DeepSeek raises about $15 billion from Tencent and CATL

DeepSeek has raised about $15 billion from backers including Tencent and CATL, Bloomberg reported on October 22 — one of China's largest private financings, and it lands three days after Washington ordered the company's technology out of federal civilian systems. The company has not confirmed the round.

Key facts

• About 100 billion yuan ($15 billion) raised, per Bloomberg's October 22 report citing people familiar with the matter — no company confirmation (Bloomberg)
• The target rose at least twice — from an initial 50 billion yuan to 80 billion, then to the final 100 billion (Investing.com/Reuters)
• CATL and Tencent were among the largest investors, according to Bloomberg's sources; neither has confirmed participation (Investing.com/Reuters)
• The round is "set to be complete in the coming days" and financial details "could still change," a second source told Reuters
• DeepSeek was valued at about 500 billion yuan (~$75 billion) before the new capital, sources told Reuters
• The DHS federal civilian removal order was published October 20, barring DeepSeek products and third-party services built on DeepSeek implementations (Federal Register)

What happened

Bloomberg News reported on October 22, citing people familiar with the matter, that the Hangzhou-based AI lab has raised about 100 billion yuan — roughly $15 billion — in its latest funding round, which the outlet called one of China's largest ever private financings. Reuters picked up the report the same day, adding that the fundraising is set to be complete in the coming days but that a second source cautioned the financial details and investor lineup are not ironclad and could still change.

The money trail is unusual in one specific way: the target kept moving. According to the Bloomberg report as picked up by Investing.com and Reuters, DeepSeek initially sought 50 billion yuan, raised the target to 80 billion, and closed at about 100 billion. Contemporary Amperex Technology (CATL), the world's largest EV battery maker, and Tencent Holdings, operator of WeChat, were among the largest investors.

Attribution matters here, and it is worth being precise about what is and is not confirmed. Every number in the round traces to anonymous sourcing — Bloomberg's "people familiar with the matter" — not to a filing or a company statement. Reuters could not reach DeepSeek for comment. Neither CATL nor Tencent has confirmed its stake. Until DeepSeek or its investors say so themselves, the $15 billion figure is a well-sourced report, not a verified fact.

Why it matters

For builders, the interesting part is not the funding number itself but what the capital is for. Bloomberg reports DeepSeek has increased spending on data centers and open-source models as it prepares for a potential initial public offering. That maps directly onto what the company has actually shipped: it open-sourced DeepInfer, the Apache-2.0 serving framework behind its production V4.1-Flash traffic, on October 20, and its open-weight release cadence through 2026 has been the fastest among major Chinese labs. A $15 billion war chest is a bet that the open-weights-plus-cheap-inference strategy continues at larger scale — more compute for training and serving, and presumably more of the serving-stack work that DeepInfer represents.

The timing also frames the open-weight supply side. The round closes days after the Department of Homeland Security's removal order for DeepSeek products across federal civilian IT was published in the Federal Register, and about a week after China proposed draft export controls covering large models, agents, and compute clusters. Whatever US policy says about government systems, the market signal from China's largest corporates is the opposite: Tencent, CATL, and roughly 100 billion yuan of additional capital are going into the same lab.

There is a gap worth naming, though. As Bloomberg's report notes, OpenAI and Anthropic have each raised more than $100 billion cumulatively. Even at $15 billion, DeepSeek has considerably less financial firepower than the US labs it is chasing. The capital narrows the gap; it does not close it.

Background

DeepSeek spent years rejecting outside capital entirely — a deliberate strategy while it was funded by the quant fund High-Flyer. That ended in June 2026, when Reuters reported its maiden funding round of about $7 billion, at a possible valuation of up to $59 billion, with Tencent and CATL already named as the biggest external investors. The current round roughly doubles the maiden one and, at about 500 billion yuan pre-money per sources talking to Reuters, roughly doubles the valuation too.

The trajectory across the fall tells you how demand exceeded plan. On October 6, Bloomberg reported DeepSeek was close to securing at least 80 billion yuan, with a final tally that could reach 100 billion. Sixteen days later, that upper bound is the reported close. CNBC's coverage of the same process added that DeepSeek is eyeing an IPO in early 2027 — again a sourced report, not an announcement.

For teams running DeepSeek models in production, the relevant questions are practical ones: whether the capital shows up as faster open-weight releases (the release cadence is the supply-side input for anyone self-hosting or routing across providers), and whether the DeepInfer serving stack gets the engineering investment to close the gap with commercial inference frameworks. Our coverage of how the official DeepSeek API differs from third-party hosts and of what DeepSeek does with your data tracks the operational side.

What's next

Three things to watch. First, confirmation: Reuters reports the round wraps in the coming days, so a filing, a company statement, or an investor disclosure confirming the 100 billion yuan figure and the full investor list should arrive shortly — until then the number stays a claim. Second, the IPO path: Bloomberg has reported DeepSeek is preparing for a potential listing, with CNBC's sourcing pointing at early 2027; any formal filing would settle the valuation question. Third, follow the compute: whether DeepSeek's data-center spending produces another frontier-class open-weight release — and what happens to the DeepInfer framework — is how this capital becomes real for builders.