Autonomy
Not investment advice. This page is for informational and educational purposes only. Sterling is not a registered investment advisor. Nothing here is a recommendation to buy, sell, or hold any security. Markets carry risk and you may lose money. Do your own research and consult a licensed professional before making any investment, securities purchase, or trading decision.

Thesis

Why this sector matters to investors right now. Structural, not market timing.

Autonomous driving in mid-2026 is no longer a binary 'will it work' question. It has split into three commercially distinct businesses: geofenced robotaxis, advanced driver-assist (ADAS), and autonomous trucking. Waymo and Baidu's Apollo Go both reached approximately 250,000 paid rides per week during 2025 (Waymo blog and Yahoo Finance December 2025; CNBC November 3, 2025). Waymo published 170.7 million rider-only miles through December 2025 and reported 92 percent fewer crashes with serious injury or worse than human-driver benchmarks (Waymo Safety Impact December 2025). The investible question is now unit economics and pace of expansion, not feasibility.

The sector has also been through a consolidation. Cruise was shut down by General Motors in December 2024 after roughly 10 billion dollars in cumulative operating losses against less than 500 million in revenue (CNBC December 2024). TuSimple delisted from Nasdaq in January 2024, wound down its US operations, and rebranded to CreateAI in December 2024 (FreightWaves, TechCrunch). Capital and talent have concentrated toward a smaller group: Waymo, Tesla, the Chinese majors (Apollo Go, Pony.ai, WeRide), Amazon's Zoox, and the picks-and-shovels layer (NVIDIA Drive Thor for compute, Mobileye for ADAS, Hesai for lidar).

Structural drivers

Forces that shape long-run demand and economics. Each driver is sourced.
  • •Robotaxi rides are scaling for real. Waymo went from 250,000 paid rides per week in May 2025 to a stated goal of 1 million per week by end of 2026, across 10 US cities and serving 1,400 square miles after May 2026 (Waymo blog; Electrek May 13, 2026). Apollo Go reached 250,000 weekly rides as of October 31, 2025 with cumulative ~17 million rides and 240 million kilometers (CNBC November 3, 2025; Carnewschina November 13, 2025).
  • •Safety data is now publicly verifiable. Waymo reports 0.02 serious injury or worse incidents per million miles versus 0.22 for human drivers, an order-of-magnitude difference at 170.7 million rider-only miles (Waymo Safety Impact, citing NHTSA Standing General Order data). The data is published, not just claimed.
  • •Compute platforms are consolidating. NVIDIA Drive Thor has been adopted by GAC AION Hyper (L4 launch in 2025), XPENG, Li Auto, ZEEKR, BYD, and through Aurora and Continental for autonomous trucking with planned mass manufacture in 2027 (NVIDIA newsroom; Globe Newswire January 7, 2025). Centralizing on a small set of compute platforms reduces software-portability friction.
  • •ADAS is already a real revenue business. Mobileye full year 2025 revenue 1.894 billion dollars, up 15 percent year over year, with operating cash flow of 602 million and a future expected automotive revenue pipeline of 24.5 billion across the next eight years, up 42 percent since 2022 (Mobileye Q4 2025 press release).
  • •Hardware costs are declining. Aurora plans to introduce a second-generation commercial hardware kit in mid-2026 that reduces hardware cost by more than 50 percent (Aurora Q4 2025 reporting via Trucking Dive and Seeking Alpha).
  • •Sensor suppliers are reaching profitability. Hesai reached industry-first full-year GAAP net income of 62 million dollars (non-GAAP 79 million) in 2025 with greater than 40 percent share of the long-range automotive lidar market (optics.org; Markets and Markets).
  • •City-level unit economics in robotaxis are turning positive in the leading markets. Pony.ai Gen-7 robotaxis reached city-wide unit economics breakeven in Guangzhou in November 2025, with daily average orders per vehicle of 232 (Pony.ai IR November 24, 2025).
  • •Autonomous trucking moved into commercial service. Aurora began commercial driverless trucking in Texas in May 2025 and surpassed 100,000 driverless miles by Q3 2025 on routes including Dallas to Houston and Fort Worth to El Paso, with night-driving validation completed (Aurora IR; ACT News).

Structural risks

Forces that could compress demand, change economics, or break the thesis.
  • •Capital intensity remains brutal. Cruise burned more than 10 billion dollars over eight years against less than 500 million in revenue before GM pulled funding in December 2024 (CNBC December 10, 2024). The shutdown showed even well-capitalized programs can lose. Roughly 50 percent of Cruise's workforce was cut by February 2025 (TechCrunch February 4, 2025).
  • •Tesla Full Self-Driving faces an active NHTSA Preliminary Evaluation (PE25012) opened October 2025, covering an estimated 2,882,566 vehicles. ODI identified at least 80 instances of FSD violating traffic rules including red-light running. Of six reported red-light crashes, four resulted in injuries (NHTSA ODI letter; TechCrunch December 5, 2025; CNBC October 9, 2025).
  • •Tesla key program delays. The next-generation AI5 chip volume production has been pushed to mid-2027, meaning the Cybercab will launch on existing AI4 hardware. The first Cybercab unit rolled off Giga Texas in February 2026 with continuous production starting April 2026 (Electrek November 20, 2025 and April 23, 2026). Musk targets unsupervised FSD reaching customer vehicles in Q4 2026, a timeline with a long track record of slipping.
  • •Regulatory regimes are tightening. California adopted new rules in April 2026 allowing police to issue citations to autonomous vehicles starting July 1, 2026 (CNBC April 25, 2025; Teslarati). The EU AI Act came into force August 1, 2024 with enforcement phasing in from August 2026, classifying most AV-related AI as high-risk and requiring integration with the EU Type Approval Framework (Squire Patton Boggs December 2025; Volvo Autonomous Solutions November 2025).
  • •Public capital markets are skeptical of pure-play robotaxi. Pony.ai shares dropped 12 percent and WeRide dropped 13 percent on their respective Hong Kong debuts in November 2025 (CNBC November 6, 2025). Kodiak AI stock fell 37 percent on a May 2026 capital raise priced at a steep discount (TechCrunch May 7, 2026).
  • •Major operator exits set the recent pattern. Cruise (shut down December 2024), TuSimple (US exit December 2023, rebranded CreateAI December 2024), Argo AI (shut down 2022), Embark (shut down 2023). The base rate of operator survival is poor.
  • •Geopolitical exposure on Chinese AV operators. US tariffs and export controls have so far limited Chinese robotaxis to non-US markets. Apollo Go's expansion plan runs through Dubai, Hong Kong, and Europe rather than the US, by necessity.
  • •Insurance and liability frameworks for L4 are still being written. Premium pricing, claims attribution, and product-liability standards have no settled large-fleet equivalent. This is a structural unknown not captured by current operator economics.

Competitive landscape

How to think about the players. Framing along axes (pure play vs diversified, incumbent vs challenger, etc). Not stock picking.

The investible universe sorts into five archetypes.

1. Geofenced robotaxi operators. Waymo (Alphabet) leads in published miles and safety data, operating in 10 US cities by 2026 with plans for Tokyo, London, and 20-plus additional cities. Apollo Go (Baidu) matches Waymo on weekly volume and is the global-expansion leader (Dubai deal for 1,000-plus vehicles, Hong Kong pilot, exploring Switzerland and Turkey). Tesla Robotaxi launched in Austin June 2025 with safety monitors, removed monitors January 2026, and as of May 2026 operates 25 unsupervised vehicles across Austin, Dallas, and Houston (Teslarati). Zoox (Amazon) operates in Las Vegas (free rides since September 2025, paid expected early 2026, San Francisco expansion Spring 2026). Pony.ai and WeRide are dual-listed (Nasdaq plus HKEX) and target international expansion.

2. ADAS and compute layer. Mobileye is the L1-L2 ADAS incumbent (230 million-plus vehicles with EyeQ, 8-year revenue pipeline of 24.5 billion). NVIDIA Drive Thor is the centralized compute platform of choice for Chinese OEMs (XPENG, Li Auto, ZEEKR, BYD, GAC AION) and Aurora. Tesla's vertically integrated FSD competes here as both a software stack and a regulatory-investigation target.

3. Autonomous trucking. Aurora and Kodiak AI are the two surviving US public players. Aurora has 100,000-plus commercial driverless miles by Q3 2025, targets 200 trucks by end 2026, and is integrated with NVIDIA Drive Thor through Continental. Kodiak AI went public via SPAC in September 2025 at ~2.5 billion valuation, operates 10-plus driverless trucks in the Permian Basin, targets H2 2026 for driverless highway operations.

4. Sensors and lidar. Hesai leads at 40-plus percent long-range market share and reached its first full-year GAAP profit in 2025. Luminar is in cost-cut mode with 2025 sales of 67 to 74 million, down from 82 to 90 million. Innoviz, RoboSense, and others occupy specific niches.

5. Exited or pivoted. Cruise (GM, shut down Dec 2024), TuSimple (delisted Jan 2024, rebranded CreateAI as AI gaming company Dec 2024), Argo AI (Ford/VW, shut down 2022), Embark (2023). Capital and talent have largely been re-pooled into the survivors.

Cross-cutting framing: the picks-and-shovels layer (Mobileye, NVIDIA, Hesai) has carried better revenue economics through the cycle than pure-play operators. Operators have larger upside if rides scale, but bear concentrated regulatory and capital-raise risk. Trucking has a different unit-economics shape (long lanes, depot-to-depot, freight-rate exposure) and should not be modeled like passenger robotaxi.

Key metrics to watch

The operational and financial metrics that matter most in this sector. Each one names its source and update cadence.
MetricSourceFrequencyWhy it matters
Weekly paid robotaxi rides per operatorWaymo blog, Baidu earnings calls and Apollo Go disclosures, Pony.ai IR, WeRide IR, Tesla Robotaxi geofence updatesQuarterly with intra-quarter updatesSingle cleanest read on commercial scale. Waymo and Apollo Go both at ~250,000 per week in late 2025 is the current frontier. Growth pace from here is the central debate.
Cumulative rider-only autonomous milesOperator safety reports, NHTSA Standing General Order dataQuarterly or per-operator publicationMiles drive insurance pricing, regulator confidence, and statistical-significance of safety claims. Waymo at 170.7 million miles through December 2025 sets the bar.
Daily rides per vehicle and city-level unit economicsPony.ai IR, Apollo Go disclosures, Waymo selective commentsQuarterlyPony.ai's reported 232 daily orders per vehicle in Guangzhou is the early signal that city-level unit economics can turn positive. Replicability across cities is what scales.
Mobileye design-win pipeline and 8-year revenue forecastMobileye quarterly press releases and IR materialsQuarterly with annual pipeline updateADAS is the largest existing AV-adjacent revenue pool. Mobileye's pipeline (24.5 billion at end of 2025) is the cleanest measure of OEM design commitments for next 5-8 years.
NVIDIA automotive segment revenue and design winsNVIDIA quarterly 10-Q automotive segment, NVIDIA newsroom on Drive Thor announcementsQuarterlyDrive Thor is becoming the de facto centralized compute platform for Chinese OEMs and Aurora. Segment revenue and named wins indicate whether NVIDIA's automotive bet compounds.
Autonomous trucking commercial miles per operatorAurora and Kodiak AI shareholder letters and press releasesQuarterlyDriverless commercial freight miles are the unambiguous proof of L4 trucking scaling. Aurora 100,000-plus miles by Q3 2025 and Kodiak 3 million-plus autonomous miles by September 2025 are the baseline.
Hesai and lidar shipments plus gross marginHesai 6-K and quarterly results, Luminar quarterly, Innoviz quarterlyQuarterlySensor unit economics and margin trends determine whether the lidar layer is a real business or just a subsidized supplier to a money-losing operator stack.
NHTSA investigation and recall activityNHTSA ODI investigation database, NHTSA recalls.gov, Standing General Order reportsContinuous (per-incident)Regulatory pressure can compress operator timelines (Cruise 2023 grounding) or expose adversarial-event rates (Tesla FSD PE25012). The single most actionable safety-signal feed.

Catalysts and milestones

Known upcoming events that could move the sector. Dated where possible.
  • •Tesla Cybercab production ramp at Giga Texas (first unit February 2026, continuous production April 2026) and Musk's stated Q4 2026 target for unsupervised FSD reaching customer vehicles. AI5 chip volume production delayed to mid-2027, so Cybercab launches on AI4. Source: Electrek November 20, 2025 and April 23, 2026; Tesla Q1 2026 earnings call.
  • •Aurora second-generation hardware kit mid-2026 (50-plus percent cost reduction) and target of 200 driverless trucks by end of 2026. Source: Aurora Q4 2025 earnings via Trucking Dive and Seeking Alpha.
  • •Kodiak AI plan to remove safety drivers from highway operations by end of 2026, contingent on validation. Source: Kodiak AI press releases via FreightWaves and Truckinginfo.
  • •Waymo international expansion in 2026 (Tokyo, London announced; 20-plus additional cities total). Freeway service rolling out to more cities. Source: Waymo 2025 year in review (December 2025).
  • •Apollo Go Dubai deployment scaling toward 1,000-plus vehicles under the March 2025 RTA agreement; expansion testing in Switzerland and Turkey. Source: Technology Magazine; KrAsia April 2025.
  • •EU AI Act enforcement phasing in from August 1, 2026, with regulatory sandboxes and automated driving corridors starting in 2026. Source: Volvo Autonomous Solutions November 2025; Squire Patton Boggs December 2025.
  • •California DMV new AV citation rules effective July 1, 2026 (police may ticket AVs). Source: California DMV; CNBC April 25, 2025.
  • •NHTSA Preliminary Evaluation PE25012 of Tesla FSD progresses through 2026. Outcome (defect determination, recall, or closure) is binary for Tesla autonomy revenue model. Source: NHTSA ODI INOA-PE25012-19171.
  • •Mobileye Surround ADAS first two customer programs converting toward production. 19 million unit future-volume pipeline. Source: Mobileye Q4 2025 press release.
  • •Toyota first solid-state EV expected 2027 and Aurora-Continental Drive Thor mass manufacture 2027 are dated downstream catalysts.

What would change the view

Conditions or evidence that would invalidate the thesis or materially shift the risk picture.
  • •Tesla unsupervised FSD publishes safety data comparable to Waymo's 92 percent reduction at 10-plus million rider-only miles. That would validate the cheaper, camera-only architecture and reset cost structures for the whole sector.
  • •A second operator (after Pony.ai) demonstrates city-wide robotaxi unit-economics breakeven on a sustained basis. That confirms the business model is replicable, not city-specific.
  • •A material NHTSA recall or commercial-grade fatality in a leading operator's fleet. That would slow permits, raise insurance costs, and reset regulator timelines (analogous to Cruise 2023).
  • •Mobileye loses a major OEM platform win to a vertically integrated alternative. That would crack the ADAS-incumbent thesis and change the competitive map for compute.
  • •A China-US deal opens (or US restrictions further close) cross-border AV operation. Either direction materially repositions Apollo Go, Pony.ai, and WeRide.
  • •Autonomous trucking unit economics turn negative versus driver-operated trucking when freight rates fall. The current sector positioning assumes long-term cost advantage; freight-cycle downside has not been stress-tested at scale.
  • •Insurance market signals: when L4 fleet insurance pricing becomes published and observable (not bespoke per-operator), the cost-of-risk component of the model stops being a guess.
  • •A new operator (or new fundraise terms for a survivor) appears with credible 1-billion-dollar-plus capital and L4 capability. That would re-fragment the field that has just consolidated.

What we are not covering

Sub-areas, technologies, or companies we are deliberately excluding from the analysis, and why.
  • •Electric-vehicle platforms as such. Many AVs are also EVs, but EV economics (battery cost, charging, market share) are covered in the separate Electric Vehicles sector.
  • •Public charging infrastructure operators. Covered in the separate Charging Infrastructure sector.
  • •Battery cell and lidar materials suppliers below the sensor-module level (lithium, gallium, indium). Materials economics belong in Batteries and adjacent commodities coverage.
  • •Last-mile delivery robots and sidewalk autonomy (Nuro, Starship, Coco). Different regulatory class (low-speed devices, often state or municipal) and different cost structure than on-road AV.
  • •Drones, eVTOL, and aerial autonomy. Different technology stack, certification regime (FAA, EASA), and capital cycle.
  • •Maritime and aviation autopilot. Out of scope for the on-road autonomous-driving thesis.
  • •Personal robotic mowers and vacuums. Consumer-grade autonomy, no transportation use case.
  • •Off-road and agricultural autonomy (John Deere, AGCO autonomy programs). Adjacent and growing but operationally and regulatorily distinct from on-road AV.

Sources

Primary sources cited in this analysis. Links open in a new tab.

Audit trail

Record of the last review and what changed. Required on every refresh.
Last reviewed: 2026-05-14
Change log
  • 2026-05-14Initial publication. All eight required SOP components populated using sourced 2025 full-year and 2026 year-to-date data. Primary sources: Waymo Safety Impact (December 2025), Mobileye Q4 2025 earnings release, Baidu and Apollo Go investor disclosures, Aurora and Kodiak AI shareholder letters, NHTSA ODI PE25012 investigation documents, BloombergNEF lidar and EV outlooks, CNBC, Reuters, Electrek, TechCrunch, FreightWaves, and the California DMV and European Commission regulatory updates. All sources accessed 2026-05-14.
Unresolved questions
  • •Outcome of NHTSA PE25012 (Tesla FSD investigation). Determines whether Tesla's autonomy stack carries a recall liability or progresses.
  • •Tesla unsupervised FSD safety record at scale. Musk's Q4 2026 target for unsupervised customer FSD has high uncertainty; track miles, incident reports, and intervention rates as they are published.
  • •Apollo Go and Pony.ai actual progression of city-level unit economics beyond the initial Guangzhou breakeven datapoint. Need a second city.
  • •Aurora and Kodiak AI cash-runway versus 2026-2027 hardware-cost reduction milestones. Both raised capital under stress (Kodiak May 2026 raise at steep discount).
  • •EU AI Act technical implementation rules for automotive (regulatory sandboxes, automated driving corridors). Detailed approval criteria still emerging through 2026.
  • •Whether Mobileye Surround ADAS first-two customer programs convert to production on stated 19 million-unit pipeline. Conversion timing is the cleanest test of ADAS-incumbent durability versus NVIDIA encroachment.

Ask Sterling

Register for a premium account to gain access to Sterling AI.

Get Started

Things you can ask Sterling:

Summarize Tesla's latest earnings reportWhy did NVIDIA's margins expand?Compare Apple vs Microsoft's cash flowWhat's driving the EV sector growth?
Autonomy - Strategy | Sterling