The economics of autonomous fleet operations
The case for autonomous fleets is usually told as a story about removing the driver. The real story is more interesting: the savings from removing the driver are partially offset by new costs you only incur once you operate at scale. Whether an autonomous fleet earns a margin or burns cash depends on how the operator handles those new costs.
The cost stack, line by line
A serviceable cost-per-mile model for a robotaxi includes ten line items. They are not all equal, and the order of magnitude matters more than the exact figure:
- Vehicle depreciation — typically the largest fixed cost, in the range of $0.15–0.25 per mile for a Tesla-class EV over a 300,000-mile useful life.
- Energy — $0.04–0.07 per mile at fleet-managed charging rates, 3–4× cheaper than gasoline equivalent.
- Insurance — currently $0.20–0.35 per mile for AV operations; expected to fall significantly as actuarial data accumulates.
- Hub operations — charging plaza, cleaning, calibration, diagnostics, real estate amortisation: $0.08–0.15 per mile depending on hub density and fleet utilisation.
- Maintenance — predictive replacement of tires, brakes, sensors, software: $0.04–0.07 per mile, comparable to a traditional taxi.
- Connectivity and compute — onboard AV compute and cellular: $0.02–0.04 per mile.
- Software platform — fleet operations OS, dispatch, energy: $0.01–0.03 per mile at scale.
- Payment processing and customer support — $0.03–0.05 per mile, similar to rideshare.
- Teleoperation and intervention — required for the foreseeable future, scales with miles: $0.02–0.06 per mile depending on the AV stack maturity.
- Regulatory and permitting — a fixed-fee allocation per mile depending on jurisdiction.
A well-run fleet lands in the $0.65–$0.90 per mile range fully loaded, against a human rideshare cost stack of $1.50–$2.00. The gap is real but it is not the order of magnitude that early AV pitches promised.
Where the operator actually wins or loses
Three of those ten line items move more than the rest combined when you adjust the operator's execution:
- Utilisation. Every percentage point of utilisation directly amortises the fixed cost stack across more revenue miles. A fleet at 60% utilisation has a 30% lower cost per mile than the same fleet at 45%. This is mostly a dispatch software problem.
- Charger handoff. The time between a vehicle ending one trip and starting its next revenue-eligible state is the largest avoidable cost in the entire model. A 90-second improvement in charger handoff time across a 500-vehicle fleet pays for the entire energy management system.
- Insurance loss ratio. Operators that can prove a sub-1.0 loss ratio to underwriters earn double-digit annual reductions in their premium. This is an asset-health and intervention-rate problem, not a "drive carefully" problem.
Why this is not a "scale solves it" story
The most expensive assumption operators make is that the cost stack flattens out at scale. It does not. Insurance, real estate, and energy all have step-functions in them — once a fleet crosses a certain density in a market, those costs change discontinuously. The operators that win plan for the step-functions rather than amortising past them.
The role of an infrastructure partner
Autzu owns the four most operationally-dense line items in this model — hub operations, energy, maintenance, and software platform. We sell access to that infrastructure on a per-vehicle basis, which lets AV operators redeploy their balance sheet into the autonomy stack itself. The result for the operator is a cost-per-mile model that is predictable from day one rather than emerging slowly through a multi-year operational learning curve.
References & further reading
- Uber Technologies — Q4 2025 Form 10-K filingssec.gov
- RethinkX — Disruption, Implications, and Choicesrethinkx.com
- McKinsey — The autonomous-vehicle racemckinsey.com
- Argo AI / Ford — AV cost-of-operation white paperargo.ai
