Europe can't own its robots, Tesla can't build them, and China just open-sourced the eyes: the supply chain triangle nobody can price

ABB agreed last October to sell its robotics division to SoftBank for roughly $5.4 billion, closing expected later this year and still pending. That follows KUKA, bought by China's Midea in 2017 and fully delisted in 2022. You cannot buy KUKA stock anymore. You can only buy Midea.

In the US, Tesla's Optimus reportedly fell far short of its \~5,000-unit 2025 target, with only hundreds built per The Information, and walked back firm timelines around the V3 redesign. The flagship humanoid project the market priced as inevitable no longer has a firm timeline.

In China, both legs are moving at once. AgiBot shipped its 10,000th mass-produced general-purpose robot in March, going from 5,000 to 10,000 in roughly three months. Unitree cleared final approval for its STAR Market IPO on July 3, on track to be China's first listed pure-play humanoid stock, with more than 5,500 humanoids shipped in 2025. The country installed roughly 295,000 industrial robots in 2024 versus about 34,200 in the US, per the IFR.

Then last week, almost buried under the Unitree IPO noise, Ant Group's robotics company Robbyant put LingBot-Vision live on Hugging Face under Apache-2.0. Per the company's own paper, none of it independently verified: four sizes from 21 million to 1.1 billion parameters. The 300-million-parameter model reportedly matches Meta's 7-billion-parameter DINOv3 on NYU depth estimation at roughly 23 times fewer parameters, though the family still trails DINOv3 on KITTI depth and ImageNet linear probing. What they kept closed is LingBot-Depth 2.0, the downstream product that handles glass and mirrors where standard depth sensors return nothing. Which reads to me like a classic commoditize-the-backbone play: give the foundation away so nobody else can charge for one, keep the revenue product.

Two numbers I cannot square. China installed nearly 9 times the industrial robots the US did last year. Yet by every account I can find, Fanuc and Yaskawa still command lead times and pricing power in automotive welding where defect tolerance is measured in microns. If hardware volume is this lopsided, how do Japanese margins on precision subsystems stay this sticky? Either the high end decouples from volume economics in a way I do not understand, or I am misreading how fast the floor drops out.

The counterargument I keep running into is Nvidia's Isaac platform, still the default stack for simulation training. The ABB sale to SoftBank is not closed, and regulatory review could stretch or alter the deal. US humanoid timelines may compress again if Tesla resolves its V3 redesign. So maybe the hardware layer was always going to consolidate to Asia and the real Western margin was always in simulation and precision subsystems anyway.

I keep going back and forth on whether any Western robotics name keeps pricing power in a world where the perception stack is free and the arms come out of Chinese factories at nine times US volume. Is there a single Western robotics name whose pricing power survives this, or is the honest answer none of them?