Capital arrives before capability
$8.6bn into humanoids this year, 1.8× all of 2025. The deployments are real, narrow, and in fixed cells. The valuations are underwritten by a generalisation nobody has demonstrated.
Humanoid startups have raised $8.6bn this year — 1.8× all of 2025 — inside a robotics sector at $55.8bn, nearly double its previous record. One company is at roughly $2.34bn raised on a $39bn valuation.
What is actually deployed
Humanoids are working at BMW, Mercedes-Benz, Hyundai and in Amazon facilities, on defined tasks: components into constrained spaces, quality inspection, materials handling. Fixed layouts, repeatable motions, structured environments. That is conventional industrial automation wearing a different chassis.
A $39bn valuation is not underwritten by installing a part in a fixed cell.
It is underwritten by the belief that the same platform generalises to unstructured environments. Nobody has demonstrated that at scale. The bet is defensible — the automotive work generates exactly the data the generalisation problem needs, in volume, under real conditions. It is still a bet, and the same argument was made for autonomous driving in 2018, where the unstructured last mile turned out to be most of the distance.
The platform work is getting more honest
Sensing skin gets a fraction of the attention that manipulation does and is closer to the actual constraint. Current safety practice is spatial — cages, light curtains, separation zones — which works by keeping the robot away from people. Contact sensing lets force limits replace distance, which is what "alongside" requires.
The remote-operator approach is the other honest answer: skip autonomy, keep a human in the loop, and hope the economics close when one operator supervises many machines. That has a long record of working and an equally long record of unit economics that only work at a ratio nobody has hit.
What this asks of you
Capital ahead of capability is not automatically a bubble — it is also how capital-intensive categories get built. The distinguishing evidence is unit counts outside automotive, and contact-safety hardware shipping rather than being demonstrated.
Watch those two numbers. Funding totals tell you about conviction, not about robots.
Humanoid Index — Humanoid Robot Funding Tracker 2026 → · Technology.org — Humanoid Robots in 2026: What Is Actually Deployed →