The Clear Path, Hidden Path, and Rocky Path of AI Fundraising

By Jiaxiang Shi

Edited by Jing Liu

Before DeepSeek and Manus captured the spotlight, China's AI sector had endured a brutal 2024.

A Dealroom report showed that over the past year, venture capital investment in US AI reached $80.8 billion — more than ten times China's total. Qu Kai, founder of 42Chapter, described the primary market before September 2024 as "basically dead water." Meanwhile, leading dollar funds only wanted to deploy capital through club deals, reserving money for the most promising founders — a group you could count on one hand. The ripple effect: funding cycles for startups were shrinking, yet the time to close individual rounds was growing longer. One AI founder noted that a "short, flat, fast" approach to fundraising now made more sense. This produced a curious spectacle in AI primary-market financing: "Round one comes from your personal network or the most active dollar funds, then relevant industry players and CVCs step in, and because no other incremental capital sources are visible, state capital enters more quickly than before." One state-affiliated investor told us outright that state capital is probably the most active source of funding in the market right now.

Beyond price, another reason investors were reluctant to pull the trigger was the perennial question: "When will you be profitable?" Whenever some AI application blew up, founders or investors would question the ultimate commercial viability — "the token costs don't add up." One investor even believed that "if you stripped out the team-investing aspect and looked purely at product, [this batch of projects] would struggle to raise a second or third round."

Yet never has any era demanded commercialization from startups so ruthlessly (of course, that was an age when capital was abundant), to the point where founders would reflexively drop "ARR" and "achieving PMF" in interviews. The rising international iron curtain and the squeezing of capital bubbles forced this cohort of AI entrepreneurs to tread carefully. "Every funding round is a choice," one AI founder who had raised four rounds told us. The profile of your previous round's investors could determine success or failure in the next.

The density and caliber of talent, stacked against endless compute costs, made this entrepreneurial wave uniquely punishing. "If you can't keep raising, all your early accumulation becomes a burden," one AI practitioner analyzed. An industry with the potential to rewrite humanity's future, yet littered with obstacles in financing — and financing is precisely the battlefield where startups need to move fast and strike hard. An era loudly proclaiming AGI, when burning cash matters most, colliding with a cycle that has the least money and the highest price tags.

So for AI entrepreneurs: where does the money come from?

The Clear Path: Embrace State Capital

In early November last year, a Visual China disclosure brought Zhipu's ownership structure into public view. Among the "Six Little Tigers," Zhipu AI's investor roster included numerous Tsinghua-affiliated entities and state capital. Though Zhipu had long cultivated a reputation of "only raising RMB," the scale of this "assembled platter" — with over 40 participating entities — still stunned observers.

It was also the first company invested in by the Beijing AI Industry Investment Fund since its establishment.

Other large model companies have followed broadly similar trajectories. StepFun's Series B had Shanghai state capital standing behind it; MiniMax maintains close ties with Shanghai International Capital; Baichuan simultaneously anchored itself to Beijing, Shanghai, and Shenzhen governments. To some degree, state capital's entry into large model companies functions as a badge of legitimacy.

Currently, aside from Kimi, all of the "Six Little Tigers" have received state investment to varying degrees. In Silicon Valley's AI wave, tech giants crowded out venture capital firms. In China, beyond the tech giants, it's state capital.

If large model companies — as infrastructure — receiving state money makes inherent sense, then a cohort of AI application companies that should have been consumer-facing and market-oriented are also finding frequent state support.

Beijing established eight closely watched government industry investment funds last year, with total scale exceeding 100 billion RMB. Guo Chuan, deputy secretary of the Party Committee, director, and general manager of Beijing Capital, revealed at year-end that the eight funds had completed investment decisions on 167 projects, with total committed capital of roughly 17 billion RMB. Among these, the Beijing AI Industry Investment Fund had invested in over 30 AI companies since its founding.

According to Anyong Waves, an AI generative application company is nearing completion of a new funding round, with investors being the Beijing AI Industry Fund and a Beijing state-affiliated entity. Yet just six months prior, they had been targeting overseas markets for commercialization. Also receiving Beijing AI Industry Fund backing are two prominent video generation companies: Aishi Technology and Shengshu Technology.

An FA familiar with state capital told us: "The 2024 primary market was down more than 50% from its peak, and Beijing state capital still carries decent name recognition and endorsement." The gist: there was no other choice.

"Going fully government-facing might be a dead end, but 'to GVC' is a viable strategy," one dollar-fund equity lawyer assessed. Many governments view AI as the next generation of investment attraction targets — just as they once pursued cloud services.

Worth noting: state capital willing to make early AI bets is almost exclusively limited to Beijing, Shanghai, and Shenzhen. "Not even Guangzhou." IT Juzi data bears this out: in 2024, Beijing, Shenzhen, and Shanghai logged 213, 123, and 110 AI funding events respectively, far outpacing other cities.

Also, currency differences make it difficult for state capital to make heavy early bets like dollar funds, and the angel projects they back typically revolve around local universities and research institutes — talent, technology, and research-oriented, with process-driven investment.

"Taking state money now is similar to taking BAT money before — what matters isn't just the money itself, and founders can't focus solely on cash; you have to bundle policy and money together," one state-capital-savvy FA partner told us. In the dollar era, money was paramount — growth, product, commercialization systems. But in infrastructure-heavy industries like large models, what matters more is upstream-downstream industrial coordination, M&A, and policy support — precisely what government can provide.

Of course, gifted presents always carry price tags. Different state entities have different demands; taking money from multiple sources requires balancing multiple interests. "Often, headquarters location depends on where support is strongest." Companies, in turn, don't release the rabbit until they see the hawk — receiving portions of funding as they land portions of operations.

This batch of AI companies that took Beijing AI Industry Fund money almost all maintain Beijing offices; one source told Anyong Waves that an AI company funded by Beijing Economic-Technological Development Area will relocate its office to Yizhuang; Hangzhou City Investment Industry Fund participated in Zhipu AI's funding, and Zhipu has established a Hangzhou subsidiary.

The aforementioned FA told us that some state entities prioritize actual local output value and downside risk — financial security and compliance — over valuation. So companies should focus more on absolute funding amounts.

The Hidden Path: International Funds and "De-Silicon Valley"

Where does the money come from? This was the question Austin, co-founder of Sapient, grappled with from day one. Sapient is a large model company attempting to challenge the Transformer architecture.

When Austin began fundraising last year, China's large model startup landscape was already set, yet their exploration required hundreds of millions in USD. At an inopportune moment, facing venture firms that had "already fired their shots," the difficulty was especially acute.

He recalls that in initial investor meetings, the most common challenge was: "Why should I invest in you instead of Zhilin Yang?" "Chinese VCs have no investment preference for technological innovation that hasn't been validated at scale in the US; they need PMF and commercialization more," Austin said. They also sought money from the Middle East and Europe.

The answer turned out to be Singapore and Japan. Last December, Sapient closed a funding round in the tens of millions of USD at a valuation exceeding $200 million, from Singapore's Vertex, Japan's largest private equity group JAFCO, Sumitomo Corporation, and other prominent overseas investors. Austin's conclusion: markets always present opportunities; there are always investors willing to believe.

Sapient's fundraising story is a textbook case of Chinese founders raising globally.

And Sapient's path represents the common profile of most founders today: Chinese founder, globally oriented from day one, majority of employees based in China, entity registered in Singapore or another country, payroll handled through service agreements, leveraging the cost arbitrage of Chinese engineers versus Silicon Valley to win starting-line advantage.

Historically, this fell squarely within dollar funds' wheelhouse. But today, the advent of venture legislation has obstructed this investment chain. So many have cast their gaze toward deeper pockets: Silicon Valley.

But actually securing Silicon Valley investment is extraordinarily difficult. A lawyer familiar with overseas fundraising told us that architecture, founder nationality, and physical presence now all need to align simultaneously to access Silicon Valley money. "Successful founders remain the minority."

Generally, such founders are either serial entrepreneurs in the US or have worked at prominent US tech companies while maintaining close investor relationships — "otherwise raising a first round of US money is very difficult," one dollar fund partner told us.

Success stories include AI video editing tool OpusClip and Pika. But OpusClip founder Yang Zhao had been founding companies in the US since 2013, while Pika has only three Chinese founders with predominantly American employees elsewhere.

More emblematic is AI video generation company HeyGen, which grew ARR from $1 million to over $35 million in just over a year. Founded in the pre-GPT era, the company received seed investment from Hongshan, ZhenFund, and IDG Capital in 2021, then completed a $60 million Series A led by Benchmark at a $500 million valuation last June.

When HeyGen comes up, most founders believe they sacrificed domestic recognition and "some valuation benefits," "made a timely decision." One AI application founder told us, not without envy, that his revenue is comparable to HeyGen's, yet their valuation is ten times his.

Following founders to Silicon Valley are Chinese dollar funds, but they too face new barriers. According to tech media outlet The Information, early last year both HeyGen and OpusClip relocated their China-based engineers to Canada and Singapore; Lightspeed China, an early OpusClip investor, and Hongshan, an early HeyGen investor, subsequently exited their boards, with rumors of forced secondary share sales.

A lawyer familiar with the Silicon Valley ecosystem explained that Silicon Valley funds are acting under duress — "otherwise they'd face constant US government scrutiny." "Chinese founders in Silicon Valley are now very particular about funding sources and registration jurisdictions," one investor told Anyong Waves.

Yet this doesn't appear to be the largest obstacle. One investor told us that when a dollar fund invested in an AI company that a Silicon Valley fund also wanted, "the partner was happy to exit." "What matters is getting good assets. If you're asked to leave early after investing, negotiating a fair price — that's not bad either," one dollar fund partner told Anyong Waves.

But for purely Silicon Valley-based founders, dollar funds' advantages aren't obvious either. Harry Wang, founding partner of Linear Capital, believes that "core entities detached from China capabilities have no special competitive advantage overseas beyond having some money."

So the likely endgame: Chinese-American founders in the US taking money from these offshore dollar funds — each side's second-best option.

Some have gone further. One AI education company was externally perceived as an exemplar of going global. But the founder told us they are not a global expansion company — they serve only the US domestic market, they are a purely American company.

This founder had been a successful serial entrepreneur in China, but in America was a middle-aged nobody. "Going to Silicon Valley to start from scratch with this crowd takes enormous resolve — you have to discard everything you had in China." He succeeded in raising Silicon Valley money.

This path is equally fraught.

The Rugged Path: CVC, M&A, or "Drip Capital"

CVC may be incremental.

Behind the large model "Six Little Tigers," internet giants stand to varying degrees — especially Alibaba, which heavily backed Moonshot AI. AI applications are similar. Jingying Technology, which launched the world's first AI short drama, received Baidu investment; AI video generation companies Aishi and Shengshu Technology, as well as search-focused Metaso, all have Ant Group behind them. Separately, we understand that an AI gaming company valued in the hundreds of millions also received Ant funding. The latest news: Tencent led the funding round for Manus developer Monica.

Brook Venture Partners launched its Soil Seed Program last year, focused on seed-stage AI application startups. Founding partner Yang Jie previously worked at ByteDance. She told us late last year that "we closed fifteen deals in the past five months." We also understand Brook invested in a well-known independent developer.

CVC's benefits need no elaboration: ample ammunition, no policy predicaments. But the downsides are equally obvious: they are themselves upstream or downstream in the value chain, in direct competitive tension with startups. One Shenzhen-based AI application founder told Anyong Waves he had declined multiple visit requests from Tencent-affiliated personnel.

The rapid entry of tech giants has also accelerated the possibility of startup acquisitions. Two AI M&A cases previously drew attention: OPPO's acquisition of Waveform Intelligence and Ant's acquisition of Biansai Technology. Both shared common features: early investors achieved exits while maintaining independent operations. "Biansai's commercialization was difficult, but from a pure technology perspective, Wu Yi has his unique strengths," one Ant strategic investment person explained.

One person close to the transactions analyzed that Ant's acquisition of Biansai Technology aimed to consolidate technical talent, while OPPO's acquisition of Waveform targeted internal business exploration. Such acquisitions generally don't price according to valuation, but rather return investors' principal plus interest to roughly break even, while giving founders and technical staff stock incentives requiring at least four years of tenure — similar to the Character.AI acquisition case.

Similarly, 01.AI was rumored to be acquired by Alibaba. Kai-Fu Lee denied this, stating that 01.AI and Alibaba established an industry large model joint laboratory: "Those wanting to work on ultra-large cluster infrastructure and training will join the joint laboratory and become Alibaba employees."

In the volatile AI application market, Harry Wang previously told Anyong Waves that AI applications won't see winner-take-all dynamics, and burning cash for market share may not hold.

Wang Bolong, founder of Huidu, resigned from a major tech company to start his own venture amid the AI wave, then sold his company and brought his team into a traditional financial firm. He told us that major tech companies currently offer positions and opportunities, but the odds of becoming "cannon fodder" are high. He also declined offers from several highly valued AI startups — "AI application layer startups in 2024 were destined to be tossed about, pivoting and FOMO-ing back and forth like retail investors constantly rebalancing in the A-share market."

This may be what sways some AI founders' calculus: if you can't become the next platform company, why not cash out while the getting's good?

One founder backed by two major tech companies candidly told Anyong Waves that one investor shares his business line and even approached investment with acquisition in mind, but he doesn't object — "if the price is right, it's responsible to investors, and in four years when I come back to start something new, I'll still be in my prime." He ran through the math for us: selling the company could net him over 100 million.

Yet when we asked a strategic investment person whether AI application M&A cases would increase significantly, the response was: "Barriers are too low; replication costs may far exceed acquisition costs."

Plants in deserts evolve leaves into spines; when consumer investment wanes, Drip Capital emerges. Some AI founders have chosen a "destined to be non-mainstream" strategy.

Luo Baishun is co-founder of Chuhaiqu, which he founded in July 2023 to provide incubation services for independent developers — solo founders or small teams going global with AI.

"Chuhaiqu" provides capital to independent developers in debt-like form early on, recouping through revenue sharing later — almost identical to Drip Capital's model. Of course, this isn't cash but rather direct offsets during development or operations (such as user acquisition).

He describes this as fundamentally different from VC logic, more about helping independent developers go from 0 to 1 rather than 1 to n. Because amounts are far below typical angel rounds, Luo calls their investments "gift money rounds." These AI projects not infrequently choose direct sale after operating for some period.

At least for now, the funding landscape's fundamentals haven't shifted much: either becoming an instrument of national will, or receiving tech giant backing — while those unwilling or unselected must scramble with every means to survive. This is the reality of AI financing today.

However, DeepSeek's recent breakout and the agent wave brought by Manus seem to introduce new variables. One founder has noticeably sensed a shift in European and American investors' attitudes toward them — "they've fully realized this is a war between Zhongguancun Chinese and Silicon Valley Chinese."

Perhaps 2025 will truly be the year of AI applications — though many said the same of 2024.

Image source | Unsplash