The Humanoid Robot Commercialisation Gap

The capability in humanoid robotics is running well ahead of the offer. Blazon has watched this pattern across a decade of frontier hardware launches, and humanoids are the sharpest version of it we have seen.

Read the sector through what the companies themselves publish, rather than through demo footage, and the picture is unambiguous. Very few humanoid robots are being sold to arms-length buyers at a published price. A small number of industrial deployments are real, contracted and conditional.

Most of the headline numbers are manufacturing counts, capacity targets, order books with milestones attached, or pilots that the buyer explicitly calls a pilot. Two exchange-listed Chinese companies are selling humanoids and filing the numbers. The Western deployment that both buyer and vendor call commercial sells hours of work rather than a machine.

That is the commercialisation gap, and it is a positioning problem before it is an engineering problem.

This is not a roster of interesting companies. If you want that, we already published one: the consumer robotics startups worth watching covers who is building what. This piece asks the buyer's question instead: what has each company actually published, what does that number mean, and what would have to be true for a humanoid to pay for itself?

Manufactured, deployed and sold are three different sentences

A useful sentence for any buyer sits on Figure's own website. Describing output from its BotQ factory, Figure states that robots shipped from BotQ are "allocated to internal research and development groups, data collection, efforts for robots to perform end-to-end housework, and commercial use-case development". On the same page it reports raising its build rate from one Figure 03 per day to one per hour, a 24x throughput improvement in under 120 days, and shipping over 350 third-generation robots from BotQ.

Read those two facts together. The manufacturer is saying, in its own words, that its output goes to internal research and to developing commercial use cases.

A robot that has been manufactured has not necessarily been deployed. A robot that has been deployed has not necessarily been sold. Almost every widely circulated humanoid statistic collapses those three states into one, and the companies themselves are more careful than their coverage.

Figure announced BotQ in March 2025 with a first-generation line "capable of manufacturing up to 12,000 humanoids per year" and a supply chain it said could "easily scale to 100,000 robots or 3,000,000 actuators in the next four years". Those are line capability and forward plan, not units produced.

In September 2025 Figure closed a Series C exceeding $1 billion in committed capital at a $39 billion post-money valuation, led by Parkway Venture Capital with Brookfield, NVIDIA, Intel Capital, Salesforce, T-Mobile Ventures and Qualcomm Ventures among others. The announcement contains no deployment numbers, no customer list and no production figures.

A billion dollars of capital raised without a single operating metric in the release is itself a data point about where the category sits.

Boston Dynamics makes the same distinction even more plainly. In January 2026 it announced that production of the new fully electric Atlas would begin at its Boston headquarters immediately, and stated that "all Atlas deployments are already fully committed for 2026, with fleets scheduled to ship to Hyundai's Robotics Metaplant Application Center (RMAC) and Google DeepMind". Additional customers are planned for early 2027.

No pricing and no unit quantities are published. Hyundai Motor Group is Boston Dynamics' majority shareholder and Google DeepMind is its AI partner, so a fully committed production year is an intra-group and partner allocation, not an order book from independent buyers. The Register noted at the time that Boston Dynamics declined to name any 2027 customers and gave no concrete 2026 volume.

The Spartanburg pilot, read from the source

BMW Group publishes operating hours and component counts rather than robot counts, and its release is where most of the sector's scale claims originate. BMW describes its Spartanburg humanoid work as a completed pilot project.

Figure 02 ran there during 2025 on ten-hour shifts Monday to Friday, accumulating approximately 1,250 operating hours, moving more than 90,000 components, performing precise removal and positioning of sheet metal parts for the welding process, and supporting production of more than 30,000 BMW X3 vehicles.

Note what is absent. BMW publishes hours, components and vehicles. It does not publish how many robots were involved.

Any robot count you have read for Spartanburg is inferred, not disclosed.

BMW has since deployed the next-generation Figure 03 at Spartanburg on a different task, sorting unsorted components into sequencing trolleys for just-in-sequence delivery to the assembly line.

Figure's founder and chief executive Brett Adcock says the eleven-month Figure 02 deployment proved that humanoids are no longer lab experiments. BMW's own release is drier, and puts the work at within ten months. No unit count is given for Figure 03 either.

BMW's first European humanoid pilot is not Figure at all. At Plant Leipzig it is testing the AEON robot from Hexagon.

BMW's release puts the initial test deployment in December 2025, a further test deployment from April 2026, and the actual pilot phase starting in summer 2026, covering assembly of high-voltage batteries and component manufacturing. BMW calls it a pilot and gives no robot count.

One buyer, two vendors, running in parallel, on different tasks, in different plants, with no quantities published. That is evaluation behaviour, not procurement behaviour.

It is also a reminder that the field is not a straight race between venture-funded pure plays. Hexagon is an industrial metrology and sensor group with existing factory relationships, and it is running alongside the most heavily capitalised humanoid startup on this list, Figure at a $39 billion post-money valuation.

The one published cost figure in the transaction materials

Agility Robotics is heading for public markets through a merger with Churchill Capital Corp XI at a $2.5 billion pre-money equity value, with more than $620 million of expected gross transaction proceeds including approximately $200 million of PIPE financing led by Foxconn, expected to trade on Nasdaq as AGLT. Going public forces disclosure, and the disclosure is genuinely useful.

The investor presentation filed for the transaction puts the current Digit v4 bill of materials at approximately $125,000, with expected reductions as production volumes increase. That figure sets the floor under every payback calculation anyone wants to run.

Agility also reports more than $300 million in multi-year contracted Digit v5 orders, more than 65,000 hours of operation, and a growing pipeline of over 30 customers, with GXO Logistics, Schaeffler, Toyota Motor Manufacturing Canada and Mercado Libre named. Agility announced in November 2025 that Digit had moved more than 100,000 totes at GXO's Flowery Branch facility, a cumulative total across a deployment that began in June 2024 rather than a single year's throughput.

The qualifier matters more than the headline. The transaction materials describe that $300 million as orders "subject to the realization of certain contractual milestones".

It is contracted backlog with conditions attached, not recognised revenue. Agility's actual current revenue is not disclosed in the coverage.

Worth noting honestly: Agility's press release lists Amazon among its strategic investors and partners, while the investor presentation filed for the same transaction lists Amazon among the enterprises Digit v4 works for. The two documents do not agree.

The commercially interesting part predates all of that. GXO and Agility signed what both companies describe as the industry's first formal commercial deployment of humanoid robots and the first Robots-as-a-Service deployment of humanoids, putting Digit into commercial operations at a GXO facility near Atlanta in June 2024 alongside Agility Arc, Agility's cloud fleet management platform, handling repetitive work such as moving totes from cobots onto conveyors.

Robots-as-a-Service is the answer to a $125,000 bill of materials. The buyer does not purchase a robot, evaluate its residual value, or carry it on the balance sheet. The buyer purchases hours of work against a task it already costs.

That is a pricing and packaging decision, not a robotics one. It is a clear example of a company solving a commercial problem rather than waiting for the hardware cost curve to solve it.

What each company has actually published

The two exchange-listed Chinese companies filing the numbers

UBTECH announced in November 2025 that its Walker S2 humanoid had begun mass production and delivery, with a first batch of several hundred full-size industrial humanoids, orders since early 2025 exceeding 800 million yuan, approximately US$112 million, a target of delivering 500 units within the year, and capacity targets of 5,000 units annually in 2026 and 10,000 in 2027.

Named customers include BYD, Dongfeng Liuzhou Motor, Geely Auto, FAW-Volkswagen Qingdao, Audi FAW, BAIC New Energy, Foxconn and SF Express, and a Zigong data centre contract was valued at 159 million yuan. UBTECH is listed on the Hong Kong exchange, so those figures sit alongside filed accounts.

Be precise about what each number is. 800 million yuan is orders received, "several hundred" is a delivery batch, and 500 units is a target.

Unitree listed on the Shanghai STAR Market, pricing its IPO at 150.80 yuan per share on 6 August 2026. It reported 2025 revenue of 1.699 billion yuan and guided to first-half 2026 revenue of 1.052 to 1.128 billion yuan, growth of 35.62 to 45.41 percent year on year.

Unitree also launched the R1 humanoid in July 2025 at US$5,900, standing about 1,210mm, weighing about 25kg, with 26 degrees of freedom and a built-in multimodal model for voice and image interaction, positioned at individual developers and early-stage research teams rather than at households.

Both companies did two unglamorous things. They published numbers anyone can check, and they sold a defined product at a defined price to a named buyer. That is the difference between selling a robot and selling a programme.

The consumer promise, in the vendor's own words

1X launched NEO in October 2025 at $20,000 outright or $499 per month, taking a $200 refundable deposit, and says it sold out its entire first-year production capacity of over 10,000 units in five days. A sell-out of capacity against refundable deposits is a reservation book. It is a strong demand signal and it is not revenue.

In April 2026 1X opened its NEO factory in Hayward, California, with stated capacity of 10,000 units annually and plans to reach 100,000 or more by the end of 2027. In that same release, 1X states that robots are "being shipped to the company's R&D Lab and Internal Home Testing" and that "first customer shipments of NEO planned for 2026".

Six months after a sold-out consumer launch, the manufacturer's own words put customer shipments in the future tense.

1X is also candid about autonomy. Its product page says that "for complex tasks NEO doesn't know, an Expert from 1X can remotely supervise its actions at scheduled times to help it learn new abilities and get the job done", that owners can "pilot your NEO from anywhere in the world through your Mobile App & VR device", and that NEO "arrives with basic autonomy for early owners and grows in capability overtime".

The consumer humanoid currently on sale is a partially teleoperated device, and that is the vendor's own description, not a critic's. It has generated a genuine privacy debate about remote operators having visual access to buyers' homes, which 1X's leadership has publicly defended.

Europe, the UK, and the order language problem

Humanoid, a London company founded in 2024 and building the HMND 01 platform with its KinetIQ AI system, raised a $152 million Series A at a $1.35 billion post-money valuation in July 2026, taking total raised to $270 million. The round was led by Prime Movers Lab with Schaeffler, Bosch, Fubon Financial Holding Venture Capital and Aglaé Ventures. Bosch is its contract manufacturing partner.

Humanoid's own announcement describes its Schaeffler arrangement as "the industry's largest publicly announced commercial agreement with Schaeffler for the large-scale deployment of thousands of humanoid robots in manufacturing environments". It gives no unit quantity, no contract value and no delivery schedule, and says beta robots roll out in Q4 2026 with deployment at several early access customers.

Schaeffler is both an investor in Humanoid and an announced customer of Humanoid, and separately a named Agility customer. Strategic investor as customer is a recurring pattern in this sector and it should be named plainly whenever a customer list is read.

Apptronik shows the depth of industrial capital behind the category. In March 2025 it closed an oversubscribed $403 million Series A, adding $53 million to a previously announced $350 million round, with B Capital, Capital Factory, Google, Mercedes-Benz, Japan Post Capital, ARK Invest and others, naming commercial relationships with Mercedes-Benz, GXO, Jabil and NVIDIA.

A further raise announced in February 2026 took its Series A past $935 million and its total capital raised close to $1 billion, with AT&T Ventures, John Deere and the Qatar Investment Authority joining.

Where the public record is genuinely thin

It is worth being explicit about what cannot be established from primary sources, because a great deal of humanoid commentary treats these as settled.

Total sector shipment figures for 2025 vary wildly between secondary estimates and sit behind paywalled vendor reports. Market sizing forecasts disagree with each other by orders of magnitude.

Per-company shipment splits circulate without traceable sources. Widely repeated unit counts for named agreements, including the Schaeffler arrangement, do not appear in the companies' own announcements.

Tesla illustrates the gap between target and delivery. Musk predicted in early 2025 that Tesla would build roughly 10,000 Optimus robots that year with around 1,000 doing productive work by year end, which did not happen, and in 2026 he said production would be "extremely slow at first" and effectively impossible to predict.

Safety standards are the other open question. There is no published safety standard specific to humanoid robots. ISO 25785-1, covering safety requirements for dynamically stable industrial mobile robots, is still in development as of writing, and the leadership of the US delegation working on it is drawn from Agility Robotics, Boston Dynamics and the Association for Advancing Automation.

Separately, Figure's former head of product safety, Robert Gruendel, has filed a whistleblower lawsuit alleging he was fired after raising safety concerns, allegations Figure denies, saying he was terminated for poor performance and that his claims are falsehoods it will discredit in court. Those are unproven allegations on both sides.

The relevant point for a buyer is structural rather than about any one company. A category selling into factories and homes does not yet have a finished standard to certify against.

What would have to be true for a humanoid to pay for itself

Strip the sector back to the buyer's arithmetic and four things have to hold.

The hardware cost has to be recoverable. The one hardware cost figure in this article is Agility's approximately $125,000 bill of materials, and the honest answer is that nobody has published a verifiable cost-per-hour comparison against human labour. Anyone quoting a payback period is estimating.

The task has to be narrow and already costed. Every real deployment in the public record is a defined, repetitive material-handling job: totes onto conveyors, sheet metal into a welding cell, components into sequencing trolleys. Nobody is buying general capability.

The commercial structure has to remove the capital decision. Robots-as-a-Service exists precisely because a six-figure machine with an unproven service life is a difficult purchase order and an easy operating expense.

The vendor has to survive procurement. That means published specifications, a defined SKU, a price or a rate, safety documentation, service terms and references. This is the part most of the category has not done, and it is not an engineering task.

What this means if you are launching hardware

Blazon does not make a humanoid pay for itself. What we do is the part this category is visibly losing: turning a capability into an offer a buyer can evaluate.

That work is concrete. It is deciding whether you sell a machine, a subscription or an outcome, and pricing accordingly, which is where our strategy engagements start. It is building proof assets that survive a procurement conversation rather than a conference demo.

It is running PR and paid media that reach an operations director with a labour line to defend, not a robotics enthusiast with a YouTube subscription. It is standing up the commerce, pre-order and reservation layer for anyone who does reach a shippable product, which is exactly what a hardware launch agency is for.

There is a specific and growing fit at the accessible end of the market. The R1 at $5,900 and NEO at $20,000 are real developer and consumer transactions, and the sub-$10,000 tier of advanced hardware is territory Blazon covers directly through 500+ product launches and 300+ crowdfunding campaigns, with $250M+ in sales driven and $120M+ raised.

1X's $200 refundable deposit is The Deposit Test run at scale. Ten thousand people paid something to hold a place, which tells you far more than any waiting list, and it is the mechanism we build into every crowdfunding and pre-order launch we run.

If your product is closer to a research platform or a frontier technology sale into enterprise, the positioning work sits with our deep tech agency practice instead. The mechanics of getting a physical product to market are covered in more depth in our guide to launching a hardware product.

Our minimum engagement is $15,000, and a positioning and offer-design sprint is the right place to start rather than a full launch programme. Get in touch through the hardware launch agency page and tell us what you have built and who you think buys it.

If you can demo brilliantly and cannot yet explain in one page who writes the cheque and why, you do not have a launch problem yet. You have a positioning problem, and of the two it is by far the cheaper to fix.

Frequently asked questions

Are humanoid robots actually being sold today?

A small number are. UBTECH has announced mass production and delivery of a first batch of several hundred Walker S2 units to named automotive and logistics customers, with orders exceeding 800 million yuan. Unitree sells the R1 at US$5,900 and reported 1.699 billion yuan of revenue for 2025. 1X has taken reservations for NEO at $20,000 or $499 per month. Most other headline figures describe manufacturing output, capacity targets or conditional orders rather than completed sales to independent buyers.

How much does a humanoid robot cost?

Published prices are rare. Unitree launched the R1 at US$5,900 in July 2025. 1X lists NEO at $20,000 outright or $499 per month. On the industrial side, Agility Robotics has disclosed a Digit v4 bill of materials of approximately $125,000, which is a manufacturing cost rather than a customer price, and its GXO deployment is structured as Robots-as-a-Service rather than a unit sale. Figure, Boston Dynamics and Apptronik publish no prices.

What is the difference between a humanoid pilot and a commercial deployment?

A pilot is an evaluation with a defined end, usually run alongside other vendors, and buyers rarely publish unit counts for them. BMW calls its Spartanburg Figure work a completed pilot project and its Leipzig work with Hexagon's AEON a pilot, publishing operating hours and components moved but never a robot count. A commercial deployment has a contract, a commercial structure and a stated commitment, as with the GXO and Agility Robots-as-a-Service agreement, which both companies describe as the industry's first of its kind.

Is the 1X NEO fully autonomous?

No, and 1X says so on its own product page. NEO "arrives with basic autonomy for early owners", owners can pilot it remotely through a mobile app or VR device, and for complex tasks a remote Expert from 1X can supervise its actions at scheduled times. That human-in-the-loop model has prompted a public debate about remote operators having visual access to buyers' homes, which 1X's leadership has defended publicly.

Which humanoid companies publish numbers anyone can check?

The exchange-listed ones. UBTECH is listed in Hong Kong and publishes order values and customer names. Unitree is listed on the Shanghai STAR Market and publishes audited revenue. Agility Robotics is heading to Nasdaq through a SPAC merger, which is why its bill of materials and its milestone-conditional order book are public at all. The privately held companies with the largest valuations publish far less operating detail.

Is there a safety standard for humanoid robots?

Not a published one specific to humanoids. ISO 25785-1, covering safety requirements for dynamically stable industrial mobile robots including legged machines, is still in development as of writing, and the leadership of the US delegation working on it is drawn from Agility Robotics, Boston Dynamics and the Association for Advancing Automation. The absence of a finished standard is a live commercial issue for any vendor selling into regulated industrial environments.

What should a robotics company do before it has a shippable product?

Define the offer. The companies that have converted did not wait for better hardware, they sold a defined thing at a defined price to a named buyer, or they removed the capital decision entirely with a service contract. In practice that means choosing the commercial model, writing the specification and the price, building proof assets that survive procurement, and reaching the operations buyer rather than the technology audience. That is the work Blazon does through our hardware launch agency practice, and it starts with positioning rather than production.