Who is Funding DeepSeek? Inside the $45B State Deal

DeepSeek spent two years insisting it had no need for outside money, flushed with cash from its parent hedge fund High Flyer Capital. That position has now shifted, with the company raising external funds for the first time at a reported valuation of $45–50 billion.

The China Integrated Circuit Industry Investment Fund (ICF), the state-backed AI fund, is in talks to lead the round, along with Tencent and Alibaba. 

But the more telling detail is who DeepSeek actively shut out—HSG (formerly Sequoia China) and Hillhouse, a Singapore-headquartered VC firm.

That puts Chinese startups in sharp contrast to their American counterparts. Chinese labs are built on state-aligned capital, while their US peers chase valuations on the back of private VC cycles.

Why the State Capital Moved In

Reports suggest that the National AI Industry Investment Fund, an AI affiliate of the ICF, is negotiating to lead the $7.35 billion (¥50 billion) round for DeepSeek. 

Tencent is contributing $833 million (¥6 billion) for roughly 2% equity, and CEO Liang Wenfeng will put in $2.8 billion (¥20 billion) of his own money, representing 40% of the round, according to Bloomberg. 

DeepSeek’s funding structure is not an isolated case. Chinese AI companies tend to follow a hybrid model of state and private capital. 

StepFun’s $717 million Series B was co-led by Shanghai's state investment arm alongside Tencent. Zhipu and MiniMax both had Shanghai state capital vehicles on their cap tables before their Hong Kong IPOs. Recently, Moonshot AI, the makers of Kimi AI models, secured roughly $2 billion led by quick commerce super app Meituan's venture arm, alongside state-backed entities like China Mobile.

A Stanford research from 2024 found that 71% of Chinese AI firms received both government and private VC funding. But this hybrid structure is also symptomatic of private venture capital in China drying up. 

“By 2024, private VC fundraising for China was the lowest in nearly a decade, and 2025 fell to less than a quarter of that. Government capital filled the vacuum,” tells Dermot McGrath, Founder and CEO of Chinese frontier tech company ZenGen Labs and Co-founder of crypto VC firm Ryze Labs, in an interaction with AIM. 

Several factors drove this contraction. 

In 2020, Ant Group's record IPO was paused after regulators required it to hold more capital against the consumer loans it was originating at scale, even though it was not regulated as a lender. In 2022, DiDi Global was investigated days after listing in New York, despite being told to wait for a data security review, because sensitive location data on hundreds of millions of citizens would have become subject to US audit and disclosure requirements. 

“Each action had specific regulatory logic, but the cumulative effect on foreign investor confidence was severe. It was seen as a sweeping ‘tech crackdown’. Overseas LPs [Limited Partners] pulled back,” adds McGrath.

The Meta-Manus episode was the latest showdown. 

Beijing blocked Meta's acquisition of the Singapore-based, Chinese-origin startup Manus, citing unauthorised technology transfer from China to the US. US export controls simultaneously made it legally risky for American VCs to invest in Chinese companies. 

Even before the Meta bid, US firm Benchmark had led a $75 million funding round for Manus and, despite the company being headquartered in Singapore, attracted regulatory scrutiny from the US government.

For private investors in China, state intervention has further constrained the already narrow exit paths. 

“Domestically, China's securities regulator slowed A-share IPO approvals because too many low-quality companies were getting listed and the flood of new supply was dragging down secondary market valuations,” says McGrath. 

Moreover, A-share IPOs in China numbered over 400 in 2022 but fell to just 100 in 2024. “The knock-on effect was breaking the exit flywheel that the VC model depends on. No exits, no returns to LPs, no successor funds,” he remarks.

Government capital has moved into that vacuum. A report from the Rhodium Group states that by 2025, government guidance funds and state-owned enterprises together accounted for 81% of LP commitments into Chinese VC and PE funds, up from 65% in 2019, and state-backed institutions participated in nearly half of all primary market deals. 

What State Capital Actually Buys

The state capital operates through two tracks. 

The first is direct investment, for example, the National AI Industry Investment Fund carries an $8.3 billion (¥60 billion) commitment over a 13-year horizon, and Beijing maintains a separate $13.9 billion (¥100 billion) AI fund with a 15-year horizon. 

“These are not structured like venture capital. They're patient capital with policy mandates to build capability, not optimise for 3–5 year returns,” says McGrath. “For Beijing, building domestic capability in semiconductors, AI, and robotics isn't optional.”

The second track is procurement. The State-Owned Assets Supervision and Administration Commission launched an ‘AI+’ initiative in 2024, directing state-owned enterprises to integrate AI across their operations. 

McGrath also states that government and enterprise AI contracts reached over $4.2 billion (¥30 billion) across 7,500 projects in 2025, up from $900 million (¥6.5 billion) across 1,520 projects in 2024. Tech giants like iFlytek led with 210 contracts worth $319 million (¥2.3 billion) and Baidu with 110 worth $124 million (¥890 million).

“That money flows primarily to established players with government delivery infrastructure,” he shares. “The frontier startups benefit more from the investment track: equity, compute access, and IPO backing.”

According to its public financial records, Zhipu is bridging both tracks, with roughly 74% of its 2025 revenue coming from localised government and enterprise deployments, as well as its frontier model work.

When Chinese AI labs need capital, compute, and customers, state capital can deliver all three. Foreign capital is welcome in Chinese AI startups, McGrath says, as long as the rules are followed and China retains its share of the returns. “Nobody is being told what to build or how to run their company. The logic is: keep the technology in China, let Chinese investors capture the upside, and build towards self-reliance.” 

He points to the Meta-Manus block, where state media framed the decision as targeting a specific pattern of cutting China out of the value its own ecosystem had created. Beijing, by contrast, backed Zhipu and MiniMax through Hong Kong IPOs, in which mainland investors can participate via Stock Connect, the programme linking the Hong Kong Stock Exchange to the Shanghai Stock Exchange.

DeepSeek’s Incentives May Not Change

Governments often treat their funds as finite economic investments rather than open-ended grants, forcing companies to reach milestones faster.

“In many industries, government-backed capital actually increases the pressure to reach profitability,” says Rui Ma, an angel investor and a China tech analyst. “It is not acceptable, politically or institutionally, to just burn public money indefinitely.” 

This stands in contrast to the long burn-rate tolerance that US private capital has extended to AI startups.

State and national capitals also operate with different incentives. “Local governments often care about very concrete things: opening offices locally, hiring locally, building facilities locally, paying taxes locally. The national government is operating at a broader strategic level. It wants companies to support national objectives,” says Ma.

In DeepSeek's case, those national objectives happen to align closely with what the company is pursuing. 

Ma describes one of the goals as freedom from NVIDIA lock-in — first at the inference layer, and eventually in training. DeepSeek is already moving in that direction by optimising its latest V4 family of models for Huawei hardware.

State fund involvement helps secure access to Huawei’s domestic compute allocation, and the company's model architecture innovations address China’s chip shortage, not by increasing demand for NVIDIA chips but by reducing the need for them altogether. Training remains heavily dependent on NVIDIA for most serious Chinese labs, but inference is where companies are pushing hardest to reduce that dependency through software optimisation, domestic chip adaptation, and more efficient use of available compute.

“That is why national-interest capital makes sense in this case: the company’s technical roadmap and the country's strategic roadmap are unusually well aligned,” says Ma.

She also says local governments are already tracking daily token usage almost as a proxy for AI adoption, productivity, and development, reflecting a broader belief that AI consumption itself is becoming an indicator of economic activity.

“Liang Wenfeng [CEO of DeepSeek] has stated publicly that he would only accept ‘national-team capital and entity-industry capital,’” says McGrath. 

Keeping US-linked VCs off the cap table, especially Sequoia, also limits DeepSeek's exposure to additional US regulatory scrutiny. 

“For many companies building at the frontier of Chinese AI right now, state capital is the preferred option,” says McGrath. “Building to a Hong Kong IPO doesn't sound so bad to many.”