In-House Robot Models Narrow Market

This is really a race to own the learning loop, not just to ship a model. Skild’s core bet is that many robot makers will buy a shared brain and feed usage data back into one horizontal system. But Figure, Tesla, Apptronik, and Galbot are increasingly doing the opposite. They are putting their own robots into real jobs, collecting the failure cases, interventions, and task data themselves, then using that closed loop to improve a proprietary model that is tuned to their hardware and customer workflow.1 2 3 4 5 6

• Figure shows what this looks like in practice. Its BMW deployment logged 1,250 plus runtime hours and 90,000 plus parts handled, and Figure says every intervention and part loaded informed the next generation of Helix and Figure 03. That is training data Skild cannot access through an API relationship.2 4 5
• The customer may also prefer one vendor over two. Galbot sells the robot, the software stack, and support together, with recurring annual software and service fees per robot. Apptronik is following a similar path through direct pilots with Mercedes Benz and GXO. Once the OEM also owns the AI layer, a horizontal model provider can get cut out of the deal.3 6 7 8
• This does not mean the horizontal market disappears. It means it narrows toward everyone who cannot afford to build the whole stack. Anvil’s interview maps the market into brain companies, hardware providers, and solution companies, but it also makes clear that the companies closest to the customer often build the moat because they own deployment data and take reliability from roughly good enough to production grade.1 9

The next phase is likely a split market. The biggest humanoid and industrial robot fleets will train in house models on their own hardware and workflows, while horizontal providers win the long tail of OEMs, form factors, and regions that need a faster plug in brain. For Skild, growth increasingly depends on becoming the default layer for everyone outside the emerging full stack leaders.1 3 8 9