DeepSeek Raises $7.4 Billion in Its First Funding Round at a $59 Billion Valuation

For two years, DeepSeek's defining boast was that it did not need anyone's money. The Hangzhou lab that rattled Silicon Valley with cut-rate, open-weight models was bankrolled almost entirely by founder Liang Wenfeng and his quantitative hedge fund, High-Flyer — no venture capital, no cloud-giant subsidies, no board to answer to. That era is over. DeepSeek has closed its first-ever external funding round, raising more than 50 billion yuan, roughly $7.4 billion, at a valuation that people familiar with the deal put between 350 billion and 400 billion yuan, or about $52 billion to $59 billion.

The raise, first reported by Reuters in early June and confirmed as closed around June 16 by The Information, instantly makes DeepSeek China's most valuable AI startup and ranks among the largest private technology financings the country has ever seen. But the more revealing story is not the number. It is the unusual architecture Liang built around it — a structure designed so that taking billions in outside capital costs him almost none of his control.

Who is in, and how much

The cap table is a who's who of corporate China. Liang himself is contributing roughly 20 billion yuan — close to 40% of the entire round — a strikingly large founder commitment that underlines how much personal conviction he has staked on the company's trajectory. Tencent is in for about 10 billion yuan and battery giant CATL for about 5 billion yuan, making the two China's largest external backers in the deal. Rounding out the late-stage discussions, according to people familiar with the talks, were China's national AI fund, gaming company NetEase, e-commerce group JD.com, Hong Kong-based IDG Capital and Monolith Management.

CATL's presence is the most telling addition. The world's dominant EV-battery maker has been pushing into energy storage and power solutions for data centers — and as AI workloads devour more electricity, reliable power has become a strategic asset in its own right. Tencent, meanwhile, has poured money into its own Hunyuan model but trails domestic leaders such as ByteDance's Doubao; a closer tie to DeepSeek buys it a seat near the front of China's AI race.

The structure nobody had seen before

What sets this round apart is governance. Commercial investors were not handed direct equity in DeepSeek at all. Instead, they were required to route their capital into a limited partnership managed directly by Liang — an arrangement that strips them of voting rights and locks their money up for five years. The lone exception is the state. China's National Artificial Intelligence Industry Investment Fund, often called the "Big Fund," received direct corporate ownership, full voting privileges and freedom from the lock-up, bypassing the limited-partnership wrapper entirely.

The practical upshot, as the Austrian outlet Trending Topics put it, is that if DeepSeek's commercial backers and its government investor ever disagreed about the company's direction, "only one party would have a formal voice." Liang reportedly also personally vetted the identities behind each investing fund, ensuring unknown — particularly foreign — capital could not slip onto the register.

That design leaves DeepSeek looking, as Tech Funding News observed, "less like a startup and more like a national strategic project," with "no external board oversight, no investor pressure, and a state fund holding a privileged stake."

A smaller war chest, a different war

The contrast with America's AI champions is stark. Anthropic is valued at roughly $965 billion after a $65 billion round; OpenAI closed $122 billion at an $852 billion valuation and is preparing for a public listing. DeepSeek's $7.4 billion is a rounding error beside those figures — but it is fighting a different war, on a different budget, by necessity.

"Western export bans mean DeepSeek cannot access frontier American silicon," said Alfredo Montufar-Helu, managing director at Ankura China Advisors in Beijing. "Without the ability to buy that hardware, they have no reason to match the multi-billion-dollar computing budgets of their U.S. rivals."

That constraint is precisely what produced DeepSeek's signature strategy. By releasing V3 and R1 with open weights and driving API prices toward zero margins, the company commoditized the model layer itself, forcing the whole market to compete on cost efficiency rather than raw scale. The new capital, the company has signaled, will fund next-generation models, engineering talent and the specialized domestic hardware needed for a sustained push toward artificial general intelligence — including its V4 model, unveiled in April, which independent evaluations rank among the strongest open-source offerings even as it trails the frontier.

What this means — and what to watch

DeepSeek's first outside raise is a milestone for China's AI sector on two fronts. It confirms that Beijing can assemble a domestic capital-and-supply-chain stack — strategic investors, a state fund, homegrown power and silicon — capable of underwriting a frontier lab without Western money or chips. And it doubles down on the open-weight playbook as a geopolitical instrument: free model weights are cheap to distribute, hard to sanction, and corrosive to rivals' margins.

The open question is whether Liang's iron grip becomes an asset or a ceiling. A founder-controlled limited partnership with a privileged state shareholder can move fast and resist outside pressure. It can also struggle to attract the kind of patient, deep-pocketed capital that near-trillion-dollar rivals command — and it tethers a commercially ambitious company to the priorities of the Chinese state.

Three things to watch from here: whether DeepSeek can convert this capital into a V4 successor that genuinely closes the gap with frontier U.S. and Chinese models; whether the limited-partnership structure becomes a template other Chinese labs copy; and whether the privileged role of the "Big Fund" complicates DeepSeek's adoption abroad, where enterprise buyers may think twice about building on a model whose only voting shareholder is the Chinese government.