A California recycler pays nothing for the robot on its sorting line. The company that built it takes half the value of whatever that robot recovers instead. That is one of three ways we found to get a robot onto a utility site without buying it, and the choice between them decides who carries the risk.
The test is ownership: does the utility end up with the robot on its books? Below are notable cases where it does not. Nearly everything else on our map of this market is a purchase, a pilot or a project campaign.
The five to know
Ordered by commercial model, strictest first.
| Company | Commercial model | Where it sits on our map | The agreement | Funding stage |
|---|---|---|---|---|
| ANYbotics | Robotics as a Service | Facility inspection across waste-to-energy, transmission, distribution and generation | Kanadevia Inova: ANYmal robots deployed into plant operations and maintenance, explicitly as a Robotics-as-a-Service offering. Gresco Utility Supply: RaaS subscriptions carried into a US utility supply channel | Series C |
| Built Robotics | Robotics as a Service | Grid construction | Sunstate Equipment: autonomous machinery made available to contractors as rentals across the US rather than sold | Series C |
| EverestLabs | Robotics as a Service | Waste sorting | Caglia Environmental: a two-year equipment lease with a 50% share of the value of the material recovered | Series A |
| Gecko Robotics | Robotic service | Thermal plant and gas facility inspection | NAES: a multi-year US power plant agreement reported at over $100M rising to $250M contingent on results. ADNOC Gas with Al Masaood Energy: multi-year predictive maintenance with an estimated $30M ceiling | Series D |
| AMP | Outsourced concession | Waste sorting | Southeastern Public Service Authority: twenty years, 540,000 tonnes a year for eight Virginia communities at $50 a ton, against roughly $200M of AMP’s own capital | Series D |
Three models hide behind one label
Only the first three are Robotics as a Service in the strict sense. The provider keeps ownership of the machine and the client runs it on its own site, through a subscription, a rental or a lease. That costs less per unit of work once volumes are steady, and it builds internal capability rather than outsourcing it, but it needs someone on staff who can operate the thing. Worth noting that only the ANYbotics agreement with Kanadevia Inova is described as Robotics as a Service on its face. The other two arrive at the same place through a rental channel and an equipment lease.
Gecko Robotics is doing something different, and our research treats it as a separate model. Gecko owns the robots and operates them itself, but the work happens on the customer’s asset. The utility is buying an inspection, not access to a machine, and pays on scope and results. For a utility with a thin technical team that is the easier thing to buy, which is part of why it is also the largest agreement in the set.
AMP’s Virginia contract sits further out again. AMP financed, owns and operates the plant, and the authority sends material and pays per tonne. Structurally that is a build-own-operate concession, the same shape as a municipal waste contract or a power purchase agreement, and the robotics inside it are incidental to the commercial form. It belongs here because the authority avoided the capital entirely, which is the outcome utilities want from RaaS, but calling it Robotics as a Service would stretch the term past usefulness. AMP’s other agreements point the other way: Waste Connections has booked or deployed fifty AMP systems and Evergreen fifteen, and those are equipment sales.
Funding stage lines up with all of this. The two companies carrying the capital themselves are the best funded, both at Series D, because carrying capital is the model. EverestLabs at Series A cannot do that, so it shifted the risk instead and takes half the value of what its robot recovers rather than a fee.
What looks like a service and is not
The trap when screening this market is the long service tail. We hold solar panel cleaning contracts carrying operations and maintenance commitments of fifteen and twenty-five years, attached to hardware the customer bought outright for between $475,000 and $8.5M. A twenty-five year maintenance obligation reads like a service relationship and is priced like one, but the asset sits on the buyer’s balance sheet and the depreciation and obsolescence risk sits there with it. In a field where the models behind the robot are revised annually, that is the risk that matters most, and it is why contract length is a poor test and ownership is a good one.
The buyers who have done this more than once
Whatever the structure, repeat buying is the signal worth tracking, because a second agreement means the first one survived contact with an operational network. A handful of utilities and majors account for a disproportionate share of the contracts in this market.
| Buyer | What the record shows |
|---|---|
| Veolia | Three separate agreements, three different suppliers, three continents |
| SUEZ | Two agreements with two suppliers, both in water networks: Solinas and Acwa Robotics |
| Chubu Electric Power Grid | Three separate engagements with the same supplier, across 2021, 2025 and 2026 |
| Enel and Enel Green Power | More than one agreement, including blade inspection data across an entire wind fleet with Perceptual Robotics |
| EDP Renewables | Four consecutive annual inspection campaigns on the same floating solar asset with BeeX |
| Adani Green, Shell and ExxonMobil | More than one agreement each |
Three questions worth asking a provider
All of the above reduces to three questions, and they are worth asking early, because the answers are hard to change once a contract is drafted.
- Who owns the robot at the end of the term? If the answer is the utility, this is a purchase with a payment plan, whatever the agreement is called.
- Who operates it day to day? That decides whether the utility builds a capability or rents one, and it is the whole difference between the ANYbotics model and the Gecko model.
- What happens when the software is a generation behind? On an owned asset that risk belongs to the utility. On a subscription it belongs to the provider, and in this field the models are revised faster than the hardware.
