We are already gathering feedback for the 2027 Spring Baiguan China Tour.
Your preferences will help shape our program. So make sure to fill in our survey here. It won’t take more than 2 minutes of your time.
2027 Spring Baiguan China Tour Survey
We’re currently designing the next Baiguan China Tour for spring 2027 — and before we lock in cities, companies, and dates, we want to hear from you.
Episode in brief
Recapping Baiguan Autumn Tour: Robert recaps the recent Baiguan Tours to Beijing and Shenzhen to visit AI and robotics companies.
Beijing is a deeper tech base than its reputation. Robert says talent (Tsinghua, Peking, the AI research institutes), proximity to policymakers and regulators, and the legacy of Baidu, Meituan and ByteDance explain why Moonshot AI and Z.ai were founded there.
Robotics looks like a bubble, but a useful one. General-purpose humanoids disappoint in open settings. Narrow, commercial use cases already work, from a robot-run pharmacy to cleaning and dish-carrying robots.
Durable value sits in unglamorous components. Dexterous hands and gears are concentrated, hard to copy and priced on cost leadership, not hype.
Commercial space is gated by approvals, not technology. Chinese rocket firms say slow government approval limits launch cadence, and officials tolerate failure less readily than a Musk-led company like SpaceX can.
China is not uniformly “full speed ahead.” Robotaxis are held to pilot zones for employment reasons even though the technology is mature.
The strongest takeaway is niche strategy. Robert argues that as AI models produce generalized, mid-distribution answers, value shifts to peripheral niches, which Smallrig illustrates in hardware and a founder’s lunch talk illustrated for AI.
Why the tours went to Beijing
Robert has just finished the first two tours of the autumn series: a week in Beijing with the independent media company Caixin, covering AI, robotics and commercial space, followed by a shorter Shenzhen trip on robotics and advanced manufacturing. Participants were foreign investors and corporate executives from the US, UK, France and Southeast Asia, with some overlap between the two tours.
Olivia notes that most people would expect an AI and robotics itinerary to center on Hangzhou or Shenzhen. Wu says participants raised the same question. His answer is the talent pool: Beijing has Tsinghua and Peking, where Hangzhou has Zhejiang University, and he says the founding teams of Moonshot (Kimi) and Z.ai trained in Beijing’s AI research institutes. He adds that demand for robotics is heavily policy-driven, so proximity to government helps, and that commercial space is highly regulated, so companies stay near regulators. The internet era points the same way, since Baidu, Meituan and ByteDance were all headquartered there.
AI: three companies, three cultures
Moonshot AI (Kimi)
Wu’s account is that Moonshot differs from DeepSeek, whose focus he describes as “high intelligence with lowest cost.” Moonshot, in his reading, is aiming at the top of the intelligence range, building models with trillions of weights and emphasizing science and medical research. It is also an open-weight company that has worked closely with international cloud service providers. Wu describes a licensing arrangement in which heavy commercial users above a threshold share revenue with Moonshot. Because the models are hosted on the provider’s own infrastructure, he says, there is no concern about a Chinese company holding the data, and customers keep more control than they would with a closed-source vendor such as Anthropic. He judges Moonshot the best of the Chinese AI firms at working with international stakeholders.
The culture was young: staff mostly under 30, meeting rooms named after rock bands, and the group met in the Radiohead room because the founder loves rock music. The schedule shifted at short notice because the model companies were busy, which put Moonshot on day one.
Z.ai and a pivoted model company
Z.ai, maker of the GLM models, felt different: a full showroom and a practiced reception style for delegations, including local governments, which Wu notes matches the state-sector clients in Z.ai’s own base.
A third model company, which Wu chose not to name, stopped training its own models about two years ago, saying it lacked the money, and became an AI application and forward-deployed engineering company. Participants were skeptical of its product that gives CEOs visibility into everything in the company, because, as Wu puts it, some executives do not want to see everything and then carry the liability. Wu still counts the visit as useful for showing where different firms sit on the value chain and why they make different choices under different constraints.
“Different companies make different decisions given the different constraints.”
Robert Wu
Robotics: a bubble with real commercial ground
What worked in the real world
Wu says the highlights were not the dancing robots. The first was an underground pharmacy run by robots, serving a steady stream of Meituan and Alibaba delivery riders. He stresses it was a live operation, not a showroom. The case for automation is a pharmacy that should ideally run around the clock, where a human night shift is costly and arguably unnecessary.
Where it fell short
At a robot-staffed convenience store, a company employee asked the robot for a yellow juice, and the robot replied that it had nothing like that despite orange juice and other yellow bottles on the shelf behind him. Wu says that was a letdown for participants. His reading is that robots cannot yet operate freely across scenarios, but within a controlled setting, such as picking medicines or moving boxes out of a disorganized pile, some general capability is clearly developed.
Revenue without subsidy
Wu cites a large service-robot maker (Pudu Robotics) with revenue he puts above RMB 1 billion. Its core products are cleaning robots and cat-shaped dish-carrying robots popular in Japan, with humanoids a secondary line. Wu says its revenue is commercial, not driven by subsidies. The company survived an early existential crisis during Covid supply chain disruption by finding a use case in cleaning robots and by focusing on higher-margin international markets from the start. Wu believes it would survive even if the robotics bubble bursts.
“We shouldn’t overestimate the intelligence of robots at this stage, but also we shouldn’t underestimate what the robots are already able to achieve from a commercial perspective.”
Robert Wu
Pockets of durable value
Humanoid making is fragmented in Wu’s account, because components are available and almost anyone can assemble a robot. Two parts of the chain are different:
Dexterous hands. The top maker claims 70 to 80 percent of the market. Wu says the hand is the most complex and highest-value part of the body, and that no supply chain for its joints and gears existed, so the maker built from scratch in Beijing and location did not matter.
Gears. Wu visited a listed company and described gears as “the chips for the industrial world.” In his telling, it has become the least-cost producer in a product boring enough that newcomers are not drawn in, which creates a durable moat.
Data collection. Origin Flow, founded about a year ago and already through four funding rounds, makes a wristband that reads hand signals. It can serve as a computer input device for disabled users or as a data collection tool for training robot hands. It has launched a first kit and plans to extend to full-body data collection.
“All throughout the robotics value chain, you find these pockets of value that are not easily involuted away.”
Robert Wu
Implication for investors. Wu’s consensus view is that Chinese robotics is entering, or is already in, bubble territory, which he previously called “necessary” and says this trip reinforced. His practical filter favors firms with clear unit economics, concentrated market position or cost leadership, over humanoid demonstrations.
Commercial space: ambition capped by approvals
The planned company visits did not receive clearance, given the sensitivity of the sector. Instead, local government hosted a seminar where Chinese space firms and Singapore companies collaborating with them shared their views. Wu, who says he and most participants were less familiar with space than with AI or robotics, was struck by the number of companies making satellites, launch vehicles and rockets.
Why launch cadence lags SpaceX
Wu argues the constraint is institutional. Policy intent is to emulate SpaceX, but a failed launch raises the question of who is responsible, and a fatal one would raise it further. Rocket firms told him directly that they are not launching as often as they want because government approval is too slow. He expects progress to be gradual, not growth at any cost.
“SpaceX is different because it’s headed by this one audacious entrepreneur and he takes all the responsibility. But in the Chinese space industry, who is going to take the responsibility?”
Robert Wu
Musk as benchmark, and the manufacturing edge
Every company Wu spoke with treated Musk as a reference point. Olivia asks whether other industries benchmark against Sam Altman in the same way; Wu says not so much, because the gap with SpaceX in launch numbers is still huge. He rejects the claim that a recent Chinese sea recovery of a reusable rocket with a net-like device was stolen technology. What China takes from SpaceX, he argues, is an example proving something can be done. On advantages, he points to manufacturing depth: recovery ships are available and shipping firms are waiting for business, and some people predict Chinese rocket costs will fall below SpaceX’s within three to five years. That is a forecast attributed to others, not a confirmed plan.
Not every sector is full speed ahead
Wu links the space bottleneck to an April tour on robotaxis. Since the DeepSeek moment, he says, a stereotype has formed that China embraces anything futuristic at crazy speed. US visitors expected robotaxis everywhere, but deployment is confined to pilot zones, not because the technology is unready but because the government is concerned about employment and applies what he calls an artificial brake. His broader point is that the state balances many factors at once.
Niches over the mainstream: the SmallRig lesson
Wu describes SmallRig as a company he had not heard of and which is large in filmmaking and photography circles. It makes camera and phone rigs and accessories, works closely with camera makers such as Nikon, and has a global sales structure that includes a North America representative and a West Coast lead for the entertainment industry. It started in car infotainment systems; user requests for a camera mount began the current line. It has spent more than 15 years building a brand and a community, and budgets for free samples of new products. Asked why Nikon does not do the same, Wu says accessories are too small for a main equipment maker to care about, yet large enough to support a specialist.
This fits a thesis Wu has held for some time: in China, business and investment opportunities are better in niches than in mainstream markets, where competitors multiply and margins are compressed. He adds that SmallRig’s user co-design was a surprise to him and to Olivia, who says few Chinese brands build community into strategy.
The lunch that resonated most
In the end-of-trip survey and farewell dinner, more than half of participants named a lunch with a startup founder, a former co-founder of an AI company (unnamed), as the most impressive meeting, ahead of the large AI companies. His argument, as Wu relays it, is that large language models give a generalized, middle-of-the-distribution answer to everything and by design will not produce peripheral ideas, so businesses that are not OpenAI or Anthropic should look for their own peripheral niche. Wu calls it the most educational takeaway of the trip.
“Instead of asking, what’s the mainstream? We want to ask, what’s the best next peripheral niche out there.”
Robert Wu
Is China better placed to find niches?
Wu says everyone starts from an equal line. Western education and culture make it easier to be a maverick, but China has a large population, many self-motivated builders, and, in his words, no lack of creative companies focused on niches. He adds that Shenzhen’s complete supply chain lets a founder test and ship quickly; Origin Flow founded in Beijing but its team spends time in Shenzhen working with suppliers. He expects the same dynamic in knowledge work, where AI makes the foundational layer cheaper and the scarce input becomes the unusual idea. Olivia observes that this cuts against the perception that China is not creative, and that intense competition pushes companies to the periphery.
