China's Investors Just Handed DeepSeek $7.4 Billion — And Got No Vote, No Exit, and No Say for Five Years

For two and a half years, DeepSeek did not take a cent of outside money. The lab that shocked the industry with V3 and R1, and again this March with a trillion-parameter V4, was funded entirely by its founder's hedge fund. That ended in June.

DeepSeek closed its first external round at more than 50 billion yuan — roughly $7.4 billion — at a valuation north of $50 billion. It is the largest fundraising in Chinese AI history and makes DeepSeek the country's most valuable AI startup.

The number is not the interesting part. The term sheet is.

Money Without a Vote

Almost every investor in this round did not buy shares in DeepSeek. They bought into a limited partnership controlled by founder Liang Wenfeng. That vehicle carries two conditions: no voting rights, and a five-year lock-up before anyone can sell.

Read that again in Silicon Valley terms. Tencent wired well over a billion dollars into a company and received no board seat, no vote, and no way out until 2031.

Who agreed to that? Liang himself put in around 20 billion yuan — about $3 billion of his own money, the largest single cheque in the round. Tencent is reported at roughly 10 billion yuan (about $1.4 billion). And in the detail that says the most about where Chinese AI is heading, the battery giant CATL put in around 5 billion yuan, close to $700 million. A company that makes electric-vehicle batteries is now a top-three shareholder in a frontier AI lab.

The One Investor Who Got Different Terms

Exactly one participant did not go through Liang's partnership. China's state-backed National AI Industry Investment Fund invested directly into DeepSeek — and took voting rights and no lock-up with it.

Nothing about that is illegal or even hidden. But it is a remarkably clear statement of hierarchy. In the largest AI financing China has ever done, private capital accepted terms that strip it of influence, and state capital did not. Tencent, one of the most powerful technology companies on the planet, sits below a government fund in the same cap table.

For anyone trying to work out how Chinese AI labs make decisions, that is a more useful document than any benchmark score.

What the Money Is Actually For

DeepSeek has not been short of cash. Liang built the lab in July 2023 on the balance sheet of his quantitative hedge fund, High-Flyer, and it stayed self-financed through every release that made it famous. Its problem was never funding. He said so himself, back before anyone outside China had heard of the company.

That is the sentence to hold onto. Raising $7.4 billion does not solve a chip embargo — but it does fund the domestic alternative to one, which is where reporting says the money is pointed: home-grown computing infrastructure rather than smuggled accelerators.

It is worth keeping the scale honest, though. Even at $50 billion, DeepSeek is a fraction of its American rivals. Forbes put Anthropic at roughly $965 billion and OpenAI at about $852 billion. China's most valuable AI company is worth around 5% of either. The gap in capability has narrowed far faster than the gap in capital.

Moonshot Closed Higher Than the Rumours

DeepSeek is not alone. On July 29, days after shipping Kimi K3, Moonshot AI closed $3.5 billion at a $35 billion valuation. The company had gone out looking for $1 billion to $2 billion.

When we covered K3 last week, the reporting had Moonshot raising at up to $31.5 billion. It closed above that. And the revenue behind it is moving in the same direction: annual recurring revenue went from about $200 million in April to more than $300 million by June.

Moonshot is already talking to investors about a further round at a $50 billion pre-money valuation, ahead of a possible Hong Kong listing. That would be a valuation increase of more than 40% in a matter of months, on a company whose flagship product you can download for free.

The Rest of the Money Went to Robots

Frontier models got the headlines. They did not get most of the capital.

Chinese robotics startups raised $5.6 billion across 176 deals in the first four and a half months of 2026 — more than the $4.3 billion the sector took in all of 2025, and level with the previous record year of 2021 before the year was half over. Counted more broadly, across all of embodied AI, first-half funding reached roughly 93.5 billion yuan (about $13 billion) over 322 deals, several times the year-earlier figure.

These are not seed cheques with a demo video attached.

A $513 million seed round is not a typo. And unlike the 2021 boom, there are exits attached: Unitree filed for a Shanghai listing in March targeting a $3 billion to $7 billion valuation, and Robotphoenix listed in Hong Kong on May 18, raising about $86 million and closing its first day up 80%.

The strategic logic is not subtle. Washington can restrict what chips enter China. It has far less leverage over robot arms, warehouse automation and humanoid platforms — a field where manufacturing capacity, not export licences, decides the winner.

And the Chips Are Going Public

The third front is silicon, and it has moved from venture funding to public markets. Biren listed in Hong Kong in January. Moore Threads filed for a Shanghai IPO in July seeking about 8 billion yuan — roughly $1.1 billion — to fund domestic GPU development. Cambricon, Moore Threads, MetaX, Biren and Iluvatar CoreX are all now valued above $6 billion.

Cambricon alone plans to more than triple output in 2026, targeting around half a million AI accelerators including as many as 300,000 units of its most advanced Siyuan 590 and 690 parts.

The ceiling is manufacturing, not money. SMIC, which fabricates most of these designs, cannot expand capacity much further while it is barred from importing critical equipment and spare parts. Capital can fund a chip design. It cannot buy a lithography machine that no one is allowed to sell you.

What to Take From This

• Read the terms, not the totals. A $7.4 billion round where the money gets no vote is a different event from a $7.4 billion round where it does. Headline numbers make Chinese and American AI financing look comparable. The governance does not.
• The state took the only seat that mattered. One investor negotiated voting rights and no lock-up, and it was a government fund. That is the clearest public signal yet about who has influence over China's leading AI lab.
• Patient capital behaves differently. A five-year lock-up removes the pressure to show quarterly progress. It also removes the discipline that comes with it. Expect longer bets and less course correction.
• Free models, expensive companies. Moonshot gives K3 away and is being valued at $35 billion heading toward $50 billion. Open weights are a distribution strategy, not a business model — and investors are pricing them accordingly.
• Watch the robots. More Chinese capital went into embodied AI in six months than into either headline model lab. If you only track chatbot releases, you are watching the smaller half of the story.