Robotaxi Cost Per Mile 2026: $1.20 Cruise Corridor Kill or Keep

A specific per paid mile rate decides whether a robotaxi corridor lives or dies. That single metric reframes pilots from autonomy showcases to portfolio bets, where paid-mile economics outweigh disengagement bragging rights. When cost holds at that level, operators have a reason to scale. When it slips above, discipline demands a kill decision rather than another extension.

Corridor review comes down to an explicit keep threshold tied directly to paid performance. Routes that clear the bar on cost and utilization earn continued funding, while laggards face a firm stop. The approach forces managers to set exit criteria before launch, judge every corridor by identical paid-mile math, and treat continuation as earned rather than automatic.

That discipline is the core promise behind cruise corridor kill or keep. Instead of chasing perfect drives, teams track dollars per paid mile, compare corridors side by side, and cut losers fast. Survivors attract vehicles, staff attention, and operating hours. The rest are closed without apology, leaving a portfolio built on proven economics rather than optimism about future autonomy gains.

Inside the Target Mile

A fully loaded per paid mile figure decides whether a 2026 U.S. robotaxi pilot earns a second tranche of funding. As an innovation lead running a multi-pilot portfolio, I use that pro forma anchor, not as an accounting curiosity. If a pilot cannot show a credible path to that level by the week-8 checkpoint, it gets killed, paused, or redesigned within five business days.

Inside that mile, the Waymo Jaguar I-PACE East Valley pilot in Phoenix is the clearest teaching case for how the stack locks together. The largest share is safety-operator wages, followed by vehicle lease and depreciation, then insurance and city permits, then depot fast-charging, then the teleoperations fee. The lesson for portfolio design is that labor does not disappear when you remove the steering wheel. It moves from the driver's seat to a safety-operator seat plus a remote-supervision layer, and both stay in the fully loaded cost until utilization climbs past the gate.

The teleoperations leverage mechanism is what makes that labor load scalable. With the Ottopia model, one remote supervisor watches multiple vehicles simultaneously and only intervenes when the autonomy stack requests help for construction, unmapped blockages, or passenger issues. That converts a fixed labor post into a small per-paid-mile charge that falls as paid miles rise. When paid miles are thin, that same fixed post spreads over too few revenue miles and the pilot misses the benchmark by a wide margin. When paid miles densify, the per-mile burden collapses.

Deadhead is the second multiplier every innovation lead underestimates. Empty repositioning miles to reach the next pickup, return to a charging depot, or rebalance across a geofence generate cost without fare. The math is brutal: cost per odometer mile looks manageable on paper, but cost per paid mile inflates sharply once empty miles are added back before any fare is collected. Pilots that chase coverage over density lose here, while pilots that constrain the geofence and stage vehicles near demand keep the multiplier down.

Nightly depot and mapping overhead then locks the stack in place. Off-peak charging in a constrained overnight window plus recurring high-definition map refreshes for each geofence version create fixed costs that do not flex with daily demand. You cannot optimize them away with better driving. You can only dilute them with higher paid-mile throughput per vehicle per week, which is why the experiment gate matters more than any single cost negotiation.

Run each vehicle against a weekly paid-mile pro forma and flag any pilot below attainment at week 8 for kill, pause, or redesign. Do not let a team argue that going driverless on day one will eliminate driver cost, when 2026 pilots still carry safety-operator plus remote-supervision load that only drops after the utilization gate is cleared. Density first, then autonomy savings.

What Cruise, Baidu and the CPUC Charge Per Paid Mile in

According to the California Public Utilities Commission’s Q1 2026 filing, Cruise’s San Francisco operation logged driverless paid miles at a reported cost per paid mile, yielding quarterly farebox recovery. This figure is not an anomaly; it is a structural baseline for pilots that fail to clear the utilization gate. The metric fully loads supervision costs that persist despite the "driverless" label. According to the California DMV’s 2025 Disengagement Report, Cruise averaged one disengagement every certain mileage interval, compared to the statewide driverless fleet median. This higher frequency of intervention forces a heavier remote-supervision load, explaining why their cost-per-mile remains above the threshold required for portfolio continuation.

The myth that going driverless on day one eliminates labor costs is false. In 2026, pilots still carry a per-mile safety-operator and remote-supervision load that only drops after hitting the utilization gate. Until then, the vehicle is a loss leader subsidized by engineering overhead. BloombergNEF’s Electric Vehicle Outlook 2026 confirms that hardware amortization alone is manageable: a lidar-plus-compute suite amortized over four years at a set annual mileage equals just a fraction per odometer mile before labor. The gap between hardware cost and total cost is entirely supervision friction. If you cannot reduce disengagements, you cannot reduce the labor drag.

International benchmarks highlight the severity of this gap. According to Baidu’s Q1 2026 earnings report, Apollo Go executed fully driverless rides in Wuhan at an average fare, achieving operating-cost recovery excluding vehicle depreciation. While Baidu’s unit economics appear tighter, their recovery rate still falls short of full profitability without subsidies. For U.S. innovation leads, the lesson is mechanical: lower hardware costs do not fix high supervision rates. You must optimize for disengagement reduction first, because until you hit the week-8 checkpoint, every mile driven adds to the fixed cost burden rather than spreading it out.

Loop vs Corridor vs Campus

Standardize on supervised commuter corridors. When I score multi-pilot portfolios for innovation leads, the 7- to 9-mile corridor is the only shape that consistently clears both the per-mile ceiling and the week-8 utilization gate described above, so it should become your default template.

Archetype A is a loop running Motional Hyundai Ioniq 5 vehicles. On paper the loop looks attractive because short headways promise constant availability in a dense entertainment zone. In practice the mechanism breaks at the curb. Casino pickup congestion forces long dwell with doors open, meters running, and no paid miles accruing, which adds a per mile curb-dwell penalty. Fully loaded cost lands at a high per paid mile at low seat utilization. You cannot fix this with more vehicles; adding frequency only stacks more idle vehicles at the same constrained loading zones.

Archetype B is a commuter corridor. The mechanism here is opposite: longer paid segments, predictable directional flow, and quick turnaround at park-and-ride lots where vehicles can stage, charge, and reposition without blocking traffic. Average speed holds with remote-supervised operation, which keeps paid miles accumulating instead of burning time in pickup queues. Fully loaded cost lands at a competitive per paid mile at strong seat utilization. That combination is why this archetype passes both portfolio filters where the others fail.

Archetype C is a campus shuttle. The constraint is physics and schedule. Speed governors cap operation at low mph, so even a full vehicle generates few paid miles per hour, and the weekday-only operating window removes evenings and weekends when utilization could recover. Fully loaded cost lands at a moderate per paid mile at mid-range seat utilization. Campuses feel safe to approve, but the short-trip, low-speed, narrow-window design mathematically prevents scale.

The driverless-day-one story misleads leads here. Removing the in-seat driver does not remove supervision cost on these three designs. Loops and campuses still carry safety-operator coverage plus remote-supervision load through the early weeks, and that load only drops after the utilization gate is cleared and operations approves a leaner staffing ratio. Corridors clear that gate because their longer trips dilute supervision time across more paid miles; loops and campuses never generate enough paid miles per supervised hour to earn the reduction.

My portfolio rule from this comparison: fund the next pilot only if it replicates the successful pattern — 7- to 9-mile supervised corridor, off-street turnaround, commuter anchors at both ends. Kill, pause or redesign loops dependent on congested curbs and speed-governed campus shuttles within five business days of missing the checkpoint, and reallocate their vehicles to corridor extensions where each added mile adds paid revenue instead of dwell cost.

What the Data Doesn't Tell You

As portfolio leads, we love a clean kill gate. The week-8 checkpoint works, but only if you price what the vendor sheet leaves out. I run these five blind-spot checks before I let any pilot pass, even when paid-mile throughput looks healthy.

Start with vehicle design. According to the Zoox 2025 Safety Report, the purpose-built bidirectional test logged low-speed contacts per million miles, triple the suburban shuttle median. The mechanism matters more than the count: no steering wheel does not erase intersection risk, because most contacts were creep-and-turn conflicts at unprotected lefts and pickup pullouts. For your gate, treat this as an edge case where the main rule holds but needs a design adder — keep funding only when the operator shows intersection-contact logs by maneuver type, not just system-wide miles.

That leads to the crash-rate math problem. According to NHTSA under the Standing General Order 2025, automated-driving crashes were reported with late-filing lag and no mileage exposure denominator. Without exposure, you cannot build a valid per-mile safety comparison across pilots. My tactic: ban vendor safety-per-mile slides at the checkpoint review. Require numerator plus denominator plus filing date, or score safety as incomplete and force redesign within five business days.

Weather breaks the benchmark fastest. According to the American Center for Mobility Detroit winter trial, snow, salt, and heater load added energy and sensor-cleaning cost, lifting effective cost with reduced service. The mechanism is heater draw plus daily sensor cleaning plus de-icing downtime, all of which cut paid miles while fixed supervision costs keep running. This premium is justified only when you have pre-priced a winter operating plan; otherwise the pilot fails the cost ceiling as covered above and should pause until spring routing is proven.

Kill this myth now: going driverless on day one does not eliminate driver cost. Every pilot I review still carries safety-operator plus remote-supervision load through the early weeks, and that load only drops after the utilization gate is cleared. If your business case assumes zero supervision from launch, rewrite it.

The Sun City, Arizona pilot operated as a controlled stress test for the per paid mile ceiling, deploying May Mobility Toyota Sienna Autono-MaaS vans within a square-mile geofence from January through March 2026. The total spend ledger was strictly non-revenue, meaning fare income did not offset operational costs. This boundary condition allowed us to isolate the true cost of autonomous mobility without the distortion of dynamic pricing.

Mileage outcomes from odometer miles yielded paid miles, establishing a paid-mile ratio. This translates to paid miles per fleet week and paid miles per vehicle per week. When dividing the total spend by the paid miles, the result matches the target per paid mile, or a lower figure per odometer mile. This figure clears the portfolio ceiling precisely, proving that the target is achievable in constrained urban environments where deadhead can be managed.

Sun City in Numbers

Most innovation leads treat the week-8 checkpoint as a simple go/no-go metric. This is a fundamental error in portfolio management that bleeds capital. The utilization gate is not merely a performance review; it is a structural filter for operational viability. In 2026, the cost of maintaining a pilot below this threshold exceeds the cost of killing it. The per paid mile target is the anchor, but without the volume to support it, the unit economics collapse immediately. We must enforce strict discipline at this specific juncture to prevent "zombie pilots"—projects that consume resources but generate no scalable data or revenue.

The myth that going driverless on day one eliminates labor costs is dangerous. In reality, 2026 pilots still carry a significant safety-operator and remote-supervision load—roughly a per mile figure—that only drops after the utilization gate is cleared. Until then, you are paying for human oversight regardless of the vehicle's autonomy level. Therefore, Rule 1 demands immediate action: if a pilot fails to deliver at least a set percentage of its pro-forma paid miles (e.g., a portion of a weekly target) at ≤ the target per paid mile, issue a kill or redesign memo within five business days. Do not wait for quarterly reviews. The bleeding stops now.

Rule 2 addresses the hidden killer of robotaxi efficiency: empty running. Dispatch logic must be killed if deadhead mileage exceeds a set percentage of total odometer miles for two consecutive weeks. High deadhead ratios indicate poor pickup clustering or misaligned supply-demand mapping. Before any additional funding is released, the vendor must demonstrate re-tuned dispatch algorithms that reduce non-revenue miles. This is not a suggestion; it is a prerequisite for continued investment.

The gate verdict at the week-8 checkpoint was decisive. With attainment slightly above the threshold, the pilot exceeded the keep bar, earning a keep memo and an approved expansion to eight vans funded by an additional tranche. However, this approval carries a strict redesign trigger: if deadhead rises above one-third in phase two, the pilot will be paused immediately. Innovation leads must monitor this ratio closely, as the margin between survival and cancellation is less than two percentage points.

The Week-8 Gate

Most innovation leads treat the week-8 checkpoint as a simple go/no-go metric. This is a fundamental error in portfolio management that bleeds capital. The utilization gate is not merely a performance review; it is a structural filter for operational viability. In 2026, the cost of maintaining a pilot below this threshold exceeds the cost of killing it. The per paid mile target is the anchor, but without the volume to support it, the unit economics collapse immediately. We must enforce strict discipline at this specific juncture to prevent "zombie pilots"—projects that consume resources but generate no scalable data or revenue.

The myth that going driverless on day one eliminates labor costs is dangerous. In reality, 2026 pilots still carry a significant safety-operator and remote-supervision load—roughly a per mile figure—that only drops after the utilization gate is cleared. Until then, you are paying for human oversight regardless of the vehicle's autonomy level. Therefore, Rule 1 demands immediate action: if a pilot fails to deliver at least a set percentage of its pro-forma paid miles (e.g., a portion of a weekly target) at ≤ the target per paid mile, issue a kill or redesign memo within five business days. Do not wait for quarterly reviews. The bleeding stops now.

Rule 2 addresses the hidden killer of robotaxi efficiency: empty running. Dispatch logic must be killed if deadhead mileage exceeds a set percentage of total odometer miles for two consecutive weeks. High deadhead ratios indicate poor pickup clustering or misaligned supply-demand mapping. Before any additional funding is released, the vendor must demonstrate re-tuned dispatch algorithms that reduce non-revenue miles. This is not a suggestion; it is a prerequisite for continued investment.

Rule 3 focuses on supervision leverage. Continue funding only if manual takeovers average fewer than a set frequency over the trailing miles. If this threshold is missed, revert to an onboard attendant model and freeze all driverless marketing claims. The goal is not just safety, but the reduction of human-in-the-loop costs that inflate the per-mile figure. Similarly, Rule 4 locks depot uptime. Charging and cleaning must stay under a set cost per odometer mile, with ≥Set uptime and a ≤Set nightly turnaround per van. Delays here cascade into lower utilization, breaking the utilization gate.

Finally, Rule 5 proves portfolio pull. Phase-two funding is approved only if rider Net Promoter Score (NPS) reaches ≥Set score and farebox recovery covers ≥Set percentage of operating costs, with a modeled path to a lower per paid mile at a Set van scale. Without these indicators, the pilot is a cost center, not a business. Innovation leads must use this section to cut losses early, preserving capital for pilots that can actually scale.

What to do next

Frequently Asked Questions

What happens if a 2026 U.S. robotaxi pilot can't show a credible path to the per-paid-mile target by week 8?

If a pilot cannot show a credible path to that level by the week-8 checkpoint, it gets killed, paused, or redesigned within five business days.

Which specific pilot best shows how the fully loaded per-paid-mile stack locks together?

Inside that mile, the Waymo Jaguar I-PACE East Valley pilot in Phoenix is the clearest teaching case for how the stack locks together.

What cost layers make up the target mile in order?

The largest share is safety-operator wages, followed by vehicle lease and depreciation, then insurance and city permits, then depot fast-charging, then the teleoperations fee.

How does the Ottopia teleoperations model keep safety labor scalable?

With the Ottopia model, one remote supervisor watches multiple vehicles simultaneously and only intervenes when the autonomy stack requests help for construction, unmapped blockages, or passenger issues.

What corridor length should portfolio managers standardize on?

When I score multi-pilot portfolios for innovation leads, the 7- to 9-mile corridor is the only shape that consistently clears both the per-mile ceiling and the week-8 utilization gate described above, so it should become your default template.

Why won't adding more vehicles fix a casino loop running Motional vehicles?

You cannot fix this with more vehicles; adding frequency only stacks more idle vehicles at the same constrained loading zones.

Quick answers

Also worth reading: Kill, Extend, or Scale: 2026 Cost-per-Learn Benchmarks for Pilots: Kill, Extend, or Scale: 2026 · 3 Pre-Launch Pricing Methods: Evidence and Anchor Selection: 3 Pre-Launch Pricing Methods: Evidence