35. He Brought Investors Managing Hundreds of Billions to China Again to See MiniMax—Sleeping at Atour, Eating at Lao Xiang Ji! - Robert | Baiguan Technology

Summary

• The episode’s biggest call comes down to one sentence: The world’s view of China is shifting from “a huge market” to “a subject to learn from.” Robert’s observation is that Chinese groups once traveled to Silicon Valley, Wall Street and Japan to study them, while “over the past two years, the rest of the world has started looking at China more and more the way we once looked at the US”—even if China is “not exactly easy to invest in” now: consumption is weak and major names such as Hang Seng Tech remain difficult, yet attention is rising across the board and expectations for 2025 have shifted materially.
• The robotics industry is “a little overhyped”; the real bottleneck is data, not movement. China may have only 10,000-20,000 commercially deployed robots, most of them dormant, versus million-scale autonomous-driving fleets collecting data; at Fourier’s data-collection center, “one guy sits there, using teleoperation to make a robot close a door, then close it again,” while a dozen-plus operators repeatedly drill one robot each. Raymond’s harsher formulation is that betting on shipment volume leading to a data pipeline and then better models means stacking “five maybes,” which is effectively impossible—and may be even more brutal than the EV industry.
• The biggest surprise for some US investors visiting China was not that Chinese technology lags, but that Chinese people cannot even get into driverless cars. In San Francisco, many people would rather hail Waymo, while China’s L4 penetration remains extremely low. Robert’s explanation is that this “clearly isn’t a technology problem”: over a 10-year horizon, aging makes autonomous driving necessary; over a 3-5-year horizon, ride-hailing is an employment backstop; so the government supports it over the long term while consciously restraining it in the short term, deliberately preventing the technology from moving too fast.
• China’s consumer market contains a notable expectation gap: the narrative is still that it has “not emerged from the gloom,” but micro data have already begun to stabilize. Robert’s favorite indicator is KFC’s average ticket in Yum China’s earnings: it declined steadily from early 2023 through Q3 last year as volumes and revenue rose but per-order spending fell, then began stabilizing and even recovering from Q3-Q4 last year into Q1 this year—“if it keeps cutting prices, no one dares raise them; conversely, if the market is going to recover, it will be the first thing to recover.”
• Eight or nine billionaire-level investors ordered the entire menu at Lao Xiang Ji; the bill came to RMB400, and even the lowest guess was RMB700-800. Robert’s point, against the overseas media narrative, is that Chinese consumers are not “living on air”; wallets are tighter, merchants are competing harder and “the overall quality of Chinese consumption has been improving”—with niche champions such as JNBY seeing their shares rise several-fold over a few years while paying 7-8% annual dividends.
• This generation of Chinese AI founders is confident because it is “in sync with the world,” and increasingly defines itself less as “China’s OpenAI.” The leading model founders—梁文锋, 闫俊杰 and Tsinghua-trained 杨植麟—are almost entirely domestically educated, while MiniMax bet on Transformer before GPT. Robert identifies the Western presumption: “If you do better than him, you must have done something wrong—whether you stole it or took it,” and DeepSeek R1, built entirely by domestic talent, open-sourced and materially innovative, is the clearest rebuttal.
• The service-oriented government at Shanghai’s Modal Space—“a problem raised in the morning can be solved by noon”—made the European delegates envious, but Raymond’s rebuttal is worth remembering: Shanghai does not represent the full picture of China’s government-business relations. What Europeans kept bringing up, something they both dislike and remain dissatisfied with, may be the Hefei model; Robert’s response was to take them to Hefei next time “so they can see how much a government can actually do.” Future trips to the median city of Linfen, the silver economy at Lu Xun Park and ordinary Chinese factories all point to the same correction: Americans still picture a “Chinese factory” as a sweatshop, but “this era is completely different.”

Deep dive

1. Spend Tens of Thousands of Yuan to Come to China—Skip the Great Wall and Go Straight to Startups

• Robert brings global investors, founders and senior corporate executives to China for study trips. People willing to pay tens of thousands of yuan typically “do not come with strong preconceptions; they come with a mindset of learning.” Others have never visited China before but skip the Great Wall and the Forbidden City to go straight to Chinese startups. One delegate called the service “educational tourism.”
• The source of participants is fairly evenly distributed globally: Southeast Asia accounted for roughly one-third of the past two trips, including Singapore, Indonesia and Thailand; Europe—especially Germany—has been steady; and the US accounts for 30%, “but only 30%.” The groups are sharply stratified. Singaporean investors have already seen Li Auto and Nio, so they skip EVs and go straight to questions about autonomous-driving regulation, testing and technical choices. Europeans are still asking more basic questions. Raymond’s analogy: “Kindergarteners and college students are sitting in the same classroom.”
• Black Lake was another case that left a strong impression. It has no obvious overseas comparable and provides work-order management and a complete internal system for small and midsize manufacturing plants. A former HKEX director ultimately urged the company to list. Robert believes China’s manufacturing base gives what might be a niche US business enough scale to become substantial; after accumulating data, it could potentially move toward agile manufacturing.

2. Fewer Are Saying “China’s OpenAI”; the Western Default Is “Without Us, You Can’t Survive”

• Robert relayed how large-model companies describe the post-85/post-90 founders: their defining trait is confidence, because “the environment they were born into was broadly in sync with the world, without a massive information gap.” The previous generation used labels such as “China’s Siemens” and “China’s Google”; Raymond added that early internet companies were often called China’s Facebook or China’s Netflix. Today, if you say, “I want to become China’s OpenAI,” even you would not sound convinced. 闫俊杰 founded MiniMax before GPT emerged and backed the Transformer direction, reflecting first-principles thinking rather than imitation. MiniMax also holds a non-consensus view in the current industry by emphasizing multimodality over the prevailing coding focus.
• The sharpest exchange came when a European delegate first asked a large-model company about its share of returnees from overseas. The answer was one-third overseas-trained and two-thirds domestic. The delegate then asked: “If the US one day stopped allowing Chinese students to study there, would that have a major impact on you?” The other side froze. Robert unpacked the premise: “A lot of Westerners start from ‘without us’—from ASML to the education system, the entire West—‘you can’t do anything.’ If you do better than him, you must have done something wrong, whether you stole it or took it.” The company’s answer was that talent comes from diverse sources; it does not follow that the company cannot survive without returnees.
• The best counterexample, both men agreed, is DeepSeek: “It is very clearly an entirely domestic team, yet when R1 came out, it was obvious that it contained some very new ideas, made major innovations in the industry—and was open source.”

3. Robots Learning to Do Human Tasks: Data Collection Is the Core, and Robert Thinks the Industry Is a Little Overhyped

• The shock at Fourier was not the performance but what happens upstream. The robots are already highly capable in terms of movement, but have little real autonomy. The biggest bottleneck is insufficient data: autonomous driving has million-scale fleets collecting data on the road, while China’s commercially deployed robots may number only 10,000-20,000, most of them idle most of the time. Customers buy the hardware and then have to train it in real-world settings; companies also need both real and synthetic data. The demonstration environment simulated a home, with robots folding clothes and washing dishes. “One guy sits there, using teleoperation to make a robot close a door, then close it again, and again,” recording the differences in closing force and joint load each time. “The scene was pretty sci-fi.”
• “I think robotics is a little overhyped,” Robert said candidly. No one can say with precision when the so-called “ChatGPT moment for robotics” will arrive. Current attention is enough to create commercial value—performance robots themselves meet a demand—but mass adoption “in tens of millions of homes” is still far away. Companies are raising capital and scaling quickly while the AI window is open. Asked whether Tesla could do it, Chinese companies offered every imaginable answer: they respect Tesla but think Musk is selling a dream, while Optimus “changes its timeline every time.”
• Raymond relayed an investor’s view that China’s overall robotics financing has already reached an absurd scale, then offered a colder calculation: the bet is on high shipment volumes creating a data gateway, with better data potentially producing better models. “After stacking five ‘maybes,’ it is effectively impossible,” and the setup may be even more brutal than the EV industry. Investors are already working backward through the supply chain: Who supplies the actuators? Is the chip from Nvidia? “Memory is already strangling AI.”

4. The Expectation Gap in Autonomous Driving: Not a Technology Problem, but a Policy Dilemma

• A major reversal came from many US delegates. Waymo has achieved high penetration in many US cities, and in San Francisco many people would rather hail Waymo than Uber. They assumed that China had “robots flying everywhere” and that ordinary people should be using autonomous driving every day. Instead, the share of Chinese consumers using pure L4 mobility in daily life is “extremely, extremely low”: Shanghai’s pilot zone is near Century Park, while Wuhan’s Apollo Go operates in outlying areas.
• Robert’s framework struck the delegates as compelling. China’s L4 technology is not materially different from Waymo’s; the dilemma lies in the time horizon. Over 10-20 years, an aging population will leave too few human drivers, making the substitution of silicon for carbon necessary. Over 3-5 years, ride-hailing is a social backstop for employment, amid heavy job pressure. So the government “supports it over the long term and consciously controls it in the short term,” deliberately slowing the technology’s development. Raymond acknowledged that this overturned his assumption: “I used to think I didn’t see Waymo-like cars around me because China hadn’t built them yet.” In reality, “where to draw the line is a policy question, not a technology question.”

5. “A Problem Raised in the Morning Can Be Solved by Noon”—and Raymond’s Rebuttal on Hefei

• Government incubators such as Modal Space offer more than premises and mentoring. Most tenants are companies across the AI value chain; support for tokens is “particularly strong,” while approvals and tax matters are handled centrally and service staff proactively visit companies to collect problems. European delegates were “pretty shocked and pretty envious” to hear that “a problem raised in the morning can be solved by noon.” Raymond cut in: European regulation “is not complicated; it is hostile.”
• The disagreement is worth preserving. Robert acknowledged that Shanghai is a selectively advanced sample, with government-business relations that are “efficient, with clear boundaries.” Raymond disagreed that Shanghai represents the most advanced model; it looks more like a Hong Kong-style service system. The model Europeans repeatedly brought up, one they neither like nor feel satisfied with, may be Hefei’s. Over the past 10 years, Hefei’s support for high-tech companies has been so aggressive that they “just took off.” Robert’s response: “Then next time we should take them to Hefei and show them how much a government can actually do.”

6. Sleeping at Atour, Eating at Lao Xiang Ji: The Micro Truth of Consumption and KFC’s Average Ticket

• The consumption group visited niche champions with market caps in the tens of billions of yuan. M&G Stationery “claims to have more stores than China has schools”; JNBY has spent more than 30 years in fashion, an industry “many people cannot survive in,” with 60-70% of customers consisting of private-domain members maintained through store managers’ WeChat groups. Its share price has risen several-fold over a few years, with annual dividends of 7-8%—Robert specifically noted that this was not investment advice. The common trait: they became small giants in a particular niche and “do not care about anyone outside their customers.” Delegates stayed one night at Atour; some thought it was no worse than a five-star hotel at a fraction of the price.
• The most widely circulated scene came in Hangzhou, where 8 or 9 billionaire-level investors walked into an ordinary Lao Xiang Ji and ordered every dish on the menu, with some ordering two portions. The total was RMB400; even the lowest guess was RMB700-800. Robert’s conclusion, countering the overseas narrative, is that Chinese consumers are not “living on air.” Wallets are tighter, merchants are competing fiercely and service quality is improving: “The overall quality of Chinese consumption has been improving in recent years. It is contradictory, but it objectively exists.” In a sense, that is building momentum for future consumption growth.
• The tool Robert uses to identify a turning point is Yum China’s earnings. KFC’s average ticket declined steadily from early 2023 through Q3 last year—revenue and volume rose, per-order spending fell, and delivery’s share increased, but “at the end of the day, the food simply became cheaper.” It began to stabilize and even recover from Q3-Q4 last year into Q1 this year. The logic is straightforward: China’s largest restaurant chain has the best cost controls. “If it keeps cutting prices, no one dares raise them; conversely, if the market is going to recover, it will be the first thing to recover.” Robert said neither domestic nor overseas markets have priced this in: “I think there is an expectation gap here.” Raymond offered his own rough corroboration—traffic jams, queues and full flights. Shanghai has recently been “both jammed and full of queues; the economy looks genuinely pretty good.” He sees sentiment as having been worst in 2022 and 2023, with “the necessary housing correction over,” while Robert believes Chinese households have returned to a normal mindset.
• A soft-power example came from Brazil, where Raymond was left speechless by a 13- or 14-year-old girl reciting Labubu’s family tree. Brazil has no official Pop Mart stores, while Chinese electronics shops in its Chinatowns display fake Labubu figures to attract customers: “If you really don’t show up, the fake will become popular.” One Brazilian macro-fund manager who reads Robert’s work later visited China; his daughter assigned him the main task of buying a particular obscure series.

7. Next Stops: Linfen, Lu Xun Park, Ordinary Factories—and the Bigger Call on the Paradigm Shift

• Robert’s 3 planned trips are designed to correct distorted views. The first is a median-city trip to Linfen, “a place representative of the broadest majority,” where visitors can decide for themselves how most Chinese people live. Raymond’s takeaway was that he began using Linfen’s GDP as a benchmark whenever he visited another city: “Most of China is not inside Jing’an Temple.” The second is the silver economy: newly retired people born in the 1960s and 1970s “may be China’s wealthiest generation,” while weekend Lu Xun Park has dozens of singing and dancing groups of retirees—“Shanghai seniors’ Coachella,” in Raymond’s words. The third is the ordinary factory. Ask any American about a “Chinese factory” and the mental image is still a sweatshop, but the automation and management of ordinary Chinese factories are “far beyond what people overseas imagine.” Last year’s most striking scene was an 8-story Hangzhou mall with many storefronts but few customers, where every shop was a livestreaming studio.
• The overarching call is that China was once understood simply as “a huge market,” but over the past two years it has become “an object of study”—“we used to look at the US this way; the rest of the world is increasingly looking at China the way we once looked at the US.” Robert does not ignore the investment difficulties: consumption is weak and its recovery date is unclear, while large names such as Hang Seng Tech remain difficult. But attention has shifted from “Can I make money?” to “Can I improve my understanding?” In 2023 and 2024, only a handful of teams produced English-language China content; since last year, businesses targeting specific niches have begun to emerge. “In 2025, expectations changed very clearly.”
• Raymond’s closing reflection was personal. He is often too tired to explain himself when confronted with prejudice, but he pointed to Robert’s Substack and his having written an editorial for The New York Times, arguing that the world needs people like Robert “to stand in the other side’s shoes … so there can be more understanding and fewer misunderstandings.” Chinese people do 10 things and say 1; in the end, they say 2, “and the other 1 is your contribution.” Robert’s answer: at minimum, they can give people who genuinely want to understand China a way to access information. “That is what we believe is the meaning and value of our existence.”