Three of the most watched humanoid programmes now have overlapping release windows: Tesla’s Optimus V3, Figure’s 04 and Apptronik’s next Apollo. All three land inside a twelve-month stretch. The interesting question is not who ships first but what the three of them will be measured against.
Each has staked a different claim, and each claim is falsifiable.
Tesla is selling credibility, not hardware
Musk said on the third-quarter call that a video of Optimus performing non-preprogrammed actions would appear within one or two months, which places it in November or December. His phrase was a demonstration people can believe. The shareholder meeting on 3 December is being treated as the likely checkpoint.
That framing is an admission. For two years the most shared humanoid clips have been choreographed dances and sparring routines, and Musk himself has said many viral demonstrations are either scripted or teleoperated. A video of a machine deciding in real time on a task it has not seen is a different category of evidence. It is also unfalsifiable in the other direction: a single successful clip proves capability, not reliability.
Figure is selling a production line
Figure locked the 04 design in May and moved into production preparation. Chief executive Brett Adcock has called it the company’s iPhone 1 moment, and outside expectations put the launch between late 2026 and early 2027. The company’s stated target is 50,000 units a year.
That number is the real claim. If it holds, Figure becomes the first US company to move a humanoid from laboratory object to catalogue product, and the signal for component suppliers is order visibility rather than another concept video. Design lock six months ahead of Tesla’s demonstration tells you where the company thinks the risk sits.
Apptronik is showing the second tier’s problem
Apollo 2 shipped quietly in June. The chief executive describes the next generation as a large step up, and the realistic window in industry eyes is 2027.
That timing is the tell. Being squeezed on both mass production and generalisation compresses the window for everyone outside the top three, and Apptronik is the strongest of that group. A 2027 launch in a market where Figure claims 50,000 units a year is a different competitive position than a 2027 launch in an empty field.
What the market is actually pricing
Primary market financing in embodied intelligence has visibly slowed, with term sheets now asking whether delivery happened and whether anyone paid. Secondary market money has moved from concept exposure towards companies with a dated production milestone.
Tesla’s demonstration answers whether generalisation is real. Figure’s line answers whether volume manufacturing is real. A slip on either one reprices the whole sector, because the two claims have been used interchangeably to justify the same valuations.
The 3 December shareholder meeting and the Figure 04 line start are the two dates that matter before the year ends. A convincing clip from Tesla without a production number, or a production number without a generalisation claim, would leave the market with exactly half of what it has been paying for.
Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-11/ART-898890-8110-30710156.html.