DeepSeek Tells Investors Not to Sign a $1.5 Billion Round at a $71 Billion Pre-Money Valuation

DeepSeek, the Chinese AI lab that built the R1 reasoning model on a training budget that rattled U.S. frontier labs in 2024, has told a group of potential investors not to sign the second-round term sheets it had expected to close within days.

The pause, first reported by Bloomberg and carried into Chinese-language press by QbitAI, 联合早报 (Lianhe Zaobao), and Investing.com 中文, sits on a round the lab had marketed at roughly 480 billion yuan (about $71 billion) pre-money and at least 10 billion yuan (about $1.5 billion) in size. A pre-money valuation is the company's implied worth before the new cash is added, so a $71 billion pre-money means the new investors are pricing DeepSeek above that figure once the raise closes. Bloomberg attributed the move to people familiar with the talks; DeepSeek, Tencent, CATL, and the National AI Industry Investment Fund have not responded to requests for comment.

The trigger, according to Bloomberg, was not a shortage of demand. It was a leak. In the days before the signing window, an internal memo the lab treated as confidential became public, and the friction that followed during investor outreach was severe enough that DeepSeek asked counterparties to hold off.

That causation is what turns a deal-ticker note into a story. "Trust issues during outreach" is Bloomberg's framing of the anonymous-sourced account, not a statement from the lab. The memo's contents are not on the record. Who leaked it is not on the record. Only the leak itself, and the pause that followed, are.

A second round at this scale, on top of the lab's first external close, would be unusual capital structure for a Chinese AI company. DeepSeek closed its first external round in June 2026 at roughly 500 billion yuan (about $74 billion), with founder Liang Wenfeng committing around 20 billion yuan of his own capital (about $3 billion), Tencent putting in about 10 billion yuan (about $1.5 billion), and CATL contributing roughly 5 billion yuan (about $740 million). The National AI Industry Investment Fund also participated. Pre-money to pre-money, that is a roughly 40% jump in implied value between the June close and the planned second signing, inside weeks.

That jump is what the leak exposed to scrutiny. Anonymous sources told Bloomberg the pause was not driven by investors walking away, by a valuation standoff, or by weak demand. If that framing holds, the story is not that DeepSeek could not find the money. The story is that a frontier-model lab moving into public-market territory chose to delay on a trust basis rather than sign on contested terms.

The pause also lands on top of an IPO file Bloomberg says DeepSeek could submit as early as this year. Liang Wenfeng had previously funded the lab mainly through his quantitative hedge fund High-Flyer, and DeepSeek only began bringing in outside capital at scale in 2026. A second close at a $71 billion pre-money would re-price the company's path to public markets and any related-party transactions the lab would have to disclose in a prospectus. The signing-window slippage, on top of an uncertain filing date, gives the IPO timing itself a moving target.

The reporting is single-sourced through anonymous channels. Bloomberg's piece is the underlying original; the three Chinese-language outlets all retransmitted the same wire. There is no public confirmation of who leaked the memo, what it contained, or whether the pause will lift on the same terms. The conditionality is itself part of the news. The round is paused, not cancelled, and a "no" today does not foreclose a re-signed term sheet on different guardrails.

If it re-signs on the marketed terms, the trust friction reads as a negotiation pause. If the signing window slides past the IPO filing window, the friction becomes a re-pricing event for the public-market debut. Either way, the operative spine is governance and trust between a frontier lab and its capital partners, in a market where the most aggressive raises of 2026 are running on relationships that can leak. The open questions are concrete: does the term sheet re-sign on the marketed terms before any IPO filing, and does the lab disclose the memo leak, or its origin, in any prospectus it submits?