The Missing Price: What Non-Disclosure Reveals About the Humanoid Market

Data as of September 2026.

Most pricing coverage in robotics reports the numbers that exist. The more useful signal sits in the numbers that do not.

We maintain a census of commercially announced robot models: 131 models from 80 makers across 15 countries, each row carrying a named source, a source URL, and an as-of date. Of those 131 models, 33 publish no price at all.

Which 33 turns out to be highly non-random. Non-disclosure is not distributed evenly across categories, across sales models, or across countries, and the pattern it forms says something specific about where each part of the market sits in its commercial life. One of those patterns runs directly against the prevailing assumption about who is open and who is not.

What counts as a price

A model enters the index when a manufacturer has publicly announced it as a product, including machines that are demonstrated but not yet sold. A price is recorded only when the manufacturer or a named, checkable source has published a figure for a specific configuration. Where a maker publishes a range, the index records the floor.

Everything else is recorded as having no public price, whether the vendor says "contact us," publishes nothing, or discusses cost in interviews without committing to a number. That last category matters. Several makers have mentioned target prices in press interviews. A target is not a price, and the index does not treat it as one.

Disclosure falls as category maturity falls

Splitting the index by category produces an ordering that is hard to read as coincidence.

The sequence tracks how long each category has had a functioning transactional market. Industrial arms have been bought and sold on published price lists for decades and are almost fully transparent. Quadrupeds have had a commercial market since the late 2010s. Service robots are more fragmented. Humanoids, the newest of the four, are five times more opaque than industrial arms.

Two smaller categories in the index do not fit neatly and are reported here rather than omitted: autonomous mobile robots show 3 of 6 models unpriced and exoskeletons 1 of 6. At n equals 6, neither cell can carry an argument, and the AMR figure in particular should not be read as a 50 percent opacity rate for that industry. They are noted because excluding inconvenient cells is how gradients get manufactured.

Disclosure tracks the sales model almost perfectly

The stronger relationship is not with category at all. It is with whether the machine is actually for sale.

Of 74 models offered for open sale, exactly one lacks a published price, a research platform from PAL Robotics in Spain. Of the 33 unpriced models in the whole index, 21 are enterprise-only and 9 are not sold to anyone.

This reframes what non-disclosure means. It is rarely a pricing strategy in the competitive sense. It is mostly a description of commercial stage. A company that has built a repeatable product with a configuration it can ship publishes what it costs, because it has to in order to sell it. A company still running pilots has no stable configuration to price, and "contact us" is an accurate description of its sales process rather than an evasion.

The practical consequence for buyers is that the absence of a price is information. It usually indicates that the vendor is not yet selling a defined product, and a procurement conversation will be a pilot negotiation rather than a purchase.

Geography: where the assumption breaks

The prevailing narrative holds that Chinese manufacturers are the opaque ones and Western vendors are the transparent ones. The census shows the reverse.

American makers are more than twice as likely to withhold a price as Chinese ones. Narrowed to humanoids, the gap widens: 6 of 10 US humanoid models carry no public price against 11 of 40 Chinese models, or 60 percent versus 28 percent.

There is an obvious objection, and it should be tested rather than waved away. The US set contains more research and internal-use machines, which naturally have no price. Tesla's Optimus, Boston Dynamics' Atlas, and Berkeley Humanoid Lite are all in the index as announced models and none is for sale.

Removing every not-for-sale machine from both sides and comparing only commercially offered models does not rescue the assumption. It widens the gap:

Among humanoids specifically, restricted to commercially offered machines, 3 of 6 US models are unpriced against 7 of 36 Chinese models. And within the enterprise-only tier, where both sides sell by quotation, Chinese vendors are the more forthcoming: 7 of 16 Chinese enterprise-only models disclose a figure against 3 of 11 American ones.

Kepler publishes $20,000 for the K2 Forerunner. UBTech publishes $68,000 for the Walker S2. Fourier publishes $125,000 for the GR-2. These are enterprise-sold machines with published numbers. On the American side, Agility's Digit at roughly $100,000 is the exception rather than the pattern, while Figure and Apptronik publish nothing for Figure 02, Figure 03, or Apollo.

The likely explanation is commercial posture rather than culture. Chinese humanoid makers are competing for unit volume in a market where price is the primary lever, and several sell through public storefronts. The best-funded American developers are competing for enterprise contracts and capital, where a published price sets a ceiling on negotiation and a floor for investor expectations. Disclosure follows the business model.

The quote-only premium

Humanoids are the one category with enough priced models across every availability tier to compare like with like.

Among humanoids sold openly, the median entry price is $23,120, spanning Unitree's R1 at $4,150 to PNDbotics' Adam at $100,000. Among enterprise-only humanoids that do disclose, the median is $60,400.

That is a 2.6x difference for machines in the same category. The caveat belongs in the sentence rather than a footnote: the enterprise-only median rests on 6 disclosed models out of 17, and the 11 that disclose nothing are plausibly the expensive end, which would push the real gap wider rather than narrower.

The gap is not simply a capability premium. It reflects what the buyer is purchasing. An openly sold humanoid is a catalog item shipped in a defined configuration. An enterprise-only humanoid usually arrives with integration work, support commitments, and a pilot structure attached, and the quoted figure bundles all of it. Buyers comparing a $23,000 catalog price against a $60,000 quotation are frequently not comparing the same transaction.

What this predicts

The reading above produces a testable claim rather than a rhetorical one.

If non-disclosure is a function of commercial stage rather than strategy, then as the well-funded American developers move from pilot deployments to volume shipment, they should begin publishing prices, and the US-China disclosure gap should narrow materially within twelve months. If instead the gap holds or widens while those companies announce shipments, the explanation is strategic rather than developmental, and the maturity reading in this piece is wrong.

That is the specific outcome we will be checking against the September 2027 cut of the index. Either result is informative. The second would be more interesting.

Methodology and limitations

This is a census of commercially announced models rather than a sample, so there is no sampling error, but there is a coverage question. A model enters when a manufacturer publicly announces it as a product and is removed when the manufacturer discontinues it. Coverage is strongest where manufacturers publish in English or Chinese and weakest for regional makers who publish in neither, which likely understates the count of smaller Japanese, Korean, and European models.

Every figure is the lowest publicly documented configuration price, sourced to a named publication or the manufacturer's own storefront, with an as-of date. All figures in this article are from the September 2026 cut. Taking the floor of published ranges biases every median in this piece downward, consistently across categories, which preserves comparisons between them while making absolute levels conservative.

Country is assigned by the manufacturer's headquarters, not by place of manufacture. Several models designed in one country are built in another, and a country-of-manufacture cut would produce a different and also defensible table.

Cells below roughly ten observations are labeled in the text wherever they appear. The two findings that carry the most weight, the sales-model relationship and the geographic gap, rest on 131 and 103 observations respectively and survive the removal of not-for-sale machines. The quote-only premium rests on 6 disclosed enterprise models and should be treated as directional.