There is an obvious problem with selling a humanoid robot to a manufacturing company in 2026. The robot costs somewhere between $70,000 and $200,000. It has been demonstrated performing specific tasks under controlled conditions. Its long-term reliability in continuous production use is unproven. Maintenance and support costs are uncertain. And the technology is moving fast enough that a unit purchased today may look dated in three years.
No experienced procurement manager at a large manufacturer is writing a capital purchase order for a $150,000 machine on those terms. The commercial question facing humanoid robotics companies right now — how do you actually structure a deal that a risk-averse operator will agree to — is receiving far less coverage than the technology itself. That is worth examining, because the pricing models that emerge from early commercial negotiations will shape both how quickly the industry scales and which companies survive long enough to find out.
The Three Broad Models
Humanoid robot companies are, in practice, offering or discussing three types of commercial arrangements. They are not mutually exclusive, and most companies are experimenting with more than one depending on the customer and context.
The first is direct sale: the operator buys the unit outright, owns the hardware, and takes on maintenance responsibility either directly or through a separate service contract. This is how the vast majority of industrial equipment has historically been sold, and it is the model that generates the cleanest revenue recognition for the manufacturer. It is also the hardest sell right now, because it requires the operator to bear the full capital risk on unproven hardware.
The second is leasing, which spreads the capital cost over time and typically includes a service and maintenance component bundled into the monthly payment. Leasing is common in industrial equipment — machine tools, forklifts, imaging equipment in healthcare — and it reduces the upfront commitment required from operators. The manufacturer or a financing partner retains ownership of the hardware, which also means they can reclaim and redeploy it if the operator relationship ends.
The third, and the one generating the most discussion in the industry, is robot-as-a-service, often abbreviated to RaaS. Under this model, the operator pays a recurring fee — typically monthly or per hour of operation — and the manufacturer retains ownership of the hardware, handles maintenance, and in some formulations guarantees a level of performance or uptime. The operator gets access to robot labour without a capital commitment; the manufacturer builds a recurring revenue stream instead of a one-time sale.
Why RaaS Is Attractive in Theory
The robot-as-a-service model has real appeal for both sides of the transaction, at least in principle. For operators, it converts a large, uncertain capital expenditure into an operating expense that can be compared directly against the labour cost it is meant to replace. If a robot costs $8 per hour under a RaaS contract and the human labour doing the same task costs $22 per hour including overhead, the economic case is legible and the commitment is limited to the contract term. If the robot underperforms or the task changes, the operator is not stuck with a depreciating asset.
For manufacturers, RaaS creates the kind of recurring revenue base that software-as-a-service businesses have used to build durable companies. A customer paying $5,000 per month per unit for a five-year contract is worth $300,000 — double what the hardware might sell for outright, and with a stronger ongoing relationship that generates data, field experience, and switching costs. The manufacturers pitching RaaS to investors are drawing explicit analogies to SaaS: high upfront cost to deploy, low marginal cost to maintain, compounding returns as the installed base grows.
The analogy has limits. Software does not fall down, wear out, or require a field technician when it breaks. The marginal cost of serving an additional software customer is genuinely close to zero. The marginal cost of supporting an additional deployed robot unit involves hardware maintenance, spare parts, remote monitoring infrastructure, and in some cases on-site support. Whether the economics of a robot fleet can actually resemble SaaS economics is an open question that current deployments are too small to answer.
The Uptime Guarantee Problem
The credibility of any RaaS proposition depends heavily on what the manufacturer can actually promise about performance. A RaaS contract that charges by the hour of operation has a natural performance incentive built in — the manufacturer only gets paid when the robot is working. But a contract with a fixed monthly fee regardless of performance gives the operator no recourse when the robot is down for maintenance or fails to complete tasks reliably.
Most industrial equipment leases include uptime guarantees or service-level agreements: if the equipment is unavailable for more than a specified number of hours per month, the operator gets a credit or the maintenance provider incurs a penalty. Applying this structure to humanoid robots requires committing to uptime numbers that current operational data does not cleanly support.
What manufacturers know about their own robots' uptime comes primarily from controlled environments and short-duration pilots. Extrapolating from a three-month pilot in a bounded warehouse zone to a multi-year production commitment with contractual uptime obligations is a significant step. The companies trying to write those contracts are navigating a genuine uncertainty: their robots' reliability in sustained production use is genuinely unknown, not just commercially undisclosed.
Some manufacturers are addressing this by structuring early contracts with lower performance guarantees and explicit provisions for improvement over time — in effect, building the learning period into the contract terms. This is commercially honest, but it also limits the economic case the operator can make for adoption: a robot whose performance is guaranteed to improve over two years is not a straightforward substitute for a human worker today.
How Pricing Is Actually Being Set
Published pricing for commercial humanoid deployments is rare. Most deals are negotiated bilaterally and treated as commercially sensitive. What has been disclosed is fragmentary, but enough to sketch an outline.
Agility Robotics, in conversations with potential customers, has described pricing for Digit in the range of $10 to $15 per hour for RaaS arrangements, depending on volume and contract length. At $10 per hour, a robot working a standard 40-hour week costs roughly $20,000 per year in direct robot costs — well below typical US warehouse labour costs when benefits and overhead are included. At full utilisation across three shifts, the economics improve further for the operator. But full utilisation across three shifts assumes maintenance windows can be managed without disrupting production, task reliability is high enough to operate without close human supervision, and the task scope justifies that level of deployment — none of which is currently straightforward.
Figure AI, targeting automotive manufacturing with its Figure 02 system, has not published pricing but has described its commercial model in terms consistent with a premium RaaS approach targeting high-value, high-consistency industrial tasks. BMW-scale manufacturing contracts would presumably involve different terms than a logistics pilot.
Unitree, whose G1 platform is priced at $16,000 for the hardware, represents a different end of the market: a robot cheap enough that direct sale is commercially plausible, but one aimed primarily at research, development, and light-duty applications rather than production deployment. The Unitree pricing model reflects a different bet on how the market will develop — volume and accessibility over premium positioning.
The Service Infrastructure Problem
Any recurring-revenue model for hardware requires a service and maintenance infrastructure capable of keeping deployed units operational. For a company with fifty robots deployed across five facilities, this is manageable. For a company with five thousand robots deployed across hundreds of facilities in multiple countries, the service infrastructure required is substantial — field technicians, spare parts logistics, remote diagnostics capability, and the organisational processes to coordinate all of it.
No humanoid robot company has yet built service infrastructure at anything approaching that scale, because no company has deployed at anything approaching that scale. The companies projecting rapid growth in their commercial models are implicitly projecting simultaneous rapid growth in service and support capability. How that gets financed — whether through the recurring revenue itself, through additional fundraising, or through partnerships with industrial service companies that already have field infrastructure — is a genuine open question.
Established industrial automation companies, including FANUC, Yaskawa, and Kuka, have spent decades building exactly the kind of global service networks that humanoid companies lack. Whether those companies become partners, acquirers, or competitors in the humanoid space will depend partly on how the technology develops and partly on which commercial model wins. A world where humanoid robots are sold as capital equipment and serviced by existing industrial automation networks looks different from a world where robot manufacturers own the entire service relationship through RaaS.
What the Pricing Model Signals
The commercial model a company chooses reveals something about its confidence in its own technology. A manufacturer willing to offer performance-based pricing — where payment is contingent on the robot actually completing tasks — is making a stronger claim about its reliability than one offering straightforward leases with no performance guarantees. The former involves the manufacturer putting its own economics at risk alongside the operator's; the latter transfers uncertainty to the buyer.
The inverse is also worth noting: companies pitching heavily on the RaaS model, with recurring revenue projections and SaaS-style unit economics in investor materials, are making claims about long-term margin structures that current deployments cannot yet validate. Investor narratives and customer conversations involve different versions of the same pitch, calibrated to different audiences. Reading the customer-facing commercial terms carefully — when they are available — provides a more honest picture of where the technology actually is than the investor-facing growth projections.
The pricing models that emerge from the next two or three years of commercial negotiation will be one of the more reliable indicators of which companies are building something durable and which are still primarily operating on the promise of future capability. How a company gets paid, and under what conditions, turns out to be a fairly direct measure of how much confidence it has in what it is selling.