When a factory manager asks a humanoid robot company how to get a few units into their facility, the answer is rarely "here's the price list." Most companies in the humanoid space are structured around a service model — sometimes called robot-as-a-service, sometimes framed as a subscription, sometimes described simply as "deployment partnerships." The language varies. The underlying logic is consistent: the manufacturer retains ownership of the hardware, the customer pays a recurring fee for access, and the manufacturer controls the software, updates, and operational parameters.
That structure is not accidental, and it's not simply a financing convenience. It reflects a set of decisions about where value accumulates in this industry, who bears risk during an uncertain period of technical development, and what kind of business these companies are actually trying to build. Understanding the model tells you something useful about how humanoid robotics will enter the economy — and what that means for the companies on both sides of the transaction.
Why the Service Model Exists
The straightforward explanation is that the hardware is expensive and the technology is changing quickly. A humanoid robot that costs $100,000 to $200,000 per unit today may cost substantially less in two to three years as production scales and manufacturing efficiency improves. A customer who buys at today's prices is exposed to rapid depreciation. A customer who pays a monthly fee sidesteps that exposure — though they also forgo any upside if the technology proves more durable than expected.
From the manufacturer's side, the service model solves a different problem. Selling a robot outright transfers the hardware liability to the buyer, but it also transfers control. The manufacturer loses the ability to remotely update the system, collect operational data from the robot's activities, or ensure that the robot is being used in ways consistent with its capabilities and safety certifications. In a period when the technology is still being refined and edge cases are still being discovered in real deployments, that loss of control is consequential.
Several humanoid companies have been explicit about this in their public communications. The argument runs roughly as follows: these robots run on software that improves continuously; deploying robots under service agreements allows manufacturers to push updates across their entire fleet simultaneously; customers benefit from improvements without procurement cycles; and the manufacturer maintains the relationship and visibility needed to ensure safe, effective operation. All of that is true. It is also true that the arrangement concentrates a significant amount of ongoing leverage with the manufacturer in ways that customers should think carefully about.
The Data Flywheel Underneath
The service model has a second rationale that gets less attention in the commercial pitch but is arguably more important to the long-term business: operational data.
A humanoid robot operating in a real facility generates continuous data — sensor readings, task outcomes, failure events, the specific conditions under which the robot succeeded or struggled. That data is the raw material for improving the robot's capabilities. A manufacturer with a fleet of robots deployed under service agreements across dozens of facilities has access to operational data at a scale that no laboratory or simulation environment can replicate. A manufacturer who has sold robots outright to customers may have much weaker data rights, depending on how the sale agreement was structured.
The competitive implications are significant. In an industry where the software stack and learned capabilities are the primary source of differentiation — where two robots with similar hardware can perform very differently depending on the quality of their training — the company that accumulates the most diverse, high-quality operational data has a structural advantage. The service model is one mechanism for ensuring that advantage accrues to the manufacturer rather than dispersing to customers who might share data with competitors or simply not collect it systematically.
This dynamic has a precedent in industrial software: enterprise resource planning systems, manufacturing execution software, and logistics platforms all moved toward subscription models partly for revenue predictability reasons and partly because the vendor's ability to improve the product depends on usage data that only active deployments generate. Humanoid robotics is following a similar logic, accelerated by the fact that the "software" in this case includes learned physical capabilities that improve directly from operational experience.
What Customers Are Actually Agreeing To
The service model shifts risk in both directions, and the specific terms of any given agreement determine which way the balance tips.
On the customer side, the recurring fee structure converts a large capital expenditure into an operating expense, which has accounting and cash flow advantages. The manufacturer's responsibility for maintenance, updates, and performance benchmarks (where those are specified in the contract) means the customer isn't carrying the full burden of keeping the technology functional. And if the robot proves unsuitable for the customer's specific environment, the downside of a service agreement is typically smaller than having bought hardware outright.
On the other side: the customer is dependent on the manufacturer for continued operation. If the manufacturer raises prices, changes the software in ways that alter the robot's behaviour, discontinues a product line, or goes out of business, the customer has limited recourse. The robot doesn't belong to them. Their operational workflows may have been designed around the robot's specific capabilities and interfaces. Switching to a competitor's system means retraining staff, potentially rebuilding integrations with other facility software, and accepting a period of reduced productivity during transition.
These lock-in dynamics are familiar from enterprise software, but they have a physical dimension in robotics that software-only products don't. A company that has restructured a warehouse operation around a specific humanoid robot platform has made capital investments in fixtures, processes, and training that are specific to that platform. The switching cost is not just contractual; it is physical and operational.
Contract terms in early humanoid deployments vary considerably and are not publicly available in detail. What industry sources describe is a pattern where manufacturers hold significant rights: to update software remotely, to collect and use operational data, to set operational parameters that constrain how and where the robot is used, and to terminate the agreement if the customer operates outside specified conditions. Customers with negotiating leverage — large enterprises with significant volume commitments — can push back on some of these terms. Smaller customers typically cannot.
The Ownership Question in Practice
There is a subset of humanoid deployments that have involved outright hardware sales or purchase options. Unitree's G1, priced at roughly $16,000, is sold as hardware. Some research institutions and industrial buyers have acquired humanoid robots through purchase agreements rather than service contracts. Boston Dynamics has sold Spot commercially since 2020.
But for the higher-capability humanoid systems designed for commercial deployment — the systems that Figure, Agility, Apptronik, and their competitors are positioning for logistics and manufacturing — the trend is clearly toward service and away from outright sale. The companies raising hundreds of millions of dollars in venture capital are doing so in part on the promise of recurring revenue from robot deployments, not one-time hardware sales. Investors understand the difference between a business that sells a product once and a business that generates monthly revenue from a deployed fleet. The service model produces the latter; outright sales produce the former.
The investor logic reinforces the operational logic: both point toward a structure in which humanoid robot manufacturers retain ownership and ongoing relationships with their hardware, rather than selling it into a market where they lose visibility and control after the transaction closes.
What This Means for the Companies Buying In
The practical implications for a business evaluating humanoid robot deployment are straightforward, even if they are rarely discussed clearly in the coverage of the industry.
First, the total cost of deployment over a multi-year horizon is the relevant number, not the monthly fee. Service agreements that look affordable on a per-unit, per-month basis can accumulate to a significant total commitment when extended over five or ten years across a fleet of dozens of units. That calculation needs to be made explicitly, not assumed to be favourable because the upfront cost is low.
Second, the data rights provisions in any service agreement deserve careful legal review. What data is the manufacturer collecting? What can they do with it? Does it include information about the customer's operations, processes, or facility layout that the customer would consider proprietary? Some early adopters have discovered that the default terms allow manufacturers to use operational data for training and improvement in ways that are broadly drawn and not limited to improving service for that specific customer.
Third, the dependency question — what happens if this manufacturer changes its terms, pivots its product, or fails — is not hypothetical. The humanoid robotics industry is at an early stage where consolidation is likely. Not every company raising large rounds today will exist in its current form in five years. A business that has built operational dependencies on a specific platform needs a contingency plan, even if the manufacturer currently seems well-capitalised and stable.
None of this argues against deploying humanoid robots or against the service model in principle. It argues for entering these agreements with clear-eyed understanding of what is being agreed to, rather than treating the structure as an afterthought to the technology decision. The robot-as-a-service framing is good marketing. It describes a real set of benefits. It also describes a set of commitments and dependencies that will matter considerably once a deployment moves from pilot to production scale.