When Unitree announced that its G1 humanoid robot would be available for $16,000, the coverage was predictably excited. The number is striking — previous commercial humanoid platforms have been priced at ten to twenty times that figure, and $16,000 is roughly what a mid-range car costs. The implication, repeated across dozens of articles, was that the economics of humanoid robotics had fundamentally shifted.

What those articles almost universally omitted was everything that happens after you buy the robot.

Purchase price is one line in the total cost of operating a humanoid system. Depending on the use case, it may not even be the largest one. Understanding why — and what the full cost picture actually looks like — matters more than any single headline number, and it's the question that will ultimately determine whether humanoid robots move from demonstration-stage technology to genuinely widespread deployment.

What "Price" Actually Means for These Systems

Humanoid robot pricing is not standardised, and comparing figures across companies requires care. The $16,000 Unitree G1 and the $200,000-plus platforms from more established vendors are not competing on the same terms. They differ in capability, in what's included in the purchase price, and in the support and integration infrastructure that surrounds them.

At the lower end of the market, a purchase price typically buys the hardware and some basic software. Integration — writing the task-specific code that makes the robot actually useful in a given environment — is the buyer's problem. Maintenance is handled through support contracts priced separately. Updates to the underlying software may or may not be included, and the support infrastructure for an operator encountering a novel failure mode is often thin.

At the higher end, purchase price often includes more: integration support, operator training, service-level agreements on uptime, and ongoing software development from the manufacturer. The robot is more capable, but it also arrives with a surrounding layer of services that are priced into the headline figure or sold alongside it. Comparing the sticker price of a Unitree G1 to a Figure AI platform without accounting for that surrounding infrastructure produces a misleading comparison.

This isn't a criticism of any particular vendor. It reflects the fact that the industry is at an early stage where product definitions are still being worked out. What "buying a humanoid robot" means commercially is not yet settled — and buyers who treat purchase price as the primary economic variable will get surprises.

The Uptime Problem

A humanoid robot that works reliably ninety percent of the time sounds impressive. In practice, for continuous production work, it is barely adequate — and current systems are not consistently reaching that figure in real deployments.

Consider the comparison with human labour. A warehouse worker on a standard shift works a defined number of hours, takes scheduled breaks, calls in sick occasionally, and requires time off. Their effective availability over a working year, accounting for all of that, is roughly eighty to eighty-five percent of scheduled hours. They do not require mechanical maintenance. When they encounter an unexpected situation — a box that fell at an odd angle, a colleague asking an unexpected question — they handle it without escalation. And if they are unable to complete a task, the consequences are usually modest and self-contained.

A humanoid robot operating at ninety percent uptime sounds comparable on paper, but the nature of the downtime is different. Mechanical failures require technician visits, spare parts, and potentially manufacturer involvement. Software failures may require remote diagnosis or updates that take the robot offline for extended periods. Edge cases that the robot cannot handle autonomously require human intervention — and if the robot is deployed in a context where intervention is not readily available, those cases cause the robot to stop entirely rather than improvise.

Current commercial humanoid systems are not operating at ninety percent uptime in uncontrolled production conditions. The figures that exist come from supervised pilot environments, and they are not routinely disclosed publicly. Companies running pilots typically describe uptime in general terms — "performing reliably," "meeting operational targets" — without publishing specific numbers. That opacity is informative in itself.

What credible industry observers suggest, based on limited data from early deployments, is that uptime in the first year of commercial operation for a humanoid platform in a real environment typically runs in the range of sixty to seventy-five percent. That figure improves as the operator gains experience with the system and as the manufacturer refines the software for that specific deployment context. But reaching steady-state reliability takes time, and the cost of the learning period is real and largely invisible in purchase-price comparisons.

The Integration Cost

The work required to make a general-purpose humanoid robot useful for a specific task in a specific environment is substantial, and it is almost always underestimated by first-time buyers.

Integration involves more than writing task software. It includes mapping the physical environment in enough detail for the robot to navigate reliably; defining the task precisely enough that the robot can execute it without human intervention for the large majority of cases; identifying and handling the edge cases that will inevitably arise; training operators on how to supervise and intervene; and establishing the monitoring infrastructure to detect failures before they cascade into larger problems.

For well-defined, high-repetition tasks in structured environments — the kind of deployment that most current humanoid pilots involve — integration typically takes several months of engineering effort. One rough figure that circulates among systems integrators is that for every dollar spent on hardware, operators should budget two to three dollars for integration, depending on task complexity and environmental variability. That is not a published standard; it is a working estimate from practitioners, and it varies widely. But it is consistently higher than first-time buyers expect.

The integration cost also has a less obvious component: the operational disruption of deploying a robot into an existing workflow. Introducing a humanoid into a warehouse or factory is not a plug-and-play process. It requires redesigning physical layouts, retraining human workers on how to operate alongside the robot, and often modifying upstream and downstream processes to accommodate the robot's constraints. Those changes absorb management time and create friction during the transition period — friction that does not appear in any hardware cost estimate.

Ongoing Costs: Maintenance, Software, and Support

Humanoid robots have many moving parts — joints, actuators (the motors and mechanisms that drive movement), sensors, gearboxes — and those parts wear. The maintenance profile of a bipedal humanoid in a real production environment is not yet well-characterised, because no humanoid platform has been running in large-scale commercial deployment long enough to generate the long-run failure data that manufacturers of conventional industrial robots have accumulated over decades.

What is known is that the mechanical components most subject to wear are the joints, particularly in the legs and hands, which absorb repeated stress over thousands of operating hours. Early indications from pilot deployments suggest joint maintenance is the most frequent service requirement, though intervals vary considerably by use case. A robot navigating smooth warehouse floors has a different wear profile than one operating on uneven terrain or climbing stairs repeatedly.

Software and AI model updates are a cost category that receives less attention than mechanical maintenance but is arguably more significant in the current period. Humanoid robots are learning systems, and the software governing their behaviour is updated regularly. Some updates are improvements; others introduce new failure modes in tasks the robot previously handled well. Managing software updates in a production environment — validating each update against existing task performance, rolling back when something breaks, retraining operators on changed behaviours — is a genuine operational overhead that does not appear in any hardware cost estimate.

Support contracts vary widely. Some manufacturers offer comprehensive agreements that include on-site technician visits within defined response windows; others operate on a best-effort basis. For operators running a small number of units, a manufacturer's support contract may be the only realistic option when something goes wrong. That means the quality of a manufacturer's support infrastructure is a meaningful factor in total cost of operation — and one that is difficult to evaluate before you've needed it.

The Labour Comparison — Done Properly

The case for deploying humanoid robots commercially rests on their eventually being cheaper than human labour for a defined set of tasks. That comparison is frequently made in ways that are too simple to be useful.

A rigorous comparison needs to account for: the fully-loaded cost of human labour in the relevant role and geography (wages, benefits, payroll taxes, management overhead, training, and turnover costs); the total cost of robot operation over the same period (purchase price amortised over expected useful life, integration, maintenance, software, support, energy, and the cost of human supervision required during robot operation); the difference in output quality, reliability, and flexibility between the robot and the human worker; and the value of adaptability — a human worker can be redeployed to a different task with modest retraining; a robot configured for one task cannot easily be reconfigured for a different one without significant additional integration work.

When the comparison is done with that level of care, the economics of current-generation humanoid robots are most favourable in a narrow set of conditions: high-wage labour markets, extremely repetitive tasks with low variability, continuous operating requirements that are difficult to staff with human labour, and a long enough deployment horizon to amortise the upfront costs. Those conditions describe some real environments — specific logistics operations, certain manufacturing settings with severe labour shortages, particular applications in agriculture. They do not describe most workplaces.

The economic case for general-purpose humanoid deployment at scale requires either a meaningful reduction in total cost of operation, a meaningful improvement in robot capability that reduces the integration overhead, or further increases in human labour costs. All three of those things are happening — but at different rates and with different degrees of certainty.

Where the Costs Are Going

Hardware costs are coming down. That is genuinely true, driven by increasing production volumes, improving manufacturing processes, and the supply chain maturation that happens when an industry moves from artisanal to industrial production. The Unitree G1's $16,000 price point would have been implausible five years ago, and credible forecasts suggest capable humanoid hardware will reach $10,000 or below within the next few years at meaningful production volumes.

But hardware cost is not the binding constraint for most operators today. Integration, maintenance, and the human overhead of supervising and managing robot operations are larger fractions of total cost than hardware for most commercial deployments. Those costs will come down more slowly, because they depend on improvements in robot capability and reliability that are harder to accelerate than manufacturing efficiency.

The cost trajectory that will matter most for widespread deployment is not the hardware price curve. It is the integration effort required per new deployment, and the uptime and task success rates that determine how much human supervision each robot requires. As those figures improve, the economics will shift. They are improving — the pilots running today are generating data that will make the next generation of deployments faster and cheaper to stand up. But the improvement is uneven, and it is not on the timeline that hardware headline prices suggest.

The $16,000 robot is a real data point. The full cost of deploying it usefully is a different number — and for now, that number is much harder to know, which is precisely why it deserves more attention than it gets.