DeepSeek Pauses $8B Funding Round Amid Founder Furor

What happened

DeepSeek, the Hangzhou-based AI lab that stunned Silicon Valley last year with efficient, low-cost models, told prospective investors in its second fundraising round that the deal was being suspended for now, Bloomberg News reported, a story picked up and syndicated by Reuters, Fortune, MarketScreener and U.S. News within hours 1791112. According to people familiar with the matter, DeepSeek verbally informed some backers that they would not be signing investment agreements in the coming days as expected, though the company left open the possibility of resuming talks later 179. Reuters, citing Bloomberg, said it could not independently verify the report 1912.

The timing points to a specific trigger: an unverified transcript, widely circulated on Chinese social media and X, purporting to capture comments founder Liang Wenfeng made to investors during the company's first funding round 7810. Chinese outlets including Yicai reported that Liang discussed DeepSeek's reliance on Nvidia chips and China's continuing lag behind the U.S. in AI sophistication 7811. Bloomberg said it had not authenticated the transcript, and DeepSeek did not respond to requests for comment 7811. The Information's account converges on the same explanation, tying the pause to the transcript's viral spread rather than to any breakdown in deal terms 10.

The money at stake

The numbers moved as the story developed, reflecting a negotiation in motion rather than a signed term sheet. DeepSeek's first round, which closed in June, raised roughly 50 billion yuan (about $7.4 billion) at a valuation Reuters initially pegged between $52 billion and $59 billion, with Tencent and CATL emerging as the largest external backers alongside China's National AI Industry Investment Fund, NetEase and JD.com 1314. That structure was unusual: most investors put capital into a limited partnership controlled by Liang rather than taking direct equity, accepted a five-year lock-up and had no voting rights — except the state-backed national AI fund, which invested directly with full voting rights and no lock-up 13.

The paused second round was targeting at least 10 billion yuan (roughly $1.5 billion) in fresh capital at a pre-money valuation of at least 480 billion yuan, or about $71 billion, according to Bloomberg's earlier reporting, later described as closer to 500 billion yuan, or $74 billion 789101112. A separate, indirect data point complicates the picture: disclosure filings from listed luggage maker Anhui Korrun implied a DeepSeek enterprise value of only about $5.2 billion based on a small stake purchased through an investment fund — a figure wildly out of step with the financing reports and likely reflecting a different valuation basis or vehicle 15.

The pause did not last. By early August, reporting indicated DeepSeek had reopened the round, seeking close to $8 billion at roughly the same $74 billion valuation, with Monolith Management — an early Moonshot AI backer — in talks to join a syndicate that also included NetEase, JD.com, IDG Capital and others 1617. By late August, the Wall Street Journal and South China Morning Post reported the financing was nearing completion at about 50 billion yuan ($7.4 billion) on a $74 billion pre-money valuation, with CATL, Shixiang Capital and Monolith participating and new names — CPE, Legend Capital, Stony Creek Capital — in discussions 18.

Why it matters beyond one company

DeepSeek's trajectory is unusual among AI unicorns. It grew out of High-Flyer, the quantitative hedge fund Liang co-founded, which financed the lab's early research without conventional venture capital 181920. Its V3 and R1 models, released in 2025, demonstrated competitive performance at a fraction of the compute cost associated with U.S. frontier labs, briefly displacing ChatGPT atop Apple's App Store and rattling Nvidia's stock 1920. That efficiency narrative is precisely what makes the current fundraising sensitive: reported comments attributed to Liang acknowledging continued reliance on Nvidia chips and a persistent gap with the U.S. cut against the story that made DeepSeek famous 7811.

The stakes are compounded by an approaching IPO. Reuters reported DeepSeek had begun early deliberations on a listing on Shanghai's STAR Market, and later reporting pointed to a possible 2026 filing with a 2027 debut 91218. A pre-IPO round priced near $74 billion sets a valuation benchmark investors and regulators will scrutinize once the company approaches public markets — making any governance or information-control stumble, such as a leaked investor transcript, more consequential than it might be for a private company with no listing plans.

The capital need is also structural. DeepSeek reportedly plans to add roughly a gigawatt of computing capacity, and its senior management has told investors the company will prioritize frontier research and open-source models over near-term commercialization while continuing to pursue artificial general intelligence 1118. That framing may appeal to strategic and state-linked backers hunting for exposure to a national AI champion, but it sits awkwardly with the venture-capital expectation of a fast path to revenue.

Where the reporting agrees

Across Bloomberg's original reporting and its syndication by Reuters, Fortune, MarketScreener, U.S. News and the Business Times, the core facts are consistent: DeepSeek paused its second-round talks, did so verbally rather than through a public statement, left the door open to resuming, and tied the decision at least in part to Liang's frustration over leaked comments from the first round 17891112. The Information's independent reporting reinforces the same causal story — a viral transcript, not a valuation dispute or investor pullback, triggered the halt 10. Multiple outlets also agree the round subsequently resumed and moved toward a close in the $74 billion valuation range by late August 161718.

Where it doesn't

The clearest divergence is in the valuation and raise figures over time, though this reflects an evolving negotiation rather than a factual contradiction: Bloomberg's mid-July figures cited at least 480 billion yuan pre-money and at least 10 billion yuan raised, while later Bloomberg-sourced reporting in August described a $74 billion valuation and an $8 billion target, and by late August the Journal and SCMP described a $7.4 billion raise closing at that same $74 billion mark 78910161718. AI Weekly's account is notably more skeptical in tone than the wire coverage, explicitly flagging that the Monolith participation, the $8 billion figure and the late-August signing timeline are single-sourced to Bloomberg and framed as reporting rather than confirmed disclosure 17 — a caveat other outlets state more implicitly or not at all.

The Anhui Korrun disclosure is the sharpest outlier: its implied $5.2 billion valuation is not reconciled with any other figure in the coverage and is presented by BigGo as a curiosity rather than a competing authoritative estimate 15. No other outlet treats it as credible against the $50 billion-plus figures from Reuters and Bloomberg's primary sourcing.

The most defensible reading

The weight of corroborating, independently sourced reporting — Bloomberg's own account plus The Information's separate confirmation — supports treating the pause as real but narrow: a temporary, reputation-driven halt rather than a collapse of investor demand. The subsequent resumption of talks at a similar or higher valuation, reported by multiple outlets weeks later, backs that reading over any interpretation suggesting DeepSeek's fundraising had genuinely stalled. The valuation figures should be read as snapshots of a moving negotiation rather than conflicting facts, with the Anhui Korrun number best understood as measuring something other than the headline financing round.