A Western VC's two-week observation of Chinese AI: hardware shocks, software pessimism, founders surprised

Author: José Maria Macedo, Co-founder of Delphi Labs; Compiler: Jinse Finance Claw

I spent two weeks in China, meeting founders, venture capitalists (VCs), and CEOs of publicly listed companies across the entire AI ecosystem. Before going, I was optimistic about this ecosystem, expecting to find world-class AI talents starting businesses for a fraction of Western valuations.

By the time I left, my perspective had become more nuanced: confidence in hardware was stronger than I had anticipated, while my outlook on software was more pessimistic, and some views I held about Chinese founders surprised me.

The Founder Issue

The great founders I’ve invested in possess highly recognizable traits: independent thinking, rebelliousness, extreme focus, and near-paranoia. They do not follow orders, constantly ask “why,” and refuse to accept established conventions. Their decisions may seem incomprehensible to outsiders, but to them, they are self-evident. They have an inner, unyielding intensity that often manifests as a long-term obsession with excellence. Their lives have a “sharpness” that allows you to identify them among the countless high-IQ talents seen by VCs.

However, many Chinese founders I encountered belong to a different archetype—this surprised me.

They are exceptionally talented—graduates of prestigious universities, have worked at ByteDance or DJI, published papers in Nature, and hold multiple patents. Achievements that in the West are only attainable by the top technical geniuses here are merely “tickets to entry.” They are also the most hardworking people I have ever met. We held meetings at any time, on weekends, and in different cities. One founder even came to meet us on the day his wife gave birth!

However, independent thinking, rebellious spirit, and the vision to go from 0 to 1 are hard to find. The founders’ backgrounds are highly similar, their business plans tend to avoid risk, and their ideas are often “impressive V2 versions” of existing things rather than truly original bets. Given the scale of technical talent produced in China, I had expected to see more unheard-of brilliant ideas.

My view is that China’s education system has produced excellence but hasn’t left enough room for “deviating from the norm.” The result is that these founders are genius executors of known problems, rather than those who can identify “problems whose existence is unknown.”

VCs Are Reinforcing This Model

Interestingly, local investors are actively exacerbating this model.

The investment logic of most Chinese funds completely revolves around supporting elite alumni from ByteDance or DJI—valuing background over sharpness, and credentials over beliefs. The backgrounds of VCs themselves reflect this: most come from large companies, consulting, or investment banking, similar to European VCs a decade ago.

Ironically, historically, China’s best founders—those who truly built epoch-making companies—never worked in large companies. Jack Ma was an English teacher who failed the college entrance exam twice; Ren Zhengfei founded Huawei after leaving the military at 43; Liu Qiangdong started from a stall in Zhongguancun; Dr. Wang Xing dropped out of school and started his entrepreneurial journey from day one. A recent example is Liang Wenfeng, who had never worked elsewhere before founding DeepSeek. These are “outliers,” lacking the so-called credentials that the current system tends to overlook.

There is real excess return (Alpha) in seeking out such outlier talent, but currently, it seems few are paying attention.

Shenzhen and the Hardware Ecosystem

What shocked me the most in China wasn’t the startup roadshows.

It was the underground hardware world in Shenzhen—where engineers systematically acquire high-end Western products, disassemble components one by one, and perform reverse engineering with extreme precision. By the time I left, I was genuinely unsure if most Western hardware founders understood the kind of competition they were up against. The network effects here are not theoretical, but physical, dense, and have accumulated over decades.

The entrepreneurs we met substantiated this point with data: over 70% of hardware investment comes from the Greater Bay Area, with nearly 100% from within China—allowing iteration cycles to reach speeds that Western hardware companies cannot even approach.

Most founders I met follow the “DJI model”: building consumer-grade hardware in a certain niche market (such as electric wheelchairs, weeding robots, next-generation fitness equipment), achieving 8 to 9-figure revenues, and then expanding into adjacent categories using their customer base or underlying technology. Some of these businesses have already far exceeded expectations. The most impressive company I encountered was TuoZhu, a 3D printing company that most Westerners have never heard of, reportedly achieving annual profits of $500 million and doubling every year.

Pessimism About Chinese Software

By the time I left, my skepticism about opportunities in Chinese software had deepened beyond my initial arrival.

At the model level, China’s open-source models are indeed impressive—but closed-source models still significantly lag behind the top levels in the West, and the gap may widen. The capital expenditure (CapEx) gap is vast, and GPU acquisition remains constrained. Western labs are increasingly strengthening restrictions on model distillation. Revenue data clearly illustrates the problem: reports indicate that Anthropic made $6 billion just in February. Meanwhile, the annual recurring revenue (ARR) of the best models in China is only in the tens of millions of dollars.

In software entrepreneurship, the mainstream portrait is of former product managers and researchers from ByteDance who are building consumer software focused on agents or environmental perception for the Western market. The talent is indeed present, but many products fall within the coverage of major firms’ native functionalities—once a major firm releases a version, these products become redundant. I was also shocked by the lack of large, fast-growing private software companies. In the West, aside from model companies, there are multiple startups (like Cursor, Loveable, ElevenLabs, Harvey, Glean) printing 9 and 10-figure ARR at astonishing growth rates. Yet, this level of breakout private software company is virtually non-existent in China—few exceptions like HeyGen, Manus, and GenSpark, once opportunities are identified, ultimately choose to leave.

Valuation Bubble

Despite the software outlook, bubbles exist in both early and late stages.

In the early stage, while top talent from ByteDance, DeepSeek, and Moonshot is still significantly cheaper than their American counterparts, median valuations have converged. It’s common for consumer startups to have valuations of $100-200 million before even launching a product. Seed pre-financings exceeding $30 million are also not rare.

In the late stage, the numbers are harder to rationalize. Minimax trades at around $40 billion in the public market, with an ARR of less than $100 million—about a 400x sales multiple. Zhizhu has an estimated valuation of about $25 billion against $50 million in revenue. In contrast, the peak financing round for OpenAI had a sales multiple of about 66x, and Anthropic about 61x.

Non-public companies like Moonshot leverage these publicly listed benchmarks to raise funds at valuations of $6 billion, $10 billion, and $18 billion within just a few months. Cryptocurrency investors would recognize this dynamic: investors are using private valuations to compare against unlocked fair prices. Moreover, part of the reason supporting Zhizhu and Minimax at these valuation levels is that they are currently the only way to gain exposure to the “China AI narrative,” thus carrying a premium. This situation will change as more companies go public and dilute this characteristic. Lastly, the IPO window often closes very quickly and without warning—you can’t be sure to complete such arbitrage before benchmark prices fluctuate.

The humanoid robot sector is similar. China has about 200 humanoid robot companies, of which about 20 have raised over $100 million, with a few achieving valuations in the billions—almost all without any revenue, most planning to IPO in Hong Kong in 2026 or 2027. If this market is real, China’s dominance in hardware makes the long-term outcome quite clear. However, commercialization may be slower than the current financing pace suggests, and I doubt the Hong Kong market can support so many humanoid robot companies with billion-dollar plans currently in the works. I have chosen to wait and see.

Asymmetries Worth Noting

One thing I didn’t expect: almost every founder I met is building products for the global market rather than just for the Chinese market. They use Claude Code and watch Dwarkesh’s videos. They are well aware of the San Francisco startup landscape, often even more so than Western investors who don’t pay much attention to the details.

The hostility of the West towards China runs far deeper than China’s hostility towards the West. Chinese founders believe that combining China’s engineering execution, hardware depth, with Western go-to-market (GTM) strategies and product visions is not contradictory. This combination, when present in the right founding teams, will generate some truly exceptional companies.

Identifying these founders—those who do not fit the “elite credential” template optimized by the local VC ecosystem—is precisely the focus of our current attention.