Embodied AI Crosses a Watershed, and Investors Everywhere Want to Sell Their Old Shares

At a group interview during this year's Bund Summit, investors from leading institutions described embodied AI as one of their main investment directions. But when a host raised recent reports that regulators were tightening IPO approvals for embodied AI companies, the room went quiet. Investors then acknowledged that the industry remains very early-stage and that the technology has not converged. The same morning, reports said regulators had informally signaled that only humanoid robot firms with recurring revenue, narrowing losses, or real technical innovation would be considered for IPO approval. The sector had been overheated for more than a year. Domestic embodied AI financing in the first half of 2026 exceeded 90 billion yuan, five times the same period of 2025, and more than 20 companies were newly valued above 10 billion yuan. Then in May UBTech launched an 'adults-only' companion robot that blurred the line between humanoid robots and adult products, and in August Unitree's market value nearly halved within five trading days of its A-share listing, deflating expectations. Primary-market enthusiasm then cooled. Investor Zhang Xin (a pseudonym) invested in a world-model startup at a 3 billion yuan post-money valuation earlier this year; by October the company was valued at 22 billion yuan. He is now trying to sell his stake at a 12 billion yuan valuation but has not yet found a buyer. Because tighter IPO rules have lengthened exit timelines and raised risk, more investors want to cash out earlier. Industry players say the main obstacle to commercialization is that the technology has not converged. Some firms have also booked data-collection-center orders as revenue, raising concerns about inflated sales through circular transactions. Intention orders for humanoid robots in China exceeded 120,000 units in January 2026, yet global shipments in 2025 were only about 18,000. Investors say the sector now looks most attractive in core component makers with high technical barriers.