AI and Mobility in 2026: Autonomous Vehicles, ADAS, Smart Cities, and the Logistics Stack

Waymo started 2025 at roughly 175,000 weekly rides and ended it above 450,000 — about a 157 percent increase in a single year — while operating a fleet of around 2,500 vehicles across Phoenix, San Francisco, Los Angeles, Austin, and Atlanta. In the same twelve-month window, GM wound down Cruise’s robotaxi operation in December 2024 and folded its autonomous work back into personal-vehicle development. Aurora Innovation began regular driverless freight runs on the I-45 corridor between Dallas and Houston in May 2025. Lilium filed for insolvency in Germany at the end of October 2024 and terminated its remaining 750 employees that December, before a reconstituted “Lilium Aerospace” emerged in early 2025. BYD passed 2.3 million vehicles equipped with its God’s Eye ADAS in China by November 2025. All of that is one industry, one year.

This post walks through what is actually shipping in AI mobility in 2026 — robotaxis, ADAS, trucking, eVTOL, and the city-scale layer underneath all of it.

The robotaxi layer: Waymo extends, Cruise exits, Zoox arrives#

Waymo is the only U.S. robotaxi operator at meaningful scale in 2026. The 450k-rides-per-week figure at end of 2025 puts it on the order of tens of millions of completed paid trips lifetime. The geographic spread by mid-2026:

• Phoenix — about 315 square miles, 500 vehicles, mature operations and the testbed for new service types (airport, freeway segments, longer-distance fares).
• San Francisco — 800 to 1,000 vehicles, the densest deployment by vehicle count.
• Los Angeles — around 700 vehicles, continuing geographic expansion.
• Austin — launched March 2025 via the Uber app at 37 square miles, expanded to 90 square miles by July 2025, around 200 vehicles.
• Atlanta — launched mid-2025 via Uber, around 100 vehicles.

The Cruise exit in December 2024 is the other half of the same story. GM stopped funding the robotaxi commercialization push, took the ADAS technology in-house for Super Cruise, and let a substantial percentage of the Cruise headcount go. The takeaway in 2026 is not “robotaxis don’t work.” It is closer to “robotaxis don’t work for an automaker that needs them to clear an automaker’s hurdle rate on an automaker’s timeline.” Waymo’s economics are still negative on a unit basis in most cities, but Alphabet has the patience and the cash to wait.

Behind Waymo, the credible next entrants are Amazon’s Zoox (custom purpose-built vehicle, limited public rides in Las Vegas and SF as of late 2025) and Tesla’s Cybercab program. Tesla’s FSD v13 rollout through late 2025 pushed end-to-end neural-network behavior into the customer fleet at scale, and the Cybercab unveiled in October 2024 is the first vehicle Tesla designed without a steering wheel. The question that matters for 2026 is whether Cybercab service moves from demonstrations to fared rides in a real city, not whether the vehicle exists.

The Chinese robotaxi and ADAS lead#

The most underweighted mobility story in Western coverage is China. Pony.ai and WeRide both completed U.S. IPOs in late 2024 — Pony.ai listed on Nasdaq in November 2024 — and operate driverless robotaxis under permit in all four Tier-1 Chinese cities. Baidu’s Apollo Go ran the largest single-city robotaxi fleet in the world out of Wuhan through 2024 and 2025, before expanding the operating model to multiple cities.

On the ADAS side, the Chinese OEMs are now ahead of most of the European stack on shipped feature volume:

• BYD God’s Eye — across more than 2.3 million vehicles by November 2025, including roughly 311,000 sold in that month alone. God’s Eye C uses twelve cameras, five mm-wave radars, and twelve ultrasonic sensors, with front-radar detection out to about 300 meters. BYD is rolling God’s Eye out across the full lineup — including the Seal 05 DM-i, which puts an ADAS package on an affordable hybrid sedan — as part of a “National Smart Driving” strategy.
• Xpeng XNGP — end-to-end neural network for urban and highway scenarios, one of the more aggressive proprietary pivots in the market.
• Huawei ADS 3.0 — Tier-1 supplier turnkey solution used by multiple OEMs without in-house capability, alongside Momenta, Horizon Robotics, and DJI spinoff Zhuoyu.

The competitive picture in 2026 is that BYD has the volume, Xpeng and NIO have the highest-end city-driving features, and Huawei is the default Tier-1 for anyone else.

The silicon underneath: Nvidia, Mobileye, Qualcomm#

The 2026 silicon stack for shipping ADAS is a three-vendor race. Nvidia Drive Thor is the high-end SoC anchoring the next generation of premium ADAS and robotaxi platforms. Mobileye continues its three-tier strategy — EyeQ for high-volume baseline, SuperVision for hands-off highway and basic urban, Chauffeur and Drive for higher autonomy levels. Qualcomm Snapdragon Ride Flex has moved from announcement to design wins across multiple OEMs through 2025. The interesting reshuffle is that several Tier-1s now ship dual-vendor reference platforms so OEMs can swap the compute layer without rewriting the perception stack.

Wayve, the UK end-to-end driving startup, sits in a different category — it is selling a model and a software stack rather than silicon, and its embodied-AI approach is increasingly cited as the architecture every other player is trying to replicate.

Autonomous trucking: Aurora moves freight, the rest catch up#

Aurora’s commercial launch on the Dallas-Houston I-45 corridor in May 2025 — with Uber Freight and Hirschbach Motor Lines as the launch customers — was the first regular driverless freight service in the United States. Aurora’s stated 2025 plan extended the service to El Paso and Phoenix by year-end. The economics are still unproven at scale but the operational pattern — long-haul highway segments, with humans handling the first and last yard — has held up.

Behind Aurora, the credible trucking players in 2026 are Kodiak Robotics, Plus, and the partnerships that Daimler Truck and Volvo Autonomous Solutions have built around their own platforms. Industry surveys from NACFE through 2025 consistently note that the autonomous-trucking case is stronger than the robotaxi case — fewer edge cases, more freeway miles per pickup-and-drop event, and an operator (the freight customer) that already knows how to measure cents per mile.

eVTOL: Joby and Archer hold, Lilium falls#

The advanced air mobility story in 2026 is two survivors and one cautionary tale.

• Joby Aviation holds the FAA Part 135 air-carrier certificate, has worked through the structural phases of type certification, and continues toward commercial passenger service with Delta as a partner.
• Archer Aviation holds Part 135 (June 2024) and Part 145 maintenance (February 2024), with a partner consortium that includes Stellantis, Boeing, and United. Archer also acquired Lilium’s intellectual property in the post-bankruptcy auction.
• Lilium filed for self-administration in Germany at the end of October 2024, terminated its 750 remaining employees on December 20, and was reconstituted as “Lilium Aerospace” in January 2025 under a new investor consortium that pledged about 200 million euros.

The economics question for eVTOL in 2026 is unchanged: route economics, vertiport availability, and the willingness of a high-density city to permit the airspace. The technology is closer than it has ever been; the operating environment is the constraint.

Smart-city and urban-mobility AI#

Underneath the vehicles is a quieter layer that has moved faster than the vehicles. Singapore’s Smart Nation programme has integrated multi-modal mobility analytics into traffic management for years; Dubai 2030 continues to push toward a stated 25 percent autonomous-transit share with sustained procurement; Helsinki’s mobility-as-a-service work remains a reference for European cities.

The vendor layer in 2026 is dominated by:

• StreetLight Data (now part of Jacobs) and Replica — mobility analytics from anonymized device and partner data, used heavily by departments of transportation.
• Optibus and Swiftly — transit operations and scheduling AI used by hundreds of agencies.
• Iteris, Rapid Flow Technologies (Surtrac), NoTraffic — signal optimization at intersection and corridor scale.

The pattern across all of these is the same: the AI is in the planning, scheduling, and optimization layer, not in the vehicles. That layer pays back faster and has fewer regulatory blockers than anything happening on the autonomous-vehicle side.

Insurance, liability, and the question nobody has answered#

The unsolved question across robotaxi, autonomous trucking, and high-autonomy ADAS is liability when something goes wrong. Waymo carries the liability itself today. Tesla’s structure pushes most liability to the driver under “supervised FSD” language. Aurora carries trucking liability through an insurance program built with Volvo and others. There is no shared framework yet, and the state-level patchwork in the U.S., the federated approval pattern in the EU, and the city-by-city permitting model in China each push the answer in a different direction. Anyone deploying AI mobility in 2026 should expect the liability framework to be a meaningful piece of the operating model — not a footnote.

Where pdpspectra fits#

Most of our mobility work is not building self-driving stacks. It is the data and infrastructure layer that fleet operators, OEMs, and city agencies need underneath any of this — telematics ingestion at scale, route and dispatch optimization, ML pipelines that move models from notebook to vehicle on a defensible cadence, and the cloud platforms that hold it all up. That work sits across our ML and MLOps practice, data engineering practice, and cloud infrastructure practice.

The conversations we have most often in 2026 are with fleet operators sizing their AI roadmap and with city agencies trying to consolidate mobility data without committing to a single vendor’s lock-in.

Related reading#

• /blog/ai-autonomous-driving-adas-2026 — the perception and end-to-end driving stack in more depth
• /blog/autonomous-trucking-2026 — long-haul freight economics and the Aurora playbook
• /blog/ai-tesla-fsd-robotaxi-2026 — Tesla’s FSD trajectory and Cybercab program

Closing#

AI mobility in 2026 is not a single industry — it is at least five overlapping ones moving at different speeds. Robotaxis are still expensive. Autonomous trucking is closer to working economics. ADAS shipped at scale in China while Western coverage was distracted. eVTOL is real but supply-constrained. The city layer is quietly the most profitable place to deploy AI today.

If you are sizing an AI roadmap for a fleet, an OEM program, or a city mobility platform and want a frank conversation about what is real and what is theatre, get in touch.