Three of the biggest American names in humanoid robotics made announcements in the same week. None of them showed its robot actually working. Figure AI paraded a factory, Apptronik an org chart, Meta an operating system. The question that ought to settle the market — what can the machine do on its own? — got, from the Western side, nothing but answers about everything except that.
Figure shows the throughput, not the task
On 29 April 2026 Figure AI announced that it had taken production of the Figure 03 at its BotQ factory from one robot a day to one robot an hour, a twenty-four-fold gain in throughput, achieved in under 120 days starting from January 2026. The company also claims a first-pass yield above 80% at end of line for the robots, and 99.3% for the batteries — manufacturer's figures, not those of a demonstrator.
The most telling detail comes down to a loop: BotQ slots Figure 03 units into its own assembly lines, where they handle key components and manage material flows. Robots building robots, in other words. And the metric that sums it all up is arithmetical: across the whole of 2025, Figure had produced roughly 150 units; the single month of April 2026 surpassed that annual total. As of 29 April, more than 350 Figure 03 units had rolled off the line.
All of this describes a foundry, not a capability. We learn how fast the object is cast, never what it accomplishes once set down at a customer's site. When you cannot film the use, you film the throughput.
Apptronik recruits a sales force, not roboticists
On 28 April 2026, Apptronik unveiled a wave of hires whose role-by-role reading is unambiguous. Daniel Chu becomes Chief Product Officer: he held the same title at Waymo, where he played a founding role in launching the world's first driverless autonomous transport service. Chirag Shah joins as VP Software, after leading multimodal AI for Alexa and Kindle at Amazon. Dave Perry takes marketing: an Emmy-award-winning executive, formerly with Paramount+ and Amazon, where he oversaw franchises such as Star Trek and Halo.
You don't build a team like that to solve a grasping problem. You build it to sell. The sequence follows in the wake of a $935m Series A closed in February 2026, at a valuation of around $5.3bn. And chief executive Jeff Cardenas sums it up in a sentence that says a great deal: the technology has, he suggests, "finally reached the level required" by the scale of the mission. That is exactly the pitch you make when a product is ready to be sold — not invented. Apptronik is professionalising the go-to-market for a robot it has yet to unveil.
Meta wants the OS, refuses the body
On 1 May 2026, Meta Platforms finalised its acquisition of the startup Assured Robot Intelligence (ARI), for an undisclosed sum. ARI, based in San Diego and New York, had around twenty employees and had developed a proprietary tactile sensor technology called "e-Flesh": 3D-printable microstructures paired with magnets and magnetometers.
But the point of the move lies not in the artificial skin; it lies in the positioning. Meta presents its offering as a software layer open to the whole industry, comparable to the role Android played in mobile telephony, and claims to be focusing on intelligence rather than on building the hardware. In other words: Meta is explicitly refusing to build the body. Here too, a non-answer to the task — this time in the shape of a platform.
The Chinese reverse shot: use, not packaging
That same week, the other end of the market was documenting precisely what the West was sidestepping. In logistics centres, RoboEra's humanoids reach up to 85% of human efficiency and run continuously, twenty-four hours a day. AiMOGA's police robots, for their part, have already been deployed in several Chinese cities before April 2026: guidance in Wuhu's school zones, support at the Jiangyin Marathon, assistance at the Jiangsu Urban Football League in Changzhou.
Those announcements put numbers on work actually done, in the field, under real conditions. It is exactly the register the three American releases of the week fail to occupy.
What the market actually puts a number on
One last data point hardens the angle. According to Roland Berger, by 2035 the segment of structural components and skeletons for humanoids could be worth up to $42bn, and that of hand systems and end effectors up to $26bn. The market being modelled is not the one for the robot that works: it is the one for the parts and the plumbing around it — the line, the brand, the OS.
The week's verdict then reads without any forcing. Unable, for now, to win on the only number that counts — that of units actually at work — the Americans have changed the ground. Figure wins the factory race, Apptronik the billboard race, Meta the operating-system race. Three ways of not saying "here is what my robot does on its own". This week, the American robot vanished from its own announcements. All that remains is the factory that builds it, the sales desk that sells it and the OS that drives it.