Six alternative branches. These are subjective probabilities from the seed, not observed frequencies or investment recommendations.
Why this trend
Load-bearing assumption
No operator publishes robotaxi unit economics - revenue per vehicle-hour, remote-assistance ratio or contribution margin - so scale and viability continue to be inferred from regulator-reported miles rather than from disclosed margin.
Preconditions
Not recorded.
More metros, same economics
Driverless ride-hail keeps growing fast from a small base and stays small relative to total travel. Waymo adds metros and vehicles, Zoox extends beyond the Las Vegas Strip, and weekly paid rides climb from roughly 500,000 toward a few million by 2030 - still under 1% of US vehicle miles travelled. Deadhead stays in the low-to-mid forties, occupancy stays near 1.4, and no operator publishes a contribution margin. The service is real, commercial and unprofitable in public.
Mechanism
The decelerating element is deadhead: the passenger-onboard share plateaued at 55-57% from mid-2025 and service-area expansion adds repositioning miles as fast as fleet density removes them. Growth comes from adding metros and vehicles rather than from utilisation, which means cost per paid mile falls slowly. Remote-assistance labour, whose ratio to vehicles no operator discloses, scales with the fleet rather than with rides.
Preconditions, early indicators and assumptions
- Calibration Basis
base_rate
- Preconditions
- No high-severity incident triggering a permit suspension
- CPUC and California DMV continue publishing mileage data
- Capital remains available for fleet expansion at a 3.50-3.75% policy rate
- Early Indicators
- Indicator
CPUC quarterly reports showing the no-passenger share of driverless deployment vehicle miles at or above 40% for the quarter ending 2027-06-30
- Source
S-10-06
- Threshold
40%
- Direction
above
- Would Be Visible By
2027-12
- Resolves
Confirms base; below 35% moves mass to disruption
- Indicator
California DMV aggregate autonomous test miles for the December 2026 to November 2027 reporting period between 9 million and 20 million
- Source
S-10-07
- Threshold
9-20 million miles
- Direction
within band
- Would Be Visible By
2028-03
- Resolves
Confirms growth from a small base; the corpus records over 9 million for the prior period
- Indicator
Alphabet continuing to report Waymo inside Other Bets with no separate revenue and operating income
- Source
S-10-15
- Threshold
no segment breakout
- Direction
fails to occur
- Would Be Visible By
2028-02
- Resolves
Confirms that the economics are not being disclosed, which is the base-case condition
- Assumptions
- Text
The deadhead plateau at 43-45% persists rather than improving materially
- Confidence
medium
- Load Bearing
true
- Basis
Peer-reviewed CPUC-data analysis, 2026-05-18: 46.4% of 86,269,177 vehicle miles carried no passenger across Aug 2023 to Dec 2025, with the passenger-onboard share flat at 55-57% from mid-2025
- If Wrong
Base and disruption swap probability mass and the disruption indicator has already fired
- Text
Weekly paid rides grow but stay under 1% of US VMT through 2030
- Confidence
medium
- Load Bearing
false
- Basis
California's entire commercial robotaxi history to end-2025 is on the order of 0.003% of one year of US light-duty travel
- If Wrong
The scale claim changes but the economics question does not
- Affected Industries
- 10
- 01
- 07
- Precedence Note
Distinguished from disruption by the regulator-reported no-passenger share, not by metro count or announced expansion. Adding cities while deadhead stays above 40% is base.
- Would Change Our Mind
Any operator disclosing deadhead below 35% or a positive contribution margin per vehicle-hour.
- What Businesses Should Do
Plan for driverless availability in specific dense metros, not for fleet replacement. Treat metro counts as marketing and regulator-reported vehicle miles as the measure; the corpus records that the fourteen-cities figure counts markets while the mileage distribution is extremely skewed, with Phoenix at 80.6 million rider-only miles against Atlanta at 5.4 million.
Utilisation improves and capital follows
The constraint that releases is fleet density within existing service areas rather than geographic expansion. Deadhead falls into the thirties as depot placement, pooling and demand prediction improve, occupancy rises above 1.4, and the cost per paid mile falls fast enough that an operator chooses to publish a contribution margin because it is favourable. Capital floods in and the fleet grows faster than the metro count.
Mechanism
Operators stop optimising for coverage and start optimising for utilisation. The specific blockage is repositioning distance, released by higher vehicle density per square mile and by depot siting. Who releases it: the operators themselves, and the evidence would be visible in CPUC mileage tables before it is visible in any announcement. New demand does not appear from nowhere; it comes from substituting for existing ride-hail trips in the same metros, which is a documented, priced demand pool.
Preconditions, early indicators and assumptions
- Calibration Basis
analogue
- Preconditions
- Deadhead falls below 35% in at least one operator's regulator-reported data
- Fleet size grows faster than service-area square mileage
- No permit suspension in a major market
- Early Indicators
- Indicator
CPUC quarterly reports showing the no-passenger share of driverless deployment vehicle miles below 35%
- Source
S-10-06
- Threshold
35%
- Direction
below
- Would Be Visible By
2028-06
- Resolves
Confirms upside
- Indicator
Alphabet reporting Waymo as a separate reportable segment with disclosed revenue and operating income or loss
- Source
S-10-15
- Threshold
segment breakout in a 10-K or 10-Q
- Direction
occurs
- Would Be Visible By
2028-02
- Resolves
A company breaks out a number when it has become material or when it is favourable
- Assumptions
- Text
Deadhead is an addressable engineering and siting problem rather than a structural feature of low-density demand
- Confidence
low
- Load Bearing
true
- Basis
The passenger-onboard share stopped improving in mid-2025 after two years of improvement, which is evidence against
- If Wrong
Upside collapses into base; nothing moves to downside
- Text
Remote-assistance ratios improve alongside deadhead
- Confidence
low
- Load Bearing
false
- Basis
No operator discloses the ratio; the corpus names this as a hidden variable in robotaxi unit economics
- If Wrong
Cost per mile falls more slowly than deadhead alone implies
- Affected Industries
- 10
- 01
- Precedence Note
Requires regulator-reported deadhead below 35%, not operator commentary on efficiency. Commentary alone leaves the outcome in base.
- Would Change Our Mind
Two more CPUC quarters with the no-passenger share above 43%.
- What Businesses Should Do
Ride-hail operators and fleet lessors should track CPUC deadhead quarterly as the single best public proxy for whether the cost line is moving.
Expansion stalls on cost per mile
The thesis holds but delivers less. Deadhead stays high, remote-assistance labour scales with the fleet, and vehicle cost plus depot capex plus a 3.50-3.75% cost of capital keep contribution margin negative. Operators slow metro additions and concentrate on the densest markets, weekly ride growth decelerates, and the sector consolidates as smaller pilot operators exit. Driverless ride-hail becomes a profitable niche in three or four metros rather than a general transport mode.
Mechanism
The deteriorating parameter is cost per paid mile. It shows up first in the pace of new-metro launches, then in CPUC quarterly vehicle-miles growth flattening, then in pilot operators ceasing to file. The transmission channel is capital: the corpus records the FOMC flagging high AI-firm valuations and leveraged infrastructure financing as a stability risk, and robotaxi fleets are exactly that kind of financing.
Preconditions, early indicators and assumptions
- Calibration Basis
base_rate
- Preconditions
- No improvement in the regulator-reported deadhead share
- Capital cost stays at or above current levels
- No step change in vehicle unit cost
- Early Indicators
- Indicator
Total California driverless deployment vehicle miles growing less than 20% year over year in CPUC quarterly reports
- Source
S-10-06
- Threshold
20% y/y growth
- Direction
below
- Would Be Visible By
2028-03
- Resolves
Confirms downside; the corpus records roughly 15% average monthly growth to end-2025, so a fall to under 20% annual is a sharp deceleration
- Indicator
Two or more operators ceasing to file CPUC pilot or deployment quarterly reports
- Source
S-10-06
- Threshold
two operators
- Direction
occurs
- Would Be Visible By
2028-06
- Resolves
Confirms consolidation; one exit is a company event
- Assumptions
- Text
Remote-assistance labour scales close to linearly with fleet size
- Confidence
low
- Load Bearing
true
- Basis
No operator discloses the ratio, so this is judgement; the corpus names it as the hidden variable in robotaxi unit economics
- If Wrong
Cost per mile falls faster than modelled and mass moves to base and upside
- Text
Vehicle unit cost does not fall sharply within the horizon
- Confidence
medium
- Load Bearing
false
- Basis
Waymo operated roughly 3,067 fifth-generation vehicles per December 2025 NHTSA filings; no disclosed unit cost
- If Wrong
The capex leg improves and the branch is narrower
- Affected Industries
- 10
- 07
- 01
- Precedence Note
Stopper: capital cost. Distinguished from failure by whether permits are withdrawn; downside is expansion slowing, failure is permission being removed.
- Would Change Our Mind
An operator launching in three or more new metros in a single year while deadhead stays flat, which would mean capital is available regardless of unit economics.
- What Businesses Should Do
Do not underwrite depot real estate or fleet finance on announced metro expansion. Require regulator-reported mileage growth in the metros already launched.
Purpose-built vehicles reset the cost base
A different mechanism wins: vehicles designed without manual controls, built to a robotaxi cost target rather than adapted from consumer cars, change the capex per vehicle-hour. NHTSA's first-ever commercial FMVSS exemption to Zoox on 2026-07-30 for up to 2,500 vehicles annually created the lawful route; a second and third exemption would make it a category. The incumbent asset that loses value is the retrofit sensor stack bolted onto a production passenger vehicle, and the cost delta is the deleted steering column, mirrors, windshield structure and occupant-protection hardware.
Mechanism
Purpose-built vehicles remove eight FMVSS-driven cost and packaging constraints and allow interior layouts optimised for occupancy rather than driving. At least two parties are pursuing it: Zoox with a granted exemption and Robomart with an application under review as of 2026-07-30. If unit cost falls far enough, the deadhead problem matters less because the asset is cheaper to leave empty.
Preconditions, early indicators and assumptions
- Calibration Basis
analogue
- Preconditions
- At least two further Part 555 commercial exemptions granted
- Purpose-built vehicles reach a production rate above the current 2,500 annual cap
- No safety event involving an exempted vehicle
- Early Indicators
- Indicator
NHTSA granting at least two further commercial FMVSS exemptions for purpose-built driverless vehicles
- Source
S-10-09
- Threshold
two further grants
- Direction
occurs
- Would Be Visible By
2027-12
- Resolves
One grant is a pilot; three is a pathway
- Indicator
An exemption application or grant covering more than 2,500 vehicles annually
- Source
S-10-09
- Threshold
2,500 annual cap exceeded
- Direction
above
- Would Be Visible By
2029-12
- Resolves
Confirms the cost-base reset requires volume, which the current cap forecloses
- Assumptions
- Text
Purpose-built vehicle cost per unit falls materially below adapted consumer vehicles at the volumes the exemption caps permit
- Confidence
low
- Load Bearing
true
- Basis
No operator discloses vehicle unit cost, and the 2,500 annual cap is far below automotive volume thresholds
- If Wrong
This branch stays a project and its mass returns to base
- Text
NHTSA treats the Zoox grant as a template to repeat rather than a pilot to evaluate
- Confidence
low
- Load Bearing
false
- Basis
One grant in the programme's history as of 2026-09-15
- If Wrong
The pathway does not open and the branch is unavailable
- Affected Industries
- 10
- 09
- 13
- Precedence Note
Requires two further exemptions. A single further grant resolves to base.
- Would Change Our Mind
NHTSA citing the A2SCEND standards work as a reason to defer further exemptions.
- What Businesses Should Do
Vehicle manufacturers and tier-1 suppliers should treat the Part 555 exemption cap, not technical readiness, as the near-term volume constraint on purpose-built designs.
Federal standards replace the patchwork
The binding constraint shifts from state permits to federal standards. NHTSA updates its AV guidance for the first time since 2017, the three-year SAE A2SCEND consortium produces the first national AV performance standards, and Part 555 exemption caps are replaced by compliance with a standard. The direction is genuinely two-sided: a workable standard is liberalising and accelerates deployment; a demanding one is restricting and advantages large incumbents over new entrants.
Mechanism
The instruments are NHTSA's interim final rule of 2026-07-30 on pre-exemption-date vehicles, the Part 555 exemption regime, the updated AV guidance covering emergency-responder interaction, safety management systems, remote assistance and post-crash behaviour, and the A2SCEND standards programme funded at USD 5m over three years. The stage to watch is from guidance to rulemaking, which is where the four-stage distinction between proposed, enacted, in force and enforced actually bites.
Preconditions, early indicators and assumptions
- Calibration Basis
base_rate
- Preconditions
- A2SCEND produces draft performance standards
- NHTSA issues updated AV guidance
- No incident forcing an emergency rulemaking, which would move this to failure
- Early Indicators
- Indicator
NHTSA publishing updated AV guidance covering remote assistance and post-crash behaviour
- Source
S-10-09
- Threshold
guidance published
- Direction
occurs
- Would Be Visible By
2028-06
- Resolves
Confirms the federal-framework leg
- Indicator
A notice of proposed rulemaking establishing AV performance standards rather than an exemption-based pathway
- Source
S-10-10
- Threshold
NPRM published
- Direction
occurs
- Would Be Visible By
2029-12
- Resolves
Confirms regulatory; exemptions alone stay in base or disruption
- Assumptions
- Text
A three-year standards programme funded in July 2026 produces standards that NHTSA adopts rather than shelves
- Confidence
medium
- Load Bearing
true
- Basis
NHTSA has not updated its AV guidance since 2017, which is nine years of stated intent without a standard
- If Wrong
Mass moves to base with the state patchwork and exemption caps persisting
- Text
Federal standards preempt rather than layer on top of state permit regimes
- Confidence
low
- Load Bearing
false
- Basis
California's CPUC and DMV regimes are independent of FMVSS and would not automatically be displaced
- If Wrong
Operators face both regimes and the compliance cost rises rather than falls
- Affected Industries
- 10
- 09
- 07
- Precedence Note
Regulatory outranks base when a standards NPRM issues, even if deployment volumes look base-like. Name the docket.
- Would Change Our Mind
A2SCEND going quiet for four consecutive quarters.
- What Businesses Should Do
Track the guidance-to-rulemaking transition, not the guidance itself. Guidance changes nothing legally; an NPRM changes what can be built.
Robotaxi winter II
A high-severity, well-documented driverless fatality in a major market triggers permit suspensions before a national standards framework exists. Because IIHS has stated the present data-collection system cannot support monitoring a scaled deployment, the regulatory response is blunt rather than targeted. Operators pause in the affected state and pre-emptively elsewhere, capital withdraws, and the sector re-runs 2023 at larger scale. The thesis stops rather than slows.
Mechanism
The historical analogue is dated and resolved: GM's Cruise held a California driverless deployment permit, the DMV suspended it on 2023-10-24, and the company ceased meaningful operation by December 2023 - weeks, not quarters. Who is left holding the asset: fleet owners with depreciating purpose-built vehicles, depot lessors, and the remote-assistance workforce. The recovery period on the Cruise precedent was terminal for that operator and roughly two years for sector sentiment.
Preconditions, early indicators and assumptions
- Calibration Basis
analogue
- Preconditions
- A high-severity event with a clear driverless attribution
- No national standards framework in force at the time
- A state regulator with unilateral suspension power
- Early Indicators
- Indicator
NHTSA opening a defect investigation into a driverless fleet with more than 1,000 vehicles
- Source
S-10-09
- Threshold
one investigation, fleet above 1,000 vehicles
- Direction
occurs
- Would Be Visible By
2029-12
- Resolves
Leading indicator for failure
- Indicator
Any state suspending an active driverless deployment permit
- Source
S-10-06
- Threshold
one suspension
- Direction
occurs
- Would Be Visible By
2029-12
- Resolves
Confirms failure on the Cruise precedent
- Assumptions
- Text
The regulatory response to a severe event would be a blunt suspension rather than a targeted condition, because the monitoring data to support a targeted response does not exist
- Confidence
low
- Load Bearing
true
- Basis
IIHS stated on 2026-07-23 that the present data-collection system is not good enough to allow continuous monitoring of a large-scale expansion
- If Wrong
The response is narrow, the branch is near-remote and mass returns to base and regulatory
- Text
The IIHS finding of 68% fewer police-reportable crashes per mile does not by itself insulate an operator from suspension
- Confidence
medium
- Load Bearing
false
- Basis
City-level variation was wide - Phoenix 76% lower, Austin 4% higher on a small sample - and Cruise was suspended despite holding a permit
- If Wrong
Independent safety evidence becomes a genuine regulatory shield and the branch narrows
- Affected Industries
- 10
- 07
- 11
- Precedence Note
Failure outranks all other branches on occurrence. The distinguishing observable is a permit suspension or a fleet-wide defect investigation, not a single crash report.
- Would Change Our Mind
A national AV performance standard reaching enforcement before 2029, which would replace unilateral suspension with a defined compliance response.
- What Businesses Should Do
Anyone with capital in this sector should know which single regulator can suspend the largest operator, and should hold that as a named counterparty risk rather than as a background regulatory risk.
Additional scenario notes
- Review Required
true
- Review Reason
Autonomous-vehicle safety and liability. IIHS has stated that the present data-collection system is not good enough to allow continuous monitoring of a large-scale expansion, which is a limitation on every safety statement in this set.
- Probabilities Sum
1
- What Must Be True To Grow
- Capability: the vehicles work in someone else's hands at scale - satisfied; 220.6 million cumulative rider-only miles through March 2026 and independent IIHS analysis of roughly 50 million driverless miles
- Economics: does it pay - unknown, and the corpus is explicit that no operator anywhere publishes robotaxi unit economics
- Supply: vehicles, depots and remote-assistance staff at the required rate - unquantified; no operator discloses the remote-assistance ratio
- Demand: paying riders distinct from promotional volume - satisfied; roughly 500,000 weekly paid rides as of March 2026, growing from roughly 50,000 in May 2024
- Permission: state permits plus, for purpose-built vehicles, a federal FMVSS exemption - partially satisfied, with one commercial exemption granted in the programme's history
- Capital: fleet finance at a 3.50-3.75% policy rate - available today to a small number of well-capitalised operators only
- What Could Stop It
- Capital cost: fleet and depot capex against an undisclosed and possibly negative contribution margin, in the rate environment the macro brief describes
- Political licence: a severe incident making the activity unacceptable before it becomes uneconomic, which is the Cruise mechanism and resolved in weeks
- Physical constraint: deadhead at 46.4% of vehicle miles and occupancy at 1.4, with the passenger-onboard share having stopped improving in mid-2025
- Measurement revision: IIHS and Waymo's own figures both rest on NHTSA Standing General Order reporting, which IIHS itself criticises as inadequate, so the safety base is not independent
- Attention withdrawal: capital rotating to other AI applications if no operator publishes a positive contribution margin
- Uncertain Assumptions
- That the deadhead plateau at 43-45% persists - load-bearing for the base case, confidence medium, and the single most decision-relevant number in the sector
- That remote-assistance labour scales close to linearly with fleet size - confidence low, and no operator discloses the ratio
- That Chinese operator claims are comparable - Baidu's Q2 2026 release gave 28 cities and over 240 million fully driverless kilometres but dropped the quarterly ride count it had previously disclosed, so disclosure got less specific as claims got bigger
- That Tesla robotaxi scale is immaterial to the sector picture - unverifiable, because Texas has no equivalent of California's reporting regime and all figures are third-party tracker estimates
- Authored
2026-09-15