After Two Weeks of Visiting China's AI Ecosystem, Delphi Labs Founder: All My Previous Judgments Have Been Overturned
- Core Viewpoint: Through on-the-ground research into China's AI ecosystem, the founder of Delphi Labs revised his original views, concluding that China demonstrates unexpectedly strong competitiveness and network effects in the hardware sector. However, it underperforms expectations in software innovation, model capabilities, and the independent, critical thinking spirit of its founders. He also points out significant valuation bubbles in the current market.
- Key Elements:
- Differences in Founder Traits: Observed that many Chinese founders possess top-tier resumes and execution capabilities but generally lack the independent thinking, rebellious spirit, and original vision for creating something from scratch that are typical of outstanding Western founders.
- Significant Hardware Ecosystem Advantage: The hardware ecosystem, represented by Shenzhen, leverages systematic reverse engineering and dense physical network effects to achieve rapid iteration capabilities far surpassing those in the West, and has given rise to large, profitable companies like Bambu Lab.
- Gap in Software and Models: China's closed-source models show a clear gap compared to top Western levels, and there is a lack of rapidly growing large private software companies. Many products are vulnerable to being disrupted by feature updates from major Western labs.
- Widespread Valuation Bubble: Both early-stage and late-stage AI companies exhibit valuation bubbles. For example, some Chinese model companies have price-to-sales ratios as high as several hundred times, far exceeding those of Western counterparts like OpenAI.
- Humanoid Robot Sector Overheating: There are about 200 humanoid robot companies in China, with several securing high levels of funding but generating no revenue. Their IPO prospects and commercialization speed are questionable.
- Information Asymmetry and Global Vision: Chinese founders are very knowledgeable about the Western ecosystem, generally prioritize exploring global markets, and are committed to combining China's engineering strengths with Western market and product thinking.
Original Author: José Maria Macedo, Cofounder of Delphi Labs
Original Compilation: TechFlow
Introduction: The founder of Delphi Labs spent two weeks intensively touring China's AI ecosystem, meeting with a large number of founders, investors, and CEOs of listed companies.
His conclusion was surprising: more bullish on hardware than expected, more pessimistic on software than expected, and his observations of Chinese founders overturned his previous perceptions.
The article also touches on hot topics such as valuation bubbles, the humanoid robotics sector, and information asymmetry between China and the West.
Full text is as follows:
I spent two weeks in China, meeting a large group of founders, VCs, and CEOs of listed companies within the AI ecosystem. Before going, I was bullish on this ecosystem, expecting to see world-class AI talent working at valuations far lower than in the West.
My view changed when I left—it became more specific: hardware is stronger than I thought, software is weaker than I thought, and some observations about Chinese founders also surprised me.
The Founder Problem
The great founders I've invested in share a common trait: independent thinking, rebelliousness, extreme focus, and obsession. They don't follow orders. They constantly ask "why" and refuse to accept second-hand wisdom. The decisions they make seem inexplicable to outsiders, but feel natural to them. They possess an intrinsic, unstoppable intensity, often manifested as a long-term obsession and excellence. As a VC who meets many smart people daily, you can spot this type in a crowd because their life trajectory has a distinct "sharpness."
Many founders I met in China were of another type, which surprised me.
They are exceptionally talented—top universities, resumes from ByteDance or DJI, Nature papers, multiple patents. In the West, these achievements are reserved for the most elite technical talent; in China, they are the entry ticket. They also work harder than almost anyone I've ever met. We held meetings at all hours, non-stop on weekends, rushing between cities. One founder met with us on the day his wife gave birth.
But independent thinking, rebellious spirit, a vision from 0 to 1—these were harder to find. Founders' backgrounds were highly similar, pitches leaned conservative, and many ideas were upgrades of existing products (impressive V2s) rather than truly original bets. Given the massive scale of technical talent China produces, I expected to encounter more people "showing up with ideas I'd never heard of."
My interpretation: China's education system cultivates excellence but doesn't leave enough room for deviation. It produces top-tier executors skilled at solving known problems, not the kind of people who "show up with a problem nobody knew existed."
VCs Are Reinforcing This Pattern
More interestingly, local investors are exacerbating this trend.
The investment logic of most Chinese funds is built on a premise: invest in the best people coming out of ByteDance or DJI. They look at resumes, not edge; at background, not conviction. The VCs' own profiles are similar—coming from big companies, consulting, or investment banking, much like European VCs a decade ago.
Ironically, historically, most Chinese founders who truly built great companies never worked at big companies. Jack Ma was an English teacher who took the college entrance exam twice before getting in. Ren Zhengfei founded Huawei at 43, having previously been in the military. Liu Qiangdong started by selling goods at a market stall. Wang Xing dropped out of his PhD to start a business. More recently, Liang Wenfeng, who created DeepSeek, never worked anywhere outside his own company. These people are outliers, the ones without "standard resumes"—precisely the people the current investment system misses.
Finding this type of person offers real alpha, but currently, few seem to be looking there.
Shenzhen and the Hardware Ecosystem
The most shocking thing I saw in China wasn't some startup's pitch.
It was Shenzhen's hardware underground workshops—engineers systematically obtaining high-end Western products, disassembling them component by component, and reverse-engineering everything in an extremely rigorous way. When I walked out, I genuinely wasn't sure if most Western hardware founders understood what they were competing against. The network effects here aren't theoretical; they are physical, dense, and built up over decades.
The entrepreneurs we met confirmed this with data: over 70% of hardware investment comes from the Greater Bay Area, close to 100% from mainland China—this means iteration cycles are something Western hardware companies simply cannot match.
Most founders I met are using DJI's playbook: make consumer hardware in a niche—electric wheelchairs, lawn-mowing robots, next-gen fitness equipment—grow revenue to 8 or 9 figures (USD), then leverage the customer base or underlying technology to move into adjacent categories. Some companies are already much larger than you'd think. The strongest company I saw this time was Bambu Lab, a 3D printing company most Westerners haven't heard of, reportedly with $500 million in annual profit and doubling every year.
Bearish on Chinese Software
When I left, I was more skeptical of Chinese software opportunities than when I arrived.
At the model level, China's open-source is indeed strong, but there remains a clear gap between closed-source models and the best in the West, and the gap may still be widening. The capital expenditure gap is massive. GPU access remains restricted. Western labs are increasingly cracking down on distillation. Revenue numbers tell the story: Anthropic reportedly did $6 billion ARR in February alone. The best Chinese model companies have ARR in the tens of millions of dollars range.
On the software startup side, the mainstream profile is PMs and researchers from ByteDance, building agentic or ambient consumer software for Western markets. The talent is indeed strong, but many of these products fall squarely within the scope of features that big labs will natively release—one update could render them redundant. Another thing that surprised me: China lacks large, fast-growing private software companies. In the West, besides model companies, there is a batch of startups already doing 9-figure or even 10-figure ARR with astonishing growth—Cursor, Loveable, ElevenLabs, Harvey, Glean. This tier of breakthrough private software companies basically doesn't exist in China—a few exceptions like HeyGen, Manus, GenSpark, which also left after gaining traction.
Valuation Bubble
Despite the bleak software picture, the bubble is real—both early and late stage.
On the early stage, the top talent from ByteDance, DeepSeek, and Moonshot AI is indeed significantly cheaper than equivalent US talent, but median valuations have converged. It's common for pre-product consumer startups to be valued at $100-200 million. Pre-seed rounds exceeding $30 million are not unheard of.
Late-stage numbers are harder to justify. MiniMax's valuation in the public market is around $40 billion, with ARR under $100 million—roughly 400x revenue. Zhipu AI is around $25 billion for $50 million revenue. For comparison: OpenAI's highest valuation round was about 66x ARR, Anthropic about 61x.
Private model companies like Moonshot AI are using these public market comps to raise funds—from $6 billion to $10 billion to $18 billion within months. Friends in crypto will be familiar with this playbook: investors compare private valuations to a public market price that is "pre-unlock." Also, Zhipu and MiniMax can maintain these levels partly because they are currently the only way to get exposure to the "China AI narrative," which carries a premium itself. But as more companies list, this premium will dilute. Finally, IPO windows have a characteristic—they can slam shut without warning. No one can guarantee you'll close this arbitrage before the comp prices move.
The humanoid robotics sector is in a similar state. China has about 200 humanoid robot companies, around 20 have raised over $100 million, several are valued in the billions—almost all with no revenue, most planning HKEX IPOs in 2026 or 2027. If this market is real, China's hardware advantage makes the long-term landscape relatively clear. But commercialization may be much slower than the current fundraising pace suggests, and I doubt the HKEX can support the dozens of multi-billion dollar humanoid robot companies currently queuing for IPOs. I'm staying away for now.
Image: Couldn't resist including a clip of a filmed humanoid robot doing a front flip
Noteworthy Information Asymmetry
One thing that surprised me: almost every founder I met is targeting the global market first, then the Chinese market. They use Claude Code, listen to Dwarkesh's podcasts, and know the San Francisco startup ecosystem inside out—often more than Western investors who haven't been keeping up.
Western hostility towards China is noticeably greater than Chinese hostility towards the West. Chinese founders see no contradiction in combining China's engineering execution and hardware depth with Western go-to-market and product thinking. When this combination materializes in the right founding team, it can birth some truly remarkable companies.
Finding these founders—the ones who don't fit the "standard resume template" optimized by the local VC system—is what we're doing now.
Special thanks to @woutergort for opening up his excellent Chinese network, @PonderingDurian for organizing this trip, and Claude for patiently editing my ramblings on the plane.
