China's state AI fund just took a direct stake in Kling
Beijing's national artificial intelligence fund put 1.4 billion yuan into Kuaishou's video-generation unit on Monday, and the size of the cheque matters less than what it signals: China's most commercially successful generative-video model is now a state-backed asset, two months after it raised the largest round in the category's history.
What actually happened
Kuaishou disclosed in a Hong Kong exchange filing that Beijing Kling — the entity that will hold the Kling AI business after a reorganization — signed accession agreements with two new investors. The National Artificial Intelligence Industry Investment Fund injected 1.4 billion yuan in cash. Charoen Pokphand Robot Limited, the robotics arm of Thailand's CP Group, put in roughly 19.29 million dollars, about 131 million yuan. Both received buyback rights under the shareholder agreement, the same protection granted to the earlier investor syndicate.
That is the whole filing, and it is deliberately dry. But it closes a loop that has been open since July 2, when Kuaishou announced a round of up to 20.45 billion yuan (roughly 3 billion dollars) at 15 billion pre-money and up to 18 billion post-money — the largest financing ever raised by an AI video model company. Reuters reported at the time that the deal left room for one additional investor within two months. That window has now been used, twice over, and the larger of the two new names is not a venture fund. It is the state.
Who the fund is, and why that changes the read
The National Artificial Intelligence Industry Investment Fund is not a soft-money regional vehicle. Established in January 2025 under the Ministry of Industry and Information Technology and the Ministry of Finance, it holds 600.6 billion yuan in committed capital with a 13-year lifespan, and its sole limited partner is the third phase of the National Integrated Circuit Industry Investment Fund — the 344 billion-yuan "Big Fund III" that is the biggest semiconductor investment vehicle China has ever built. Its mandate covers the full AI stack: compute, algorithms, data and applied deployment, with embodied intelligence named as a priority.
That pedigree is the point. The fund's portfolio so far reads like a hardware-resilience list — GPU maker MetaX via a strategic IPO allocation, DPU designer Yunsilicon, quadruped-robot company Deep Robotics — companies that reduce China's exposure to US export controls. Kling is a different kind of bet. It is a consumer-and-advertising software business whose main input is inference compute, not fabs. Putting national semiconductor money behind a video generator signals that Beijing now treats generative video as strategic infrastructure in its own right, not merely as an app category that happens to be popular.
The commercial case, and the hole in it
Kling has the strongest revenue line of any Chinese video model. Revenue went from 150 million yuan in Q1 2025 to 340 million in Q4, then to 650 million in Q1 2026 and 850 million in Q2 2026 — a more than 200 percent year-over-year jump that we tracked in Kling AI's revenue jumps 200% as Kuaishou's core business stalls. Kuaishou has now broken the unit out for two straight quarters while its legacy livestreaming business shrank 13.5 percent, which tells you exactly what management wants investors to look at.
The losses run ahead of the revenue. Kling booked roughly 1.1 billion yuan of revenue in 2025 against a net loss near 1.9 billion yuan, up from about 500 million the year before, and it ended 2025 with negative net assets of around 9 million yuan. Against a March 2026 annualized run rate near 500 million dollars, an 18 billion dollar post-money valuation works out to roughly 36 times forward revenue. That is a pricing you can defend for a category leader and cannot defend on unit economics, which remain unproven.
The buyback clause is the tell. Investors can demand their money back at principal plus 8 percent simple annual interest if Kling has not listed by October 30, 2031, or if the internal asset restructuring and algorithm filings slip. This is not venture capital betting on an outcome — it is structured, downside-protected capital with a state entity now sitting inside the same protections. Kuaishou reported holding roughly 117.7 billion yuan in available funds at the end of Q1 2026, so it did not need the money for liquidity. It needed a balance sheet built to survive listing diligence, and investors who will not run.
Context matters here: nine of the ten top-ranked text-to-video models on Artificial Analysis are Chinese, a lead we examined in Chinese labs claim nine of the top 10 text-to-video spots. Kling is the one with the clearest commercial proof.
Who wins, who loses
Kuaishou wins the obvious way: it moves a compute-hungry, loss-making unit off its own income statement without losing control, keeping about 68.33 percent and consolidating Kling's losses against a slowing core. It also buys political cover at a moment when Beijing is tightening its grip on the sector.
The harder question is what state capital costs Kling abroad. Roughly three-quarters of its revenue reportedly comes from outside China, and it has been pushing hard into Western creator and enterprise markets — including a slot inside Adobe's Firefly model lineup. A sovereign-backed cap table does not by itself block those deals; nothing in the disclosed structure conditions Kling's international business. But it gives every foreign procurement team, every studio compliance officer and every US regulator one more reason to ask questions, and it arrives in the same year that Chinese labs have faced theft-of-IP accusations in Washington. The 1.14 percent stake is small enough that this is a governance flag rather than a control event — but flags get read.
The skeptics' case is straightforward. The valuation is set by a restructuring, not a market clearing, and the buyback makes the "18 billion" figure less a price discovery than a financing term. The 36-times multiple rests on a March run rate that has not been independently audited. Video generation is also the most brutally price-competitive corner of AI: inference costs are high, switching costs are near zero, and Kling's three biggest investors — Tencent, Alibaba and Baidu — all ship competing video models. Strategic money in a rival's round is a hedge, not a conviction.
What to watch
Three dates and one disclosure. Whether the Hong Kong IPO lands within the twelve-month window management has signaled, or slips toward the 2031 buyback trigger. Whether the nine-month asset reorganization actually completes, since that is a second, separate buyback condition and the harder one to satisfy. Whether the state fund takes a follow-on position or treats this as a toe-hold — a 1.14 percent stake is a signal, not a commitment. And whether Kling's overseas revenue share holds above three-quarters now that its cap table carries a name that reads as policy.
China's biggest video model now has state money behind it — does that make it stronger, or just harder to sell abroad? Tell us in the comments.
Sources: Kuaishou HKEX filing via Odaily · Reuters on the $2.8B round · TechNode on the July syndicate · South China Morning Post · CNBC · National AI fund background (36Kr)
