A humanoid robot company releases a demonstration video. The robot folds laundry, carries a box, climbs stairs. The coverage arrives within hours. Investor interest follows. The company's valuation climbs. What the coverage almost never addresses is the question that determines whether any of this translates into a viable business: how does an actual company, with a real operations problem to solve, decide to purchase one of these things?
The procurement process — the evaluations, the financial approvals, the legal reviews, the operational pilots — is the unglamorous mechanism that connects impressive hardware to commercial deployment. And right now, for humanoid robots, that mechanism is largely unbuilt. Not because buyers aren't interested, but because the process of evaluating and committing to technology this new, this expensive, and this operationally consequential is genuinely difficult, and neither buyers nor sellers have fully figured out how to do it.
Who Is Actually Making These Decisions
The first thing worth understanding is that the decision to deploy humanoid robots at scale doesn't belong to a single person or department. In any organisation large enough to be a meaningful customer for humanoid robotics — a major logistics company, an automotive manufacturer, a large retailer — a procurement decision of this magnitude involves multiple stakeholders with different incentive structures, different risk tolerances, and often different views on whether the technology is ready.
Operations leadership cares primarily about throughput and reliability. Will this robot do the job consistently enough to justify the disruption of introducing it? If it fails in production, what happens to the workflow it was supposed to support? The operations team has seen enough automation projects underdeliver that scepticism is the default posture, and they are usually the ones who have to manage the consequences when something goes wrong.
Finance wants a credible return-on-investment calculation. With humanoid robots, this is harder than it sounds. The capital cost is significant — units from the current generation of commercial humanoid platforms run from roughly $70,000 to well over $100,000 each, depending on capability and contract structure. Maintenance costs are uncertain because the operational track record is thin. The labour savings the robot is supposed to generate depend on assumptions about uptime, task success rates, and supervision requirements that real-world deployments haven't yet fully validated. Constructing a financial case for a technology this new requires making assumptions in almost every input, which makes the resulting numbers less persuasive to anyone who understands how they were built.
Legal and risk management teams have their own set of concerns. Who is liable if a humanoid robot injures a worker? What does the service contract actually guarantee, and what happens when the robot fails to meet those guarantees? Does deploying this technology create obligations under occupational health and safety regulations that haven't yet adapted to humanoid robots specifically? These are not hypothetical objections. They are real questions that procurement teams are raising and that the robot companies do not always have clean answers to.
The Pilot Trap
The standard response to procurement uncertainty in enterprise technology is the pilot: a time-limited, scope-limited trial that lets the buyer evaluate the technology before committing to a full rollout. Pilots are sensible risk management. They are also, in the context of humanoid robotics, frequently inconclusive — and both buyers and sellers are beginning to understand why.
The problem is that a meaningful pilot for a humanoid robot requires conditions that are difficult to establish quickly. The robot needs to be deployed in an environment representative of actual operating conditions, not a specially prepared test cell. The task it's performing needs to be one that matters operationally, not a demonstration task chosen because the robot handles it well. The evaluation period needs to be long enough to capture how the robot performs across variations in workload, environment, and edge cases. And the metrics used to evaluate success need to be defined in advance and tied to real business outcomes, not to impressive-looking demonstrations.
Most early pilots for humanoid robots have not met these conditions. The timelines are short, the environments are controlled, the tasks are selected to showcase the robot's strengths, and the success metrics are often vague. The result is pilots that generate enthusiasm and positive press but don't actually answer the questions that procurement teams need answered: Can this robot run reliably for six months at our facility, doing our most common task, without requiring more human oversight than the task we're trying to automate?
Several large manufacturers have described a cycle of repeated pilots — each one generating interest, each one falling short of the evidence threshold needed for a volume commitment — that is exhausting both the patience of internal champions and the commercial timelines of the robot companies they're evaluating. The pilot is necessary but not sufficient, and the gap between what a pilot produces and what a full procurement decision requires is wider than either side typically acknowledges at the outset.
What the Successful Early Deals Actually Look Like
The humanoid robot partnerships that have progressed furthest — Agility Robotics and Amazon, Figure AI and BMW, Apptronik and NASA — share a structural feature that distinguishes them from standard procurement relationships. In each case, the customer was involved in the development process, not just the evaluation process. Amazon invested directly in Agility. BMW partnered with Figure AI before the robot was production-ready. NASA worked with Apptronik on the hardware specification itself.
This is not conventional procurement. It is closer to a joint development relationship, where the customer contributes operational knowledge, facility access, and in some cases capital, in exchange for early access to a technology they believe will matter and some influence over how it develops. The customer takes on more risk than in a standard procurement — the technology might not work, the timeline might slip, the company might not survive — but in exchange gets a relationship with the supplier that gives them advantages a late-adopting competitor won't have.
For companies that don't have the scale or risk appetite for this kind of relationship, the commercial entry point is much less clear. The large platform deals absorb most of the early production capacity and most of the robot company's implementation attention. A mid-sized manufacturer that wants to evaluate humanoid robots for a specific task in a specific facility is not the customer these companies are prioritising right now, even if it represents a significant addressable market over time.
The Comparison Problem
Procurement decisions for capital equipment usually involve comparison shopping. You evaluate multiple vendors, compare specifications and pricing, check references, and make a decision based on competitive analysis. For humanoid robots, this process is genuinely difficult because the products are not interchangeable and the market is not mature enough to have established comparison frameworks.
The relevant comparison for humanoid robots is not always other humanoid robots. For a given task in a given environment, the real comparison might be a conventional industrial robot arm, an autonomous mobile robot, an exoskeleton that augments a human worker, or additional human labour. Each option has different capital costs, different operating costs, different flexibility, and different risk profiles. Constructing a fair comparison requires the kind of detailed operational data that most potential customers haven't had the time or occasion to collect.
The humanoid robot companies have an interest in framing the comparison in ways that favour their product. Comparisons that emphasise flexibility — the humanoid's ability to perform multiple tasks in human-designed spaces — tend to favour humanoids over fixed automation. Comparisons that emphasise reliability and cost-per-operation tend to favour conventional robots for any task that conventional robots can actually perform. Both framings are accurate in their own terms. Neither is the complete picture.
A few independent organisations have begun developing evaluation frameworks that try to establish consistent comparison criteria — the International Federation of Robotics has working groups on this, and some academic institutions have published early methodology papers. But the field is too new for those frameworks to have been validated against real-world deployments at scale, and the companies with the most operational data have competitive reasons not to share it.
The Contract Structure Question
One thing that has become clearer as humanoid robots approach commercial deployment is that the traditional capital equipment purchase model — you buy the machine, you own the machine, you maintain the machine — does not map cleanly onto how robot companies want to go to market, or onto what risk-averse buyers are willing to accept.
Most humanoid robot companies are gravitating toward service-model contracts: the customer pays a monthly or annual fee that covers the hardware, software updates, maintenance, and support. This structure is attractive to robot companies because it creates recurring revenue and keeps them close to the deployment, able to push software updates and respond to issues quickly. It's attractive to buyers because it converts a large uncertain capital expense into a more predictable operating expense, and it puts the performance risk back on the vendor — if the robot doesn't meet agreed uptime or task success targets, the contract terms create remedies.
The challenge is that service contract terms for technology this new require negotiating around uncertainties that neither party can fully characterise. What constitutes acceptable uptime? What task success rate triggers a remedy? Who is responsible for performance degradation caused by changes the customer makes to the operating environment? These questions don't have established industry answers, which means every contract negotiation is partly a negotiation over how to handle a future that neither party can predict with confidence.
Where the Buying Decision Actually Stands
The gap between "companies are interested in humanoid robots" and "companies are committing capital to humanoid robots at scale" is real and currently wide. It exists not because of a single obstacle but because of a cluster of procurement challenges that are each individually manageable but collectively slow the process significantly: financial models built on uncertain assumptions, legal questions without established answers, comparison frameworks that don't yet exist, and contract structures being invented in real time.
None of this means that widespread commercial deployment won't happen. It means the path from here to there runs through a procurement process that is less glamorous than the demo videos and harder to accelerate than the engineering. The companies that figure out how to make the buying decision tractable for cautious operations teams and conservative finance departments — not just technically impressive for early adopters — will have a meaningful advantage over those that don't.