Autonomous vehicle commercial deployment: Autonomous vehicles are shifting from demos toward commercial deployment across passenger and freight markets

Overview

This thematic field-trip report examines how autonomous driving is progressing from technical development toward commercial use in robotaxis, passenger vehicles and freight. BofA concludes that the opportunity is broad and multi-player, but that safety validation, vehicle supply and fleet operations will determine the pace of scale.

Core views

BofA's central conclusion from meetings, demonstrations and ride-alongs with more than 30 executives across 15 companies is that autonomous vehicles are moving from development toward commercial deployment in both passenger and freight markets. The institution does not view autonomy as winner-take-all: companies are pursuing end-to-end AI, modular systems with sensor redundancy, vertically integrated fleets, partnerships and software-licensing models. The key debate has shifted from whether autonomy can work to whether operators can validate safety, secure autonomous-ready vehicles and build the charging, maintenance, mapping, remote-operations and regulatory capabilities needed to scale. AI and foundation models are the common technological driver. Tesla and Wayve favor end-to-end architectures trained on large real-world data sets, while Aurora, Zoox, Nuro, Kodiak and Rivian combine AI with redundant sensors, simulation and safety layers. Wayve's AV2.0 strategy seeks a generalized driving model across vehicle types, locations and sensor configurations; it cited a 2025 test in 500 cities where vehicles drove in many locations with little or no local training data. Wayve intends to license its software to OEMs, initially focusing on L2++ driver assistance before robotaxis, with Nissan Leaf deployment expected from 2027 and a supervised Uber robotaxi service in London launched in September 2026 with 15 vehicles. The report sees improving economics as a major enabler. Falling sensor costs, improving model capability and driver shortages support the case for automation. Tesla expects Cybercab all-in cost at scale of about $0.30 per mile, versus about $0.50 for a Model Y fleet vehicle and $2-3 per mile for Uber or Lyft. Tesla cited 125,000 units of installed capacity and battery and drive-unit durability of up to 500,000 miles. Its unsupervised miles rose from about 380,000 cumulatively at the earnings call to more than 1 million the prior week, with a target for double-digit percentage week-over-week growth through year-end. The report identifies software validation and the data feedback loop around seven technology tracks in version 15 as the near-term bottleneck, while Tesla prioritizes customer experience and fleet building over near-term robotaxi profitability. Freight autonomy offers a separate productivity case. Aurora estimated Driver-as-a-Service costs of about $0.85 per mile versus roughly $1.30 per mile for human-driver wages and benefits, before indirect labor costs, and highlighted potential operation of about 20 hours per day versus 8-9 productive driving hours for human drivers. Aurora launched driver-out operations in April 2025 and currently operates 25 commercial-freight trucks; it expects roughly 200 trucks exiting 2026 before larger subsequent fleet expansion as OEM supply improves. It estimates a $1 trillion long-term trucking addressable market, but says vehicle availability and scalable operations are now more important constraints than technology readiness. Kodiak operates 35 driverless trucks in the Permian Basin for Atlas Energy Solutions and reported driver-out miles increasing from 91% to 93%; it emphasizes cameras, radar and LiDAR, probabilistic safety assessment, simulation and real-world validation. The report also highlights different commercial models. Zoox is vertically integrating vehicle design, autonomy software, manufacturing and fleet operations around a purpose-built bidirectional robotaxi, with initial revenue in Las Vegas and operations in San Francisco and Las Vegas. Lucid, Nuro and Uber represent an ecosystem model: Uber has committed to buy at least 35,000 autonomous vehicles, while Nuro estimates its hardware stack at about $10,000 per vehicle and seeks recurring software-licensing revenue while retaining autonomous-driver product liability. Rivian plans to move from driver assistance toward autonomous ride-hailing, using multimodal sensing, in-house AI, custom chips delivering 1,600 TOPS and an Uber relationship targeting up to 50,000 autonomous vehicles. Gatik focuses on middle-mile logistics, expects about 100 driverless vehicles by year-end, cited more than $600 million of contracted backlog, and described typical contracts producing roughly $200,000 of annual revenue per truck. Beyond vehicles, BofA argues that autonomous driving is one of the largest real-world training environments for physical AI. Wayve is exploring manufacturing, logistics and humanoid robotics through its embodied-AI lab, and Tesla is leveraging autonomy infrastructure for Optimus. However, the report records differing views on transferability: Rivian cautioned that autonomous-driving models may not readily transfer to humanoid robotics because commercial environments and required data differ. The broader ecosystem discussion stressed that mapping, charging, maintenance, parking, fleet operations and remote supervision may become important value pools as AV utilisation rises. Personally owned autonomous vehicles may also converge with fleet robotaxis, potentially expanding demand beyond replacement of incumbent transport models.

Analysis framework

The report synthesizes management meetings, site tours and ride-along demonstrations across the autonomous-mobility ecosystem. It compares technical architectures, safety-validation approaches, deployment progress, unit economics, utilisation, vehicle supply, operating models and infrastructure requirements across passenger robotaxis, consumer ADAS and autonomous trucking.

Methodology notes

• Industry AnalysisSupply-demand framework

  Autonomous-mobility economics assessed through driver shortages, vehicle availability, operating costs and utilisation.

  The report links worsening driver supply constraints and longer autonomous operating hours to lower cost per mile and potentially higher transport demand.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Autonomous-mobility ecosystem analysis.

  The report connects AI and vehicle platforms with mapping, charging, manufacturing, fleet operations, logistics and ride-hailing networks needed for commercial deployment.

• Other

  Safety validation using simulation, real-world testing, redundancy and probabilistic risk assessment.

  Companies use these methods to demonstrate that autonomous systems can operate safely enough for scaled deployment; the report treats validation as a key gating factor.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Wayve

  Autonomy-software provider pursuing an OEM licensing model.

  Strengths

  End-to-end generalized AI architecture, hardware flexibility and asset-light licensing strategy.

  Weaknesses

  Robotaxis remain a longer-term opportunity relative to its near-term ADAS focus.

  Comparison

  Contrasts its single end-to-end neural network with legacy modular, rules-based AV stacks.

  Risks

  OEM adoption, deployment validation and scaling from assisted driving to robotaxis.

• Aurora

  Autonomous-trucking provider using a Driver-as-a-Service model.

  Strengths

  Commercial driver-out operations, OEM relationships and a lower reported cost-per-mile proposition.

  Weaknesses

  Fleet scale depends on autonomous-ready vehicle availability.

  Comparison

  Emphasizes verifiable AI and safety constraints rather than purely end-to-end autonomy.

  Risks

  Operational scaling, OEM vehicle supply and safety validation.

• Tesla

  Robotaxi and consumer-autonomy platform pursuing the Cybercab model.

  Strengths

  Large real-world data feedback loop, low targeted Cybercab cost and durable vehicle components.

  Weaknesses

  Current robotaxi pricing is not representative of the intended scaled model.

  Comparison

  Targets materially lower cost per mile than Model Y fleet vehicles and Uber/Lyft.

  Risks

  Software validation and scaling the unsupervised-mile data loop before fleet expansion.

• Gatik

  Commercial autonomous middle-mile logistics operator.

  Strengths

  Existing driverless operations, contracted backlog, take-or-pay contracts and fixed-fee revenue model.

  Weaknesses

  Focused on middle-mile logistics rather than the broader long-haul market.

  Comparison

  Uses fixed-fee transportation contracts rather than mileage-based pricing.

  Risks

  Route setup, operational expansion and manufacturing-partner execution.

Key data

• Field-trip participation>30 executives across 15 companiesBofA autonomous-vehicle field trip
• Wayve generalisation test500 cities globally2025 test; vehicles operated in many locations with little or no prior local training data
• Tesla Cybercab cost target~$0.30 per mileExpected all-in cost at scale
• Aurora Driver-as-a-Service cost~$0.85 per mileVersus approximately $1.30 per mile for human-driver wages and benefits
• Autonomous-truck operating time~20 hours per dayVersus roughly 8-9 productive driving hours for human drivers
• Kodiak driverless fleet35 trucksOperating in the Permian Basin for Atlas Energy Solutions
• Tesla unsupervised miles>1 million cumulative milesUp from about 380,000 at the earnings call
• Gatik contracted backlog>$600 millionSupported by long-term take-or-pay customer agreements

Impact & implications

BofA sees autonomy creating opportunities across passenger mobility, trucking, logistics, vehicle platforms, AI software and fleet infrastructure. Lower cost per mile and higher utilisation could expand mobility demand, but commercial scale depends on safety proof, regulatory execution, vehicle production and operational infrastructure rather than software progress alone.