Capital

A Leaked Call Paused a 480 Billion Yuan Round

DeepSeek suspended its second funding round on 25 July after its founder's remarks leaked. What he said matters more to a European buyer than the pause does.

CapitalBy Servola Tech Desk2026-07-263 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

A Leaked Call Paused a 480 Billion Yuan Round

Key takeaways

DeepSeek halted a round targeting a pre-money valuation of at least 480 billion yuan not because the numbers moved but because its founder's private remarks became public, which means the disclosure risk in your AI supply chain now sits in what a vendor says off the record rather than in what it publishes.
Read the content of the leak rather than the fact of it: the founder is reported to have described dependence on Nvidia chips and acknowledged that China still trails the United States, and neither statement appears on any model card you can download.
Treat a vendor that suspends a financing to control a narrative as a vendor with a defined information posture, and assume the same posture will apply during an outage or a security incident.
Ask any model supplier, in writing, which accelerators its training and inference depend on and where those are physically hosted, because that single answer is what a leaked call just revealed for free.

The pause was about candour, not valuation

On Saturday 25 July, DeepSeek verbally told prospective investors it would not be signing investment agreements in the coming days, suspending a second fundraising round. The company had been targeting a pre-money valuation of at least 480 billion yuan. Its first round closed in June and raised about 7 billion dollars.

Why it matters: the trigger reported by Bloomberg and echoed across Fortune and other outlets was not a due diligence problem or a market move. It was that remarks attributed to founder Liang Wenfeng, made to investors during the earlier financing, circulated online. He is reported to have discussed the company's reliance on Nvidia chips and to have acknowledged that China still trails the United States in AI sophistication.

Both of those are unremarkable as analysis. Plenty of observers have said the same. What made them consequential is who said them and in what room. A frank assessment from a founder to his own investors carries weight that the identical sentence from an analyst does not, which is precisely why it moved.

The leak is more useful to you than the round ever was

If your organisation is evaluating or already running this model family, the funding round is close to irrelevant. The two disclosed statements are not.

The first is a hardware dependency. A lab that depends on a specific American accelerator inherits every export control, allocation queue and supply constraint attached to it. That is a continuity question for anyone building on the model, and it is not answered anywhere in a model card, a licence or a pricing page.

The bottom line: the second is a candid capability ranking from the person best placed to know. You are entitled to weigh a founder's private assessment of his own model against the benchmark table on his website, and where those two disagree, the private one is the better input to a procurement decision.

What to put in writing before you standardise on any model

Ask which silicon, and where. One question, in writing: which accelerators do your training and inference depend on, and in which jurisdictions are they hosted. A vendor that will not answer has told you something. A vendor that answers has given you the input you need to model an export-control or allocation shock.

Ask what happens if that supply is interrupted. Not whether it might be, but what the documented fallback is, and whether service levels survive it. This is the same question you would ask a component supplier, and there is no reason to ask a model supplier a softer version of it.

Watch how a vendor behaves when information escapes. Suspending a financing round to regain control of a narrative is a legitimate commercial decision, and it is also a data point about disclosure culture. The firm that goes quiet when a private remark leaks is the firm that will go quiet when a model degrades or a breach occurs. Price that into how much of your stack you are willing to make dependent on it.

Frequently asked questions

Does a paused funding round mean the model is at risk?

No, and it would be a mistake to read it that way. The company closed roughly 7 billion dollars in June and reporting indicates it may resume the process later. The operational signal here is about information handling and hardware dependency, not about solvency.

Why does the Nvidia dependency matter if the model works today?

Because it determines who else has to agree for it to keep working. Accelerator supply is subject to export policy, allocation between very large buyers, and manufacturing constraints. A model that runs perfectly today can still face a capacity or cost change driven entirely by decisions made outside its vendor.

Is it fair to weigh leaked remarks in a procurement decision?

Weigh them as reporting, with appropriate caution about attribution, and never as the sole basis for a decision. What makes them useful is that they were not written for you. Marketing material is authored to persuade a buyer, and a candid internal assessment is not, which is exactly what gives it evidential value.

The round can be restarted whenever the company chooses. The two sentences that stopped it cannot be withdrawn, and they told a prospective customer more than any datasheet on the site.

DeepSeekLiang WenfengNvidiaAI fundingmodel dependencyexport controlsvendor risk

Nvidia Now Backs the Voice AI It Also Powers

Gradium, the Paris voice-AI startup spun out of Kyutai, extended its seed round past 100 million dollars with Nvidia joining as a backer. When your chip supplier also owns the voice vendors, neutral infrastructure starts to blur.

3 min read

Capital

Paradigm's $1.2 Billion Fund Takes Crypto Money Into AI and Robotics

Paradigm closed a 1.2 billion dollar fund on July 8, 2026, widening from crypto into AI and robotics. What the close means for founders raising in 2026.

4 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.