Liang Wenfeng's private remarks about DeepSeek's dependence on Nvidia chips and China's lag behind the United States spread across Chinese social media. Days later he verbally told backers that the second funding round, targeting a pre-money valuation of at least 480 billion yuan, was suspended.

Liang Wenfeng was talking to investors, in a room, with the door closed.

That is the setting for every candid thing a founder ever says: the private meeting, the small audience, the assumption that nobody is writing it down. What Liang said in that room, according to Chinese media reports of a transcript, was that DeepSeek still leans heavily on Nvidia's chips, and that China remains behind the United States in the sophistication of its AI.

Neither statement is exotic. Both are widely believed by people who work in the field. One of them is the reason American export controls exist.

The transcript went online anyway. Within days, the man who had just become the richest AI model creator in the world picked up the phone and told prospective investors that the funding round they expected to sign was suspended.

Bloomberg reported the pause on Saturday, July 25. DeepSeek did not immediately respond to Bloomberg's emailed request for comment about the transcript or the fundraising.

The Round Was Days From Signing

DeepSeek verbally informed some of its would-be backers that they would not be signing investment agreements in the coming days as they had expected, according to people familiar with the matter cited by Bloomberg.

The suspension stemmed in part from Liang's frustration over online reports about his comments to investors during the first financing deal. Bloomberg has not verified the authenticity of the posts, which concerned a transcript of a meeting Liang held with unidentified parties. Chinese outlets including Yicai reported the contents in the days before the pause.

The second round was not small. DeepSeek was seeking at least 10 billion yuan in fresh capital.

The two rounds, side by side

First round, closed June 2026: roughly 7 billion dollars raised, at a valuation of about 50 billion dollars. Backers included Tencent, battery maker CATL, and China's National Artificial Intelligence Industry Investment Fund. It set a record for first-time financing by a Chinese tech startup.

Second round, suspended July 2026: at least 10 billion yuan sought, at a pre-money valuation of at least 480 billion yuan, roughly 71 billion dollars. Talks began weeks after the first round closed.

Negotiations remain fluid and the company may still proceed. It is not clear whether DeepSeek has communicated its intentions to every prospective investor in the deal.

The Leak Landed on the Most Sensitive Ground Available

Beijing has promoted a specific narrative for the better part of two years: Chinese AI labs are closing the gap with American ones, and export controls have failed to stop them.

Liang's reported remarks cut against both halves of that narrative from the single most credible source available, which is the founder of the company most often cited as proof of it. DeepSeek's breakthrough model showed that a Chinese lab could build a frontier-class system with fewer computing resources than its American rivals, and that demonstration became the centerpiece of the argument that restrictions were not working.

A founder saying privately that his company still depends on Nvidia hardware is not a scandal. It is an inconvenience with a very large audience.

DeepSeek has been designing its own inference chip to reduce that dependence. That effort is ongoing, and it is the practical answer to the problem the transcript described.

The Money Is Not the Reason

The instinct is to read a suspended round as a valuation problem. The reporting does not support that reading.

Investor appetite was the opposite of scarce. DeepSeek closed a record first round in June, drew state-backed capital alongside Tencent and CATL, and opened a follow-on at a pre-money valuation roughly 40% above it within weeks. Liang's own net worth more than doubled after the first round, to 36 billion dollars from about 16.7 billion, according to the Bloomberg Billionaires Index. That puts him above Anthropic co-founder Dario Amodei and OpenAI's Greg Brockman among creators of AI models.

The company also has an exit in motion. DeepSeek has begun preparations for an initial public offering and may file as soon as this year.

What appears to have stopped the round is that a private conversation stopped being private. Founders who are raising capital tell investors things they do not tell the market. When the wall between those two audiences comes down, the immediate rational move is to stop talking until you understand who is listening.

The Other Reading: This Is About Research, Not Investors

There is a second interpretation, and DeepSeek's own behavior supports it.

The company's senior management has told potential investors that it will prioritize frontier AI research over short-term commercialization. Liang pledged in at least one investor meeting to keep developing open-source models while pursuing artificial general intelligence.

That is an unusual pitch. Most labs raising at a valuation above 70 billion dollars are describing a revenue path. DeepSeek is describing a research agenda, and pairing it with aggressive pricing and free weights, a combination that has won it an enormous developer following while making the economics of a large private round genuinely harder to underwrite.

Read that way, the pause is less about a leaked transcript and more about a founder deciding he does not need the money badly enough to accept the terms, the scrutiny, or the loss of control that comes with it. Whether the interruption is brief or structural depends on how quickly the controversy fades.

Why a Private Chinese Funding Round Matters to Practitioners

DeepSeek is not an abstraction to the people who build with these models. It is a line in a config file.

Teams have been moving real production traffic onto DeepSeek models on price and latency grounds, a shift we documented when one startup moved 100% of its AI traffic from Claude to DeepSeek. The company's engineering output has been consistently practical, including the release that made its models 85% faster with no new hardware.

Three things follow from this week.

The funding pause is a roadmap signal. DeepSeek said it was raising to expand computing capacity. A delayed round is a delayed capacity plan, and capacity plans determine how aggressively a lab can keep pricing inference.
The IPO changes the disclosure regime. A listed DeepSeek publishes financials, compute commitments, and risk factors. Right now the only visibility into its chip supply comes from leaked transcripts.
The chip dependency is confirmed from the inside. Anyone building a supply-chain assumption around Chinese labs achieving hardware independence should treat the founder's private assessment as the most reliable data point available.

The Bottom Line

A founder said in a closed room that his company depends on American chips and that his country is behind. Someone published it. He paused a round worth billions of dollars.

Nothing in that sequence is about technology. It is about a lab that has spent two years being held up as evidence in an argument between two governments, and a founder who appears to have decided that being evidence is more expensive than being funded.

The models will keep shipping. The weights will keep being free. What just got harder is knowing anything true about the company that makes them, because the last person who spoke honestly about DeepSeek in private is unlikely to do it again.