The showroom is the strategy

Chinese automakers want their humanoids to meet the public. Not work behind the scenes on a factory line. Rather, be in the showroom, where potential customers are deciding whether to trust the badge on the grille.

Chery says its AiMOGA robot is already working in 4S dealerships¹ in Malaysia, greeting customers, explaining vehicles, serving drinks and arranging test drives. The Wall Street Journal reports that XPeng plans to put IRON in domestic stores as a shopping assistant in early 2027. BYD is the newest confirmation of this trend. In June, executive vice-president Stella Li told Business Insider she wanted two or three robots in every store. Four months later, China’s patent office published design drawings for what appears to be the Xiao Di robot BYD has shown publicly.

The industry treats humanoids as a solution to the factory labor problem. The showroom gets treated as a novelty. I think it is the strategy for these Chinese automakers.

BMW reported that Figure 02 handled more than 90,000 parts during its Spartanburg pilot, inserting sheet metal into fixtures for welding. Mercedes-Benz has been testing Apptronik’s Apollo in manufacturing settings in its Berlin plant. Hyundai says Boston Dynamics Atlas is planned for parts sequencing at its Georgia plant from 2028. These are sensible first jobs: known objects, repeatable movements, a workplace already designed around processes. The public is not part of the test. In a showroom, customers can ignore a robot, ask it something unexpected, or tell their friends about it. But they can’t not notice it.

Now look at who owns the robots. Chery has AiMOGA. XPeng built IRON. BYD is developing its own system. In the publicly documented cases, Chinese automakers using someone else’s humanoid, including NIO, Geely/Zeekr, Dongfeng and FAW-Volkswagen with UBTECH Walker models, have kept the work in factories. That pattern may change, but it reveals something already. You could buy a robot to move material. Or you could put your own robot in the showroom and make the machine part of the brand.

In issue 9, “Humanoid labor: same technology, different strategy,” we drew the line between automakers buying robots for efficiency and automakers building robots as products. But there's another split in the market: where they deploy first. A factory-first company is testing whether a humanoid can do a job. A showroom-first company is testing whether people will accept one as part of buying a car.

BYD’s patent did not invent this strategy. It did make the experiment hard to ignore. The showroom is where they find out whether people will share a room with one of these things and feel fine about it.

Physical Intelligence has raised more than $2 billion to build AI models for robots. It builds no robots itself. Its bet is that the intelligence can work across different machines, humanoid or otherwise. In February 2026, it published its strongest public evidence yet of what that intelligence can do in the real world: two continuous-shot videos of robots at work. One folded laundry. The other packed boxes in a warehouse. The robots in the footage weren't humanoids. Both were stationary systems built around specific tasks.

At Ultra, a Brooklyn startup, Physical Intelligence's π0.6 model ran through a continuous eight-hour shift at a live customer site, operating autonomously 96.4% of the time. This is a company whose investors include Jeff Bezos, Sequoia, Thrive and CapitalG, building the intelligence that could power a generation of robots. Its clearest public evidence of that intelligence working at scale came from machines that don't walk, don't balance, and don't need to navigate a room. They stay put and do a specialised job.

On October 9, Ultra announced $62 million in funding. Its chief executive, Jon Miller Schwartz, told Fortune that fixed industrial systems “are having more real-world impact than humanoid robots, which receive most of the attention.” That statement comes from the leader of a company deploying Physical Intelligence’s technology, not a critic taking a swipe at humanoids.

Physical Intelligence also works with humanoid makers, including AgiBot, so this isn't a company rejecting the form. It's a company that claims its intelligence can travel across hardware, but its own deployment evidence points to something less ambitious in both form and narrative. Stationary robots designed for one job have demonstrated what a general-purpose humanoid is supposed to deliver: useful work, sustained over a real shift, outside a controlled demo.

The core argument for humanoids is that a machine shaped like a person promises to operate in spaces built for people, using tools designed for people, doing jobs that vary from hour to hour. That flexibility justifies the added complexity. Physical Intelligence’s investors are financing that future. But the benchmark its own technology has set in production, 96.4% autonomy over a full shift, belongs to a machine that doesn’t walk, doesn’t balance, and never leaves its station.

This week's Signal is about where Chinese automakers are actually putting their humanoids: the showroom, not the factory, and it's not an accident. Below the Fold is about what Physical Intelligence's own deployment evidence says about the humanoid form factor its investors are funding.