Delphi Labs Founder's Observations and Reflections on Two Weeks in China's AI Ecosystem

Original article by: Delphi Labs Cofounder, @ZeMariaMacedo

Compiled by: Big Claws | PANews Lobster

I spent two weeks in China, visiting founders, venture capitalists, and CEOs of listed companies within the AI ​​ecosystem. Before entering China, I was optimistic about this ecosystem, expecting to find world-class AI talent making significant strides at valuations far lower than in the West.

When I left, my perspective had become more nuanced: I was more optimistic about hardware than expected, more pessimistic about software , and I also had some surprising views about the Chinese founders.

Founder problem

The outstanding founders I've invested in all share a common trait: independent thinking, a rebellious spirit, intense focus, and an almost obsessive dedication. They don't follow the crowd, constantly asking "why," and refusing to accept secondhand wisdom. Their decisions may seem inexplicable to outsiders, but they see them as perfectly logical. They possess an innate, relentless drive, often manifested in a life journey filled with unwavering dedication and excellence. Their life trajectories have a distinctive "edge," making them instantly recognizable among the vast number of highly intelligent individuals I encounter as a VC.

Many of the Chinese founders I've met are a different type—which surprised me.

They are incredibly talented—graduates of top universities, former employees of ByteDance and DJI, published papers in Nature, and hold numerous patents. These achievements, typically reserved for top-tier technical talent in the West, are merely the entry ticket here. Their work ethic also surpasses that of almost anyone I've ever met. We meet anytime, on weekends, and across cities. One founder even came to see us on the day his wife gave birth!

But independent thinking, a rebellious spirit, and a vision to build something from scratch—these are hard to find. The founders' backgrounds are remarkably similar, their business plans are more conservative and risk-averse, and their ideas are more like refined versions of existing things than truly original gambles. Given that China has cultivated such a large pool of technical talent, I expected to encounter more people proposing ideas I'd never heard of before.

My assessment is that China's education system can cultivate excellence, but it doesn't leave enough room for "deviation." It produces founders who can execute known problems to the extreme, rather than those who emerge with a problem that "no one has ever realized is a problem."

Venture capital is reinforcing this model

What's even more interesting is that local investors are actively amplifying this trend.

Most Chinese funds base their entire investment logic on investing in the best alumni of ByteDance or DJI—valuing prestigious backgrounds over individuality and seniority over judgment. The composition of VCs is also similar: most come from large companies, consulting, or investment banking backgrounds, just like European VCs a decade ago.

Ironically, some of the greatest Chinese founders in history—those who truly built multi-generational companies—never worked for a large corporation. Jack Ma was an English teacher who failed the college entrance exam twice; Ren Zhengfei founded Huawei after retiring from the military at 43; Liu Qiangdong started as a street vendor and founded JD.com; Wang Xing dropped out of his doctoral program to start his business from scratch. Most recently, Liang Wenfeng had never worked anywhere before founding DeepSeek, only in his own company. These are all outliers, people without impressive resumes—precisely the kind of people the current system chooses to skip.

Finding these people is where the real rewards lie, and in my opinion, almost no one is looking in that direction right now.

Shenzhen and Hardware Ecosystem

In China, what amazes me most is not any startup roadshow.

Rather, it was Shenzhen's "hardware underground"—in those workshops, engineers systematically purchased high-end Western products, disassembled them component by component, and reverse-engineered them with meticulous precision. As I left, I was truly unsure whether most Western hardware founders genuinely understood what they were competing against. The network effects here weren't just theoretical; they were real, inseparable, and the result of decades of accumulation.

The entrepreneurs we met confirmed this with more than 70% of the hardware raw materials come from the Greater Bay Area, and nearly 100% come from mainland China—which makes the iteration cycle far exceed the capabilities of Western hardware companies.

Most of the founders I've met are following DJI's strategy: focusing on consumer hardware in a niche market—electric wheelchairs, lawnmowers, next-generation fitness equipment—building eight- or nine-figure revenues, and then expanding into adjacent categories based on their customer base or underlying technology. Some of these companies have grown to a scale far exceeding your imagination. The most impressive company I've encountered is Bambu, a 3D printing company virtually unknown to most Westerners, which reportedly generates $500 million in annual profit and doubles its profit every year.

pessimistic about Chinese software

When I left, I was more skeptical of the opportunities in China's software industry than when I arrived.

At the model level, China's open-source achievements are indeed impressive—but closed-source models still lag significantly behind the best Western models, and this gap is likely to widen further. The capital expenditure gap is enormous, GPU acquisition remains constrained, and Western labs are intensifying their efforts to suppress model distillation. Revenue data speaks for itself: Anthropic reportedly achieved $6 billion in revenue in February alone. China's best model, ARR, is still in the tens of millions of dollars range.

At the software startup level, the mainstream profile consists of former ByteDance product managers and researchers who are developing intelligent agent or environmentally aware consumer software for the Western market.

The talent pool is real, but most of these products fall within the scope of functionalities natively integrated into large labs—making them vulnerable to being rendered obsolete by a single product release. I am also deeply shocked by the overall lack of large-scale, high-growth private software companies in China.

In the West, besides model companies, several startups have achieved nine-figure or even ten-figure ARR with astonishing growth rates—Cursor, Loveable, ElevenLabs, Harvey, and Glaan. Disruptive private software companies at this level are virtually nonexistent in China—and the few exceptions, such as HeyGen, Manus, and GenSpark, all ultimately chose to leave once they found a breakthrough.

Valuation bubble

Despite the poor state of the software industry, a bubble is real—both in its early and late stages.

In the early stages, while the cost of top talent from ByteDance, DeepSeek, and Dark Side of the Moon remains significantly lower than comparable talent in the US, median valuations have converged. Valuations of $100 million to $200 million for pre-product consumer startups are commonplace, and seed rounds exceeding $30 million are not unusual.

In later stages, numbers become even less convincing. MiniMax is valued at approximately $40 billion in the public market, while ARR is less than $100 million—a price-to-sales ratio of about 400. Zhipu AI is valued at approximately $25 billion with revenue of about $50 million. In comparison, OpenAI's peak valuation was about 66 times that of ARR, and Anthropic's was about 61 times.

Private modeling firms like Moonshot are using these publicly traded companies as benchmarks, raising funds at valuations of $6 billion, $10 billion, and $18 billion in just a few months. Those in the crypto world should be familiar with this dynamic—investors are comparing private placement valuations to pre-IPO share prices.

Furthermore, part of the reason Zhipu and MiniMax maintain their current valuations is that they are currently the only way to gain exposure to the Chinese AI narrative, which inherently carries a premium. However, this scarcity will change as more companies go public and the scarcity is diluted. Finally, the IPO window has a history of closing quickly and without warning—before you've completed your arbitrage, the benchmark you're referencing may have already changed, and there's no guarantee against it.

The same situation exists in the humanoid robot sector. China has approximately 200 humanoid robot companies, with about 20 having raised over $100 million and several valued at billions of dollars—almost all in the pre-revenue stage, and most planning to list on the Hong Kong Stock Exchange in 2026 or 2027. If this market is real, China's hardware dominance makes the long-term trajectory clear. However, commercialization is likely to be much slower than the current pace of funding suggests, and I have serious doubts about whether the Hong Kong stock market can accommodate the numerous billion-dollar humanoid robot companies currently queuing up. I choose to wait and see for now.

Asymmetry worth noting

One thing I didn't expect was that almost every founder I met prioritized the global market over the Chinese market. They used Claude Code, followed Dwarkesh, and were extremely knowledgeable about the San Francisco startup ecosystem—often more so than Western investors who hadn't paid close attention.

The West's hostility towards China far outweighs China's hostility towards the West. Chinese founders see no contradiction in combining China's engineering execution and hardware depth with the West's market development capabilities and product vision. When this combination is realized within the right founding team, it can give rise to some truly remarkable companies.

Finding founders—those whose qualifications don't fit the optimized "templates" of the local VC ecosystem—is exactly what we focus on.

Special thanks to @woutergort for sharing his amazing network of contacts in China, and thanks to @PonderingDurian.

I'm grateful to Claude for organizing this trip and for patiently helping me sort out the fragmented thoughts I had on the plane.

Note: This article was compiled under the leadership of Ai.