Three numbers that are not the same number
A thousand robots built. More than ten thousand deployed. Fifteen thousand cumulative. These get stacked in the same paragraph as though they were comparable, and the comparison is where humanoid robotics coverage goes wrong.
BotQ built its thousandth Figure 03 on 23 July at roughly one robot per hour. Elsewhere Figure is reported at more than 10,000 deployments, and AgiBot at 15,000 cumulative. Three figures, three different quantities.
The three quantities
Units built is a manufacturing fact. It is verifiable at the factory, it says something real about production capability, and it says nothing about whether anyone is using the output.
Units deployed is a placement fact, and it is the softest of the three. It may mean sold, leased, pilot-installed, or sitting in a customer’s facility under an agreement that involves no payment at all. Without a definition, a deployment number is not checkable.
Cumulative units is a lifetime total, which quietly includes everything retired, returned, superseded or scrapped. It is the largest of the three by construction and the least informative about current state.
Reporting that stacks these in a league table is comparing a rate, a placement and a lifetime sum as though they were an inventory count.
The one number that is hard to fake
Roughly one robot per hour is a throughput figure, and throughput is the honest measure of an early manufacturing operation. It bounds annual output arithmetically, it can be improved only by real investment, and it cannot be inflated by redefining a word. If you follow one humanoid metric, follow that one.
Which explains the forecast spread
When the present is this poorly defined, the future is unconstrained — and the forecasts show it. Goldman Sachs puts humanoids at $38 billion by 2035; Morgan Stanley puts them at $152 billion by 2040. Different end years, so not directly comparable, but the implied trajectories are far apart regardless. Against roughly $4.3 billion of humanoid-specific startup funding, that spread is not a rounding difference. It is the whole investment thesis, unresolved.
What would settle it
Deployment definitions in the announcement rather than in a follow-up clarification. Utilisation figures — hours worked, tasks completed — rather than unit counts. And, eventually, repeat orders, which is the only number that demonstrates a robot did something worth paying for twice.
Until then the correct posture is that humanoid robotics has entered real manufacturing, at a rate that is measurable and modest, and that every larger number in circulation should be read with the question: counted how?
Technology.org — Humanoid Robots in 2026: What Is Actually Deployed → · VaaSBlock — Humanoid Robotics 2026: Figure, Optimus, 1X Commercial Reality → · EVST — Top 8 Humanoid Robot Companies to Watch in 2026 →