The autonomous-mobility thesis has crossed its first meaningful verification threshold: Waymo is running a multi-city commercial robotaxi service at genuine scale, and Aurora is converting signed carrier commitments into driverless lane-miles in commercial freight. For the portfolio the question is no longer whether physical AI can operate at scale but whether the capex that built the capability translates into the recurring revenue margin the broader AI-investment thesis requires — and the current operating prints are beginning to answer that, partially, at the wrong layer.
The operating prints
Waymo's verified disclosure as of 31 March 2026 is the highest-signal data in this cycle: 170 million cumulative fully autonomous miles, 4 million weekly miles, 500k+ weekly trips across 10 US cities, and 92% fewer crashes causing serious or fatal injuries versus the human-driver baseline [https://waymo.com/safety/]. By 13 May 2026 coverage had expanded to more than 1,400 square miles across 11 cities [https://waymo.com/blog/]. Florida opened to all residents in Miami and Orlando on 15 April 2026; Waymo announced autonomous operations in London on 14 April 2026 and a Tokyo deployment with Nihon Kotsu and GO on 31 March 2026. What this reveals is that the regulatory bear case — that geofenced deployment would stall at the city-by-city California DMV review pace — is no longer the binding constraint at the marginal city. The bottleneck has shifted to HD-map ingest, depot construction, and supervisor-staffed handoffs at first launch. That shift moves the durable competitive advantage one layer towards operational infrastructure and away from the regulatory calendar.
Aurora's autonomous-trucking ramp is the industrial-freight parallel. The company tripled its driverless network to ten routes in February 2026 and has since added McLane Company for restaurant supply chain (6 May 2026), a Volvo Autonomous Solutions and DSV partnership in Texas (13 May 2026), and a Volvo plus Aurora Oklahoma City launch (4 May 2026) [https://ir.aurora.tech/news-events/press-releases]. The fleet-scale anchor is the 30 April 2026 disclosure that a leading carrier committed to scaling an autonomous fleet to 500 trucks. Aurora is exploiting Texas's permissive AV jurisdiction — no separate state permit required beyond NHTSA compliance — in a way that no Sun Belt-dependent competitor currently replicates at the same pace.
Where the economics settle
This is not a story about the compute layer enriching itself. Aggregated Waymo, Aurora, and Mobileye AV-segment revenue rounds to under $3 billion per year; the gap between that figure and 2026 hyperscaler capex is still sixteen to one. Revenue capture is occurring, but it is accruing at the operational and validation-data layer, not at the training run.
The most durable position in this stack is the combination of fleet-miles, regulator-recognised safety case, and city-by-city permit history. Waymo's 170 million autonomous miles inside Alphabet (GOOGL) constitute that moat; no competitor substitutes it within the product cycle. The second most durable position is AV inference silicon: NVIDIA's (NVDA) DRIVE Thor (~2,000 TOPS, Blackwell-derived) and Mobileye's (MBLY) EyeQ6H are the binding capability gate. Mobileye's vertically-integrated stack — REM crowdsourced mapping plus camera-radar-lidar fusion sold OEM by OEM — is the leverage point because it scales without the depot-by-depot capital lift Waymo absorbs. Mobileye's design-win catalogue is wide: a Polestar 4 Chauffeur integrating Luminar lidar tracking to production, Volkswagen Group's SuperVision plus Chauffeur production series, and a second top-ten automaker added to the EyeQ6H platform in January 2026 [https://www.mobileye.com/news/]. In Oslo, Ruter and Holo selected MOIA with Mobileye Drive for deployment beginning spring 2026 — the platform-agnostic counterweight to Waymo's vertically-integrated approach — and Lyft (LYFT) announced commercial fleet operator integration with Mobileye in November 2024.
The position I regard as most under-priced at the 2026 horizon is the network-rake layer. Uber (UBER) captures the matching and payments rake on Waymo trips in Austin, Atlanta, and Phoenix without owning a single lidar unit. Its partnership-with-everyone strategy — Waymo, Aurora, Wayve — is structurally favoured by the multi-vendor reality: no global monopolist emerges within five years, which makes the indifferent dispatcher the default beneficiary. On the other side of the ledger, pure-play sensor suppliers face structural margin compression. Luminar (LAZR) is down more than 70% year-to-date into May 2026, and the mechanism is Hesai's (HSAI) Chinese-domestic cost-down creating a price floor that Luminar's unit economics cannot clear — the same dynamic the prior humanoid scan identified between Suzhou Green Harmonic and Harmonic Drive Systems. That China-commodity-tier versus Western-sovereign-tier dynamic in a physical component neither OEM nor operator can easily bypass is becoming a recurring structural pattern.
Tesla (TSLA) is the most consequential wildcard and the clearest absence. The Robotaxi service in Austin remains supervised at the time of this scan, and no GAAP Robotaxi revenue line appears in the Q1 2026 Form 10-Q, against a CEO target of fully unsupervised operation by end-2026. The gap between aspiration and regulatory status is not narrowing at a pace visible in the primary-source record, and at roughly 176 times forward earnings the stock is pricing a probability of unsupervised scale that the operating record does not support.
The second-order displacement is worth naming even if its timeline is longer. The directly displaced cohorts — roughly 3.4 million heavy and tractor-trailer drivers (BLS OEWS 53-3032) and 1.8 million taxi and rideshare drivers (BLS 53-3054) — represent a labour-displacement second wave running on a five-to-ten-year lag behind the white-collar first wave. It is a different political coalition, skewing towards middle-aged male workers without college degrees, and the payroll-tax mechanism fires through a separate channel rather than compounding the same one.
What would change my mind
A serious-injury crash at scale involving Waymo or Aurora, followed by an NHTSA enforcement action suspending operations for more than 30 days, would reset the regulatory calendar back towards the stall-at-city-by-city thesis I have largely dismissed. If Tesla Robotaxi's active vehicle count remains below 1,000 at year-end 2026 disclosure, or unsupervised approval is pushed past Q1 2027, the end-to-end-vision path closes as a live competitive threat for this cycle and the optionality embedded in TSLA's multiple becomes very difficult to defend. A state-level AV moratorium in Texas or California — the two jurisdictions Aurora and Waymo respectively depend on most — triggered by a high-visibility incident would do equivalent structural damage to the base case.
What I'm watching next
Mobileye Chauffeur series-production unit disclosures — the first print that converts the design-win catalogue into an actual revenue line. Aurora's Q2 2026 route count and any named-carrier confirmation on the 500-truck commitment. NHTSA's pending rulemaking on AV exemptions under Part 555, which sets the federal floor distinguishing supervised from unsupervised operation in practice. Two adjacent gaps this scan surfaces but cannot resolve within its scope: the personal-auto versus commercial-fleet insurance pool re-pricing at Allstate (ALL), Progressive (PGR), and Travelers (TRV), which warrants a dedicated scan; and the Apollo Go, Pony.ai (PONY), WeRide (WRD), and Hesai complex as a China-AV rotation, where HFCAA delisting risk and the Outbound Investment Order create a valuation overhang distinct from anything in the US-centric thesis.