Autonomous Driving and ADAS in 2026: Waymo, Cruise's Exit, Wayve, Mobileye, and the L4 Reality Check
Where commercial L4 actually works, why Cruise wound down, and how Wayve, Mobileye, Nvidia, and Qualcomm split the rest of the ADAS market.
The autonomous driving market in 2026 is two markets in a trench coat. One is the L4 robotaxi market, which is real but narrow, dominated by Waymo, Baidu Apollo Go, Pony.ai, and WeRide, with most of the volume in a handful of geofenced metros. The other is the L2-plus ADAS market, which is much larger and much more boring, and which Mobileye, Nvidia, Qualcomm, and a smaller cluster of Tier 1s actually run. The 2024 wind-down of GM Cruise made the split obvious. The rest of 2025 and the first half of 2026 has been the industry digesting what that means.
This post walks the map.
Waymo: commercial expansion, network reality#
Waymo through 2025 and into 2026 ran paid public service in Phoenix, San Francisco, Los Angeles, Austin, and Atlanta, with weekly ride volume crossing what was reported as roughly two hundred fifty thousand by late 2025. The expansion playbook is consistent — pick a metro, run a long shadow-driving period with safety drivers, switch to driverless in a tight geofence, expand the geofence quarter by quarter. The economics are still gated by per-vehicle hardware cost (Pacifica, I-PACE, and the new Geely Zeekr platform that began trickling in through 2025), per-mile remote-operations overhead, and weather and freeway limitations that the Driver has been steadily but not fully extending past.
The headline read on Waymo is that the technology works inside the operating domain it has been validated in, and the network business is real but capital-intensive. Profitability per metro is the open question.
Cruise’s exit and what it meant for the L4 thesis#
GM announced the wind-down of Cruise as a robotaxi business in December 2024, folding the technology into GM’s personal-vehicle ADAS roadmap. The proximate cause was the October 2023 San Francisco pedestrian incident and the regulator fallout, but the underlying cause was that GM’s board was not willing to fund another three to five years of operating losses against a service-business unit economics curve that had not closed.
The lesson the rest of the industry took: robotaxi is a venture-scale capital commitment with a long payback. Companies without a clear funding ladder — strategic partner, sovereign wealth, or a parent willing to absorb losses — should not be playing in pure L4 robotaxi.
Wayve and the Embodied AI bet#
Wayve, the London-based company, raised more than one billion dollars across 2024 and into 2025 led by SoftBank, Nvidia, and Microsoft. The technical pitch is end-to-end embodied AI — a single foundation driving model that can be deployed across vehicle platforms with minimal HD-map dependence. Wayve’s commercial wedge is licensing the model and the data pipeline to OEMs rather than running a robotaxi service. Nissan and Uber were named partners through 2024 and 2025.
The Wayve thesis is the closest non-Tesla version of the camera-forward, end-to-end neural network approach. If it works across geographies without per-city training, it is the most capital-efficient path the industry has seen.
Mobileye’s three-tier strategy#
Mobileye has carried the volume of the global ADAS market for more than a decade and has restructured the product line into three tiers: SuperVision (eyes-on hands-off), Chauffeur (eyes-off in defined ODD), and Drive (driverless). The 2025 customer wins were uneven — Polestar, Volkswagen, Audi, and several Chinese OEMs continued the SuperVision rollouts, while Geely and Zeekr publicly de-emphasized parts of the relationship through 2024 and 2025. Q4 2024 guidance came down, and the recovery through 2025 was slower than the post-IPO narrative had implied.
The strategic point is that Mobileye still ships more ADAS units per quarter than any other vendor and has the deepest validation library. The L4-versus-L2-plus split is a feature, not a bug, of how the company has positioned.
Nvidia Drive Thor, Qualcomm Ride Flex, and the SoC race#
Nvidia Drive Thor — the successor to Orin, with the Blackwell architecture — began design wins through 2024 and shipped reference platforms in 2025. Drive Hyperion 9 is the integrated reference stack. BYD, Xpeng, Li Auto, Volvo, Polestar, Mercedes, and Hyundai are among the named customers. Qualcomm’s Snapdragon Ride and the Ride Flex platform — which unifies cockpit and ADAS on a single SoC — picked up BMW, GM (post-Cruise pivot), and Renault wins through 2024 and 2025.
The SoC market is now a duopoly with Mobileye as the specialized third option and Tesla’s in-house silicon as the captive fourth. Tier 1s and OEMs are largely choosing one of those four for the next platform cycle.
Autonomous trucking and Apollo Go#
Aurora Innovation launched commercial driverless freight on the Dallas-to-Houston I-45 corridor in April 2024 with Volvo VNL Autonomous and Paccar Peterbilt platforms, and through 2025 expanded the lane network. Plus.ai consolidated with other players through 2024 and 2025. Pony.ai and WeRide both completed US and Hong Kong listings across 2024 and 2025, providing public-market visibility into the unit economics.
Baidu Apollo Go ran the largest commercial robotaxi service by ride volume in Wuhan through 2025, with expansion to additional Chinese cities. The Apollo Go fleet on the RT6 vehicle reached per-mile economics that the US robotaxi operators have not matched.
Regulators, lidar, and the camera-only debate#
The regulatory map matters. NHTSA’s standing general order on automated-driving crash reporting continued through 2025 and 2026 with tightened requirements. The EU GSR-2 phase added ADAS feature mandates across new vehicles. UN ECE WP.29 R157 framed automated lane-keeping system approval in Europe and Japan. Japan METI continued its L4-ready zone program.
The lidar consolidation continued — Hesai and RoboSense became the volume Chinese suppliers, Luminar restructured through 2024 and 2025 against a difficult cash position, and several US and EU startups exited. The camera-only-versus-sensor-fusion debate sharpened after Tesla FSD v13 shipped, but most non-Tesla programs continued to spec lidar plus radar plus camera, citing redundancy and validation cost rather than perception ceiling.
Where pdpspectra fits#
We work with automotive-tech buyers and Tier 1 suppliers on the data-engineering, ML and MLOps, and inference-cost problems that sit underneath ADAS and L4 programs, through /services/ml-mlops and /services/data-engineering. The labeling-pipeline, simulation-validation, and on-vehicle-inference cost problems are all ones we do regularly.
Related reading: the Tesla FSD post, the BYD AI strategy post, and the autonomous trucking post.
Robotaxi is a niche. ADAS is the market. Talk to our team about your automotive AI and data platform.