Three ways a humanoid gets paid for, priced in one week

Between August 28 and September 1, 2026, three sets of humanoid deployments were shown to be funded by three entirely different balance sheets. A government committed a budget line. A customer bought equity in its supplier. And a listed manufacturer reported what factories and other commercial buyers had paid it for 921 full-size robots. None of these is a better business than the others in the abstract, but they price the machine differently, and they fail differently.

The state

Korea’s 2027 budget proposal, released on September 1, lifts the physical AI line from ₩900 billion this year to ₩3.1 trillion, and itemises roughly 2,000 domestic AI robots for public use: 650 for research and education, 214 for defence, 369 for police and fire, 200 for care, and 210 for agriculture and food.

Procurement buys volume, which is what a young manufacturing base needs and what no commercial market is currently offering. What it does not buy is price discovery. A ministry with a unit target and a localisation goal will pay what the plan requires. The number that comes back — robots delivered, at a stated unit cost — tells you about the policy, not about the product’s value to a buyer who could have said no.

The customer

On August 31 Donut Robotics of Tokyo announced a capital and business alliance with Charm Care Corporation, an Osaka operator of paid nursing homes. Charm Care is subscribing to ¥300 million of Donut Robotics common stock, and has been trialling the company’s humanoid, cinnamon, in its facilities since July 2026 for conversation with residents, with nurse-call response and other facility tasks planned. The figures come from Donut Robotics’ own release; no independent coverage of the terms was found at the time of writing.

This is the most committed of the three, because the buyer has put capital at risk on the same asset it is being asked to operate. A care operator that subscribes to shares in its robot supplier has priced both the product and the supplier’s survival — and has an incentive to keep the deployment running past the point where a pilot would normally be quietly discontinued. The corresponding weakness is that one customer’s validated need is not a market, and equity from your only reference account is a concentration risk on both sides.

The invoice

UBTech’s interim results, released on August 28, show revenue from full-size humanoid products and services up 1,445% to RMB 590.3 million on 921 units sold, 46.5% of group revenue. Group revenue doubled to RMB 1.27 billion and gross margin rose to 44.7%, but the company still posted a net loss of RMB 338.8 million. KrASIA’s report ties the Walker S line to automotive, logistics and aviation work, including an Airbus purchase earlier in the year for aircraft-manufacturing trials.

This is the only one of the three where a buyer with alternatives — a human worker, a conventional robot cell, or nothing — signs an ordinary purchase order. Divided out, the segment brought in roughly RMB 640,000 per unit, though that figure includes services and is not a list price. The weakness is what the buyers are doing with the machines: purchases such as Airbus’s are framed as exploring whether humanoids fit a production line, and repeat orders at scale, not first orders, are what will show the price holds. The seller’s own loss says the price has not yet covered the cost of building the business.

What to watch

The three models converge only if a humanoid becomes cheap and reliable enough to be an ordinary operating expense — at which point procurement targets stop mattering and supplier equity becomes unnecessary. Until then, the honest way to read humanoid revenue claims is to ask which of the three the money came from, because they are not comparable and the industry routinely reports them as if they were.

The near-term test is repeat business. Korea’s procurement will deliver units regardless of what they are worth to an end user. What matters is whether UBTech’s full-year results show follow-on orders from the same industrial customers, and whether a second Japanese care operator writes a cheque into a robot maker without a government programme behind it. Both would be signs that a humanoid has been priced by someone with an alternative.