeepSeek stock. NVDA is tagged as the closest listed read-through — a company this size, training and serving models at this scale, is a GPU demand signal, and DeepSeek's growth also feeds the China chip-demand story that runs through Nvidia's numbers.
The open-weight underdog just got a $15B war chest from CATL, Tencent and friends. The closed labs' pricing power has a new problem.
Why It Matters
A $15 billion war chest backed by China's biggest battery maker, its biggest social platform and more than a dozen other industrial and financial heavyweights says DeepSeek is done being a research curiosity. The strategy is visible: pair the capital with the DeepInfer open-source release — win developers with free, frontier-grade infrastructure, then monetize at scale. Free plus $15 billion of staying power is a pricing problem for every closed lab that charges for access.
For the listed market, the tension cuts through $NVDA in both directions: a bigger DeepSeek means more total demand for AI compute (bullish for the GPU suppliers feeding both sides of the Pacific race) and more pressure on the API prices that justify US labs' valuations (bearish for the closed-model premium).
Analyst's Take
The timing is the story. Days after Washington ordered DeepSeek off federal systems, Beijing's industrial champions doubled the fundraising target. This is the US–China AI race expressing itself as capital formation: closed-lab moats on one side, an open-weight challenger with a $15 billion balance sheet on the other. The moat question for OpenAI and Anthropic is no longer just “are your models better” — it is “can your prices survive a competitor that gives the infrastructure away and has $15 billion to burn while doing it?”
DeepSeek's inference pricing — around $0.60 per million tokens, a fraction of US rivals — is now backed by one of the largest war chests in the industry. Until the STAR Market IPO produces audited numbers, investors are pricing a narrative rather than a profit-and-loss statement. But narratives with $15 billion behind them have a way of becoming financial statements.
By The Numbers
FAQ
Who is backing the round?
CATL and Tencent are the anchor investors. The wider list of 15-plus includes Midea, Meituan, Geely, GigaDevice, Boyu Capital, CPE and Legend Capital — a mix of China's biggest industrial companies and major investment firms.
What will DeepSeek do with $15B?
Fund the commercial expansion: compute for training and serving models, the DeepInfer serving stack it just open-sourced, and the push from research lab toward revenue-generating company — with a STAR Market IPO in preparation via CITIC Securities.
How does this affect US labs' pricing?
DeepSeek charges around $0.60 per million tokens for inference — using its models — undercutting closed US labs. A $15B balance sheet lets it sustain low prices far longer, pressuring the pricing power that supports US labs' valuations.
What is the STAR Market IPO plan?
Reuters reported in September that DeepSeek hired CITIC Securities to prepare a listing on Shanghai's STAR Market, China's board for technology companies. An IPO would publish audited financials for the first time — until then, the company's economics are estimated from the outside.
Source: Bloomberg / Reuters via BusinessWorld