Why won't you give me one total-cost-of-ownership number?

Because it would be fiction. Integration, supervision and support — the buckets that dominate — depend entirely on your task, your labor rates and your country. We give you the structure and multipliers so you can build a number you can defend, then quote the parts we can actually pin down.

Is the "replaces one worker" payback real?

Only after you subtract the supervision cost. During the POC and early production you are paying for a human teleoperator or supervisor for a meaningful share of the robot's hours, on top of maintenance. Run the payback with that line included; run two shifts and the math improves substantially, which is why humanoids pencil out first in high-labor-cost, multi-shift operations.

How much should I budget beyond the hardware in year one?

As a planning floor, expect a working deployed unit to cost roughly 2×+ its purchase price once operations are included, with integration and commissioning being the largest first-year lines. Use the cost-bucket table to allocate; don't treat the hardware quote as the budget.

Can I skip the POC and just buy a few units?

Not sensibly. The 3–6 month, 2–3 round POC exists because the robot has to be made to do your task, and the ~12 months of validation exists because uptime and economics have to be proven before you commit capital to 50 units. Skipping it converts a managed pilot into an expensive experiment.

What's the most-forgotten cost?

Human supervision and teleoperation through the pilot, followed by integration engineering. Both are labor, both are recurring or large, and both are invisible on a spec sheet — which is exactly why naïve TCO estimates come in low.

Does buying from China change the TCO structure?

It changes the risk structure more than the buckets. Hardware is competitively priced, but you must get spares pricing, warranty, in-country RMA and English SDK support onto the proforma invoice before payment — see our import playbook. The integration, supervision and downtime buckets are the same wherever the robot is made.

Total Cost of Ownership: Deploying a Humanoid Robot in a Factory

Who this guide is for

This is written for the buyer who has already decided whether to pilot a humanoid and now needs to budget what it actually costs to keep one working. If you are still comparing a $5,900 biped to a $16,000 one on spec sheets, read How Much Does a Humanoid Robot Cost in 2026? first — this guide starts where that one ends.

We will not give you a single all-in figure. Anyone who quotes you a precise total cost of ownership for a factory humanoid in 2026, sight-unseen of your process, is guessing or selling. What we can give you is the structure: the cost buckets, which ones dominate, and how they shift from year one to steady state.

Hardware is the anchor, not the answer

Start with the hardware, because it is the only number you can actually pin down today. From the catalog, single-unit EXW China:

Platform	Price band	Reference figure	Class
Unitree R1	$5–15k	¥39,900 · $5,900 US list	Entry / education biped
EngineAI PM01	$5–15k	$12,000 (business & education)	Compact research
Unitree G1	$15–50k	from ¥99,000 (≈$14k CN) · $16,000 US list · EDU to ≈$74k^†	Default research humanoid
Unitree H1	$50k+	≈$90,000 class historically^†	Full-size platform
Fourier GR-2	POA	quoted case-by-case	Advanced research / healthcare
AgiBot Yuanzheng A2	$50k+	reseller-reported ≈$100–190k by variant^†	Deployment-class

Two things to take from this table. First, the spread is enormous — a factory-capable humanoid is a $50k+ or POA conversation, not a $6,000 one; the sub-$15k machines are development and locomotion platforms, not workers. Second, and more important: for a deployment project the hardware is typically the smaller half of your first-year spend. Everything below is why.

The cost buckets nobody quotes you

Integration and engineering

This is usually the single largest first-year line and the one most often forgotten. A humanoid does not arrive knowing your process. You are paying — in your own engineers' time, the vendor's, or an integrator's — for fixtures and end-of-arm tooling, safety fencing or sensing, MES/PLC hookup, cell layout, and the perception and grasp work to make the robot do your task rather than the demo task. This scales with process complexity, not with hardware price. A single station doing one repetitive pick can be light; a multi-step assembly cell with tight tolerances is heavy.

If your task needs dexterous manipulation, the hands are both a hardware and an integration cost. A pair of capable hands — for example the Inspire RH56 class at $5–15k†, or higher-DOF units quoted POA — adds to the bill of materials and to the perception/grasp engineering. See our dexterous hands category for the range.

Human supervision and teleoperation

Here is the honesty that separates a real pilot budget from a fantasy one: most "autonomous" humanoid demonstrations in 2026 are teleoperated or scripted. During your POC and early production, plan to pay for a human — a teleoperator, a supervisor, or a data-collection operator — for a meaningful fraction of the robot's working hours. This is not a defect; it is how you collect the demonstration data that eventually reduces the supervision. But it is a real, recurring labor cost that undercuts the "replaces a worker" pitch until autonomy matures on your specific task. Budget it explicitly. If you don't, your pilot's economics will look far worse than expected halfway through.

Batteries, spares and consumables

Runtime per charge is a 1–4 hour class figure — the AgiBot A2 is reported at ≈2 h continuous†. Any multi-shift operation therefore needs battery swaps and a charging strategy, which means buying spare packs and a fast charger up front. Beyond batteries, the parts that wear are fingers/hands, actuators, cables and feet. Price the spares list at PO time, not after the first failure — a spare-parts quote you can't get before you buy is a warning sign about after-sales support. Individual joint modules run from under $1k each (see actuators & joint modules), but a full-hand replacement is a $5–15k-class event.

Maintenance and support

Budget an extended warranty / support contract in the range of 8–15% of hardware value per year, and insist the terms cover your country with a stated RMA process and who pays return freight. For a machine shipped from China, "who fixes it and how fast" is a cost and a risk you negotiate before payment — our import playbook covers getting this onto the proforma invoice.

Software, SDK and data

The SDK is usually bundled, but the ongoing costs hide here: edge compute upgrades for onboard vision/VLA models, any cloud or licensing fees, the data pipeline to collect and label demonstrations, and the engineering to retrain policies as your task drifts. Verify English SDK and documentation before payment. Also confirm where data goes — if the robot phones home, that is a compliance question in the EU (GDPR) and increasingly elsewhere; verify with your counsel.

Downtime and training

Downtime is a cost even when nothing breaks: commissioning, retraining after a process change, and scheduled maintenance windows all take the robot offline. At today's maturity, plan for supervised operation with maintenance windows, not 24/7 unattended running. Training is a one-time-heavy, recurring-light line: your maintenance staff and operators need real hours on the platform, and turnover resets part of that investment.

First-year vs steady-state: the honest table

The table below uses qualitative multipliers against your hardware cost, not invented totals — because your totals depend on your process, your labor rates and your country. Use it to sanity-check a vendor's proposal and to see where the money actually goes.

Cost bucket	First-year weight	Steady-state weight	Notes
Hardware (robot + hands)	1× (baseline)	Amortized / replacement only	The anchor; often the smaller half of year one
Integration & engineering	1–3× hardware	Low (change-driven)	Usually the largest year-one line; scales with process complexity
Human supervision / teleop	High, recurring	Declining as autonomy matures	The line that breaks naïve payback math if ignored
Batteries & spares	Moderate up-front	Recurring, moderate	Multi-shift needs swaps; price spares at PO
Maintenance & support	8–15%/yr of hardware	8–15%/yr of hardware	Insist on in-country RMA terms
Software / compute / data	Moderate	Low–moderate, recurring	Edge GPU, licensing, retraining pipeline
Downtime & training	High (commissioning)	Low–moderate	Retraining resets with process change / turnover

The shape to internalize: first-year cost is dominated by integration, supervision and commissioning — not hardware. Steady state is dominated by supervision (falling), maintenance and spares. A common industry rule of thumb puts a working, deployed unit at roughly 2×+ its purchase price once operations are included; treat that as a floor for planning, not a promise.

The timeline is a cost too

Time-to-value is the cost buyers underestimate most. The realistic path below reflects commonly-cited industry channel research on Chinese robot deployments†:

Phase	Duration	What happens
Proof of concept	3–6 months, 2–3 rounds	Robot proves it can do your task, under supervision, with iteration
Validation	~12 months	Reliability, uptime and economics validated on the floor
First order	after validation	Typically under 50 units per factory
Scale	later	Pilots grow toward 50–100 units per customer only after validation

From first serious meeting to stable deployment is on the order of ~2 years. Every one of those months carries engineering, supervision and opportunity cost. This is the single best filter for whether you should start now: if you need production output in six months, a humanoid pilot is not your path this year. If you are building a two-year capability, the clock starts at the POC.

How to decide if you're ready

You are a good candidate to start a pilot now if: you have a high-labor-cost market and multi-shift operations (the payback math works first here); you can staff a supervisor/teleoperator through the pilot; you have engineering capacity for integration; and you can carry a ~2-year timeline to stable deployment. Structuring the pilot as Robots-as-a-Service can shift integration and uptime risk to the vendor for 1–10 unit pilots — worse long-run unit economics, better risk profile early.

You should wait if: you need 95%+ unattended uptime in year one, you have no engineering resource for integration, or your business case collapses without immediate labor replacement. There is no shame in waiting — hardware cost is falling and autonomy is improving; the buyer who starts in 2027 with a mature task definition may spend less to reach the same place.

When you are ready to price a real project, send us the task, quantity and destination country via the RFQ form — a realistic scoped estimate needs your process, not a catalog.