The autonomous vehicle industry is approaching a commercial inflection point in 2026 as new legislation and fresh regulatory frameworks remove barriers that have historically constrained large-scale deployment.
According to the “Autonomy 2.0” white paper from Stifel’s in-house market research team IRIS, cumulative Level 5 driverless mileage has reached 536.3 million miles with Waymo, WeRide and Apollo Go together responsible for more than 99 per cent of recorded driverless operations.
More than 1,500 robotaxis already operate commercially in the US, a figure that industry projections suggest could rise to 35,000 by 2030, alongside a parallel expansion to 25,000 autonomous trucks.
Mahaut Arnaud, Research Analyst at Stifel IRIS and co-author of the paper, points to several key developments behind a pivotal 2026.
Regulation reaches a critical juncture
In the United States, the Self Drive Act grants the National Highway Traffic Safety Administration exclusive authority over performance standards, eliminating the state-level patchwork that had constrained interstate autonomous trucking, large-scale fleet deployments and institutional investment.
“One of the major hurdles for companies so far was that they were limited to city environments,” Arnaud tells Zag Daily. “Now you can go across states and that means you can also operate on road types that are more easily drivable such as highways and across different vehicle categories including trucks.”
In Europe, the UN Global Technical Regulation on Automated Driving Systems published earlier this year introduces the first harmonised technical blueprint covering Level 4 and Level 5 autonomy and models its mandatory Safety Management System closely on aviation.
For Arnaud, the UN regulation is particularly significant because it removes the cost and complexity of navigating country-by-country approval processes: “It introduces a degree of simplicity that should make deployment considerably less costly for companies operating across borders.”
While the white paper concentrates on robotaxis and personal vehicles, Arnaud acknowledges this leaves out parts of the market that may scale faster. UK company Oxa, which builds systems for ports and logistics, exemplifies a segment she expects to move quickly because, in industrial and logistics settings, “the business model and the economic case are somewhat easier to defend”.
Hardware costs fall as AI architecture matures
Central to the paper’s investment thesis is the concept of the AI-Defined Vehicle, a vehicle architecture in which unified neural networks map sensor inputs directly to driving actions.
The paper describes this as the product of a broader shift in artificial intelligence from digital environments to the physical world, which it calls Physical AI. For years, a key limitation of autonomous systems was the lag between observation and decision-making. Arnaud describes the requirement as one for “very quick response and human-like thinking,” comparing the challenge to the way the human brain operates without any perceptible gap between decision and action.
The emergence of end-to-end learning architectures, which train across entire driving sequences rather than individual perception modules, is presented as the technology resolving that bottleneck.
Waymo’s evolution is offered as a case study, beginning with highly detailed HD maps, then moving to a mapless approach before integrating Gemini-based Physical AI, a trajectory the paper presents as indicative of where the industry is heading.
Hardware economics are also shifting in ways that underpin the paper’s commercial forecasts. Solid-state LiDAR costs are approaching the sub-$200 threshold, a level at which the sensor becomes economically viable for mass-market vehicles, and the paper considers the long-running debate about camera-only versus sensor-fusion architectures to be largely settled in favour of fusion.
“For many years there was debate about building autonomous cars on camera only because it’s less expensive,” Arnaud notes. “It’s no longer the case and it doesn’t work very well.”
Sensors are commoditising and vehicle architectures simplifying – including purpose-built designs without steering wheels or unnecessary components which Arnaud highlights as particularly significant for margin potential. Robotaxi operators have cited gross margin targets of 40 to 50 per cent, figures Arnaud considers credible at the per-vehicle contribution level in high-utilisation cities once hardware costs have stepped down, but considerably less so as a blended corporate margin while operators are still building out on subscale fleet.
Those projections are offered with explicit caveats. Arnaud is direct about the data constraints the team faced. “In many cases, data is not that reliable and the data that companies share is not always reliable either,” she says, pointing to recent controversies around Tesla’s Full Self-Driving figures as an illustration.
“Unless you have global standards in place, it’s almost impossible to have something homogeneous on which you can build reliable statistics.”
The paper’s market estimates are therefore built on modelled replacement ratios over time rather than observed deployment rhythms, and the analysts caution that many published total cost of ownership figures in the sector are company-specific and business model-specific to a degree that makes them difficult to apply broadly.
Software providers emerge as the value chain’s power centre
The question of who ultimately captures value in an autonomous ecosystem is one the paper addresses at length.
Arnaud argues that software, and specifically fleet management and data management systems, will sit at the top of the value hierarchy, as OEMs have repeatedly demonstrated they struggle to develop compelling autonomy software independently.
General Motors’ Cruise is cited as a cautionary example. “I think some of those have already chosen to rely on Google or the established platforms, and they’ve realised that doing it alone is no longer the most effective path,” Arnaud says. “Software providers will be setting the price and setting the terms of the relationship.”
Investors, she suggests, will focus on “niche players, flexible players, innovative players” – companies able to develop client-specific solutions and maintain long-term partnerships with OEMs, particularly given that many European OEMs are currently capital-constrained.
“One example of such a software player in Europe is Wayve,” Arnaud adds.
Ride-hailing platforms are also positioned as likely gatekeepers if robotaxis prove to be the first autonomous business model to achieve viability at scale.
The paper makes a guarded case for personal autonomous vehicle ownership eventually materialising but Arnaud declines to put a firm timeline on it, suggesting ten years is reasonable while noting that a single high-profile accident could set the sector back significantly.
She sees the economic and liability frameworks around personally-owned AVs as far less tractable than those around commercial fleets. Insurance, she notes, is among the most significant remaining hurdles and one where progress depends on robotaxi operators first accumulating the kind of operational data that can be credibly scrutinised.
Since the paper’s publication, Arnaud says the clearest validation has come from capital markets. Public AV companies that previously struggled to demonstrate commercial revenues have seen substantial stock price recovery over the past two to three quarters as deployment activity has made their operational progress more legible to investors. Aurora is cited as one example, with Arnaud describing its stock as having roughly doubled over the past year.
That shift in sentiment, she believes, reflects a market beginning to price in what scaled deployment actually looks like, with real operational progress now visible.
