For a month, this thread has dismantled the machinery of the humanoid lie layer by layer. First the rigged demo — the preloaded Tai Chi, the Atlas piloted through a VR headset, the prospectus backflip. Then the word: "autonomous", brandished five times in a single week and cracking at every third of a sentence. Then the stage: a televised gala where the broadcaster happened to be a shareholder in the very robots it was making dance. At each layer, the same reflex: step back a notch towards whatever seemed more solid. One last refuge remained, the only one thought incorruptible — the hard metric, the delivery figure, the number of units rolling off the factory floor. This week, the deception got in there too.

Between 27 January and 2 February, three players put out, in six days, three mutually exclusive "world number one" rankings. Each manufactured its own scoreboard. Stop asking who is number one: ask who is holding the calculator, and who is paying for it to show that number.

Three crowns for a single head

First to take the crown is AgiBot. On 27 January, the Chinese manufacturer issues an official statement citing a ranking from the analyst firm Omdia that places it first in the world for humanoid robot deliveries in 2025: 5,168 units, including 3,588 bipedal humanoids from the A and X series and 1,412 mobile manipulators from the G series. Three days later, on 30 January, AgiBot opens its first European event in Milan, at the Magna Pars venue, and arrives there girded with the same figure: 39% market share worldwide according to Omdia, out of a total of roughly 13,000 units sold over the year. The statistical coronation is not a PR detail: it is the very sales pitch for the entry into Europe. You are not selling a robot, you are selling a rank.

Except that the rank is contested — by the very company Omdia had ranked second. On 29 January, Unitree Robotics puts out its own statement and claims to have delivered more than 5,500 pure humanoid robots to customers in 2025, excluding wheeled robots. The firm had credited it with only 4,200. It is the first time Unitree has published delivery figures — and it does so not to submit to the benchmark, but to reject it publicly. The trade press recorded the rupture without ambiguity: Caixin runs "Unitree Defends Robot Sales as Rival Claims Market Crown", Gasgoo asks "Annual Champion Changes Hands?". The company, rather than entering the ranking through the door, has dismantled its lock.

The only man with no figure promises the biggest figure

On the other side of the Pacific, on 28 January, the third player plays a card of a different nature. During Tesla's fourth-quarter 2025 earnings call, Elon Musk admits he has been unable to provide any Optimus production figure for 2025, with actual output remaining well below the announced targets. He specifies that the few Optimus present in the factories perform only simple tasks — sorting battery cells, for instance — for learning rather than production purposes, and that the programme remains in a research and development phase: the robots are not deployed "in any material way". No significant volume is expected, he adds, before probably the end of 2026, and he gives no quantified production target for 2026.

The same man, on the same call, nonetheless drops the biggest figure of the entire week: a target capacity of one million Optimus units per year at the Fremont plant. To achieve it, he sacrifices his prestige cars. "It's time to basically bring the Model S and X programs to an end", he declares, adding: "We expect to wind down S and X production next quarter." The Model S and Model X are scheduled to stop in the second quarter of 2026, and the Fremont line will be repurposed for robot manufacturing. Zero verifiable units on one side, one million promised on the other — from the same mouth, in the same sentence.

The Davos promise collapses in the spreadsheet

The swing is all the sharper because it contradicts a scene this thread noted the previous week. At Davos, six days earlier, Musk was promising "complex tasks by the end of 2026" and a "public sale by the end of 2027". The earnings call of 28 January brings that promise back down to earth: no Optimus is doing any useful work today, and no volume will arrive before probably the end of 2026. The press flagged it bluntly — Electrek sums up: "Musk admits no Optimus robots are doing useful work — after claiming otherwise." The promise made on the stage does not survive the passage through the spreadsheet.

The pattern no one cross-checks: there is no consensus count

Line up these three claims and one detail leaps out: none rests on the same count. Omdia puts forward 13,318 units worldwide for 2025. The China Machinery Robot Association, for its part, estimates Chinese deliveries alone for the year at around 20,000 units — that is, on their own, more than Omdia's entire global total. The classification scopes differ: what is called a "delivered humanoid robot" does not have the same boundary depending on which office is counting. Unitree over-counts itself by 1,300 units relative to Omdia's ledger; Tesla does not count itself at all, for lack of production. The metric this thread spent a month digging out from under the folklore has, in its turn, become the folklore: every manufacturer is now its own statistics office.

This arithmetical disorder is not merely an accountants' quarrel. It thrives on a technical uncertainty that financial analysts, for their part, do not paper over. Wang Feili, an industrial analyst at UBS Securities China, reckons that "brain technology" remains the main bottleneck for the commercialisation of humanoids, and that an "electric car moment" for the sector is unlikely within the next five years. As long as real-world use stays thin, the only available battleground is the figure — and the figure, when no one arbitrates it, gets manufactured.

The calculator rather than the crown

After the demo, after the word, after the stage, it is therefore the number that lies. Three "world number ones" mutually exclude one another in the same week; the man who can produce no production figure promises the biggest figure of all. The only genuinely verifiable thing from these six days is, in fact, not a ranking: it is an admission — Musk's, conceding that none of his robots is working. The rest is a matter of competing spreadsheets.

The formula this thread has carried since December — do not ask whether the robot is autonomous, ask who is holding the stopwatch — moves up a notch. This week, the stopwatch became a calculator, and everyone brought their own.