A year ago, when the investor Ryan Cunningham suggested a trip to China in a group chat of robotics backers, the reply was a joke about smuggling a humanoid home. This year, Business Insider reports, the jokes have turned into itineraries. Cunningham is organising an April trip for ten investors, and nearly every venture group he belongs to now swaps notes on who has been to Shenzhen and what they saw. There are organised tours charging around $10,000 a head, pairing factory visits with translators, five-star hotels, and dinners with local executives. The striking part is why they are going. It is not to find Chinese companies to fund. It is to see, up close, what their American portfolio companies are losing to.
The numbers behind the pilgrimage are sobering. According to the research firm Omdia, the Chinese firms Unitree and AgiBot together shipped 71% of the world's humanoid robots last year. A German think tank, the Mercator Institute, puts China's share of the key companies in the humanoid supply chain at 63%. Stanford's 2026 Emerging Technology Review notes that most American humanoids are still in development, not on the shelf, and that US firms largely depend on Chinese suppliers for parts or assembly. In a 2023 policy document, Beijing compared humanoids to computers and smartphones and set a target of deploying them across the economy by 2027. The lead did not appear overnight; China spent years building toward it while Silicon Valley's attention was elsewhere.
What changed the mood, several investors told the outlet, was DeepSeek. The Chinese model's 2025 release punctured a comfortable assumption: that China might out-manufacture the United States but America would always build the better AI. Once that belief cracked, the hardware gap became harder to wave away. "We knew they were ahead," said Neel Mehta of G2 Venture Partners, "but it doesn't really hit you until you're on the ground." At one site he watched hundreds of workers in pods teleoperating robots around the clock, generating the real-world training data that robotics models are starved for. If data is the true bottleneck, he concluded, China may end up with better models than anyone.
The visitors do not come home uniformly alarmed, which is what makes their accounts useful. Charly Mwangi, a former Tesla and Rivian executive now at Eclipse, offered the trip's sharpest line: "In the US, we've figured out vibe coding. In China, they've figured out vibe manufacturing." Yet he also watched a robot that could dance and perform martial arts but could not pick up a bottle and drop it in a bin. The physical supply chain is formidable; the machines' actual cognition, their ability to reason about the messy real world, is not obviously ahead. The gap is in atoms, not entirely in brains.
That distinction is where the story points next. The tours are already reshaping how these firms invest: both G2 and Eclipse say they now weigh a China-equivalent risk into every robotics bet, asking whether a US startup has any durable advantage against a rival that can build faster and cheaper. Washington is treating the same question as national security, with the FCC banning new foreign-made "advanced robotic devices" in July. The investors flying east are making a quieter calculation. Before you can decide whether to compete, you have to be honest about the size of the thing you are competing with, and that, it turns out, is worth $10,000 and a fourteen-hour flight to see with your own eyes.