DeepSeek Suspends Fundraising Round After Founder's Leaked Remarks Expose the Compute Gap

DeepSeek told its prospective investors to hold off this week. Verbally, on July 25, the Chinese AI lab informed some backers they would not be signing investment agreements in the coming days as expected, suspending a round that was targeting a pre-money valuation of at least 480 billion yuan, roughly $71 billion 1.

The trigger was not a failed audit or a regulatory blockade. It was a leak. Days before the pause, remarks widely attributed to founder Liang Wenfeng from a roughly four-hour investor meeting had spread through Chinese tech and investor circles 2. Bloomberg, which first reported the suspension, has not verified the transcript's authenticity 1.

What makes the freeze remarkable is who DeepSeek is. This is a lab that built its global reputation on open-weight model releases, letting researchers worldwide download, inspect, and build on its work. When DeepSeek released its V3 model, it demonstrated that cutting-edge AI could be trained with a fraction of the computing resources Western labs were pouring into their own models 1. That transparency became DeepSeek's sharpest competitive weapon. Now the same instinct for openness has turned inward: private remarks from the company's founder, remarks not meant for public consumption, went viral and cost the company control of its own fundraising timeline.

The suspension reportedly stemmed in part from Liang's frustration over the spread of his comments 1. Liang, whose net worth sits at $36 billion after the first round 1, reportedly holds the large majority of DeepSeek's equity and nearly all of its voting rights, meaning the decision to halt was effectively his alone 2. The backers wiring in billions had little say over the timing.

Then there is what the leaked transcript actually says, unverified as it remains. Chinese media outlet Yicai reported that Liang discussed DeepSeek's reliance on Nvidia Corp. chips and China's persistent lag in AI sophistication relative to the US 1. According to quotes attributed to Liang in the transcript and circulated online, the gap between the two countries comes down to computing resources, not talent. The people are the same, he reportedly said; the silicon is not 3.

The same attributed remarks make the compute gap look structural rather than a simple quantity problem. According to the transcript circulated online, Liang said that during V3 training, DeepSeek still used Nvidia GPUs but had already abandoned Nvidia's software ecosystem, meaning the CUDA software stack that locks most of the world's AI developers into Nvidia hardware was something his engineers had worked around 3. The implication is that China's disadvantage is not an engineering problem DeepSeek's talent can solve at a whiteboard. It is a procurement problem: they can write their own software stack, but they still need Nvidia's physical chips, and US export controls limit how many reach Chinese shores.

This is the tension underneath the fundraising pause. DeepSeek closed $7 billion in its first outside round in June 2026, drawing backers including Tencent Holdings Ltd., Contemporary Amperex Technology Co. Ltd., and the state-backed National Artificial Intelligence Industry Investment Fund 1. Founded in 2023 and owned by hedge fund Zhejiang High-Flyer Asset Management, the company had been preparing for an initial public offering that could file as soon as this year 1. If a single leaked document can freeze a process aimed at one of China's most ambitious tech valuations, that valuation has not been tested by anything harder than investor enthusiasm. The market is pricing confidence in DeepSeek's efficiency story. It has not priced the risk that a founder willing to halt billions in commitments over how his private words got repackaged might respond the same way to a difficult quarter.

DeepSeek may still resume the round 1. But the episode already exposes something the term sheets did not: a founder who pledged to keep developing open-weight models as a matter of strategy 1 is also one who will shut the door the moment transparency reaches him instead of his competitors.