Will AVs Replace Every
Car on the Road?
Waymo is completing 450,000 rides a week. Uber has announced 25 or more AV partnerships in 18 months. Nvidia's CEO says everything that moves will eventually be autonomous. The technology is accelerating. But will AVs eventually replace every car on the road? The data tells a more nuanced — and still compelling — story.
On March 16, 2026, Nvidia CEO Jensen Huang stood at GTC in San Jose and said it plainly: "The autonomous vehicle revolution is here — the first multitrillion-dollar robotics industry. Everything that moves will eventually be autonomous." That same week, Uber and Rivian announced a $1.25 billion partnership to deploy 50,000 autonomous R2 robotaxis across 25 cities. A few days earlier, Uber, Wayve, and Nissan announced a robotaxi launch in Tokyo before year-end. The pace of news is no longer incremental.
But there is a difference between "the AV revolution is real" and "AVs will replace every car on every road." The first is now supported by substantial evidence. The second is a considerably more complex proposition — and the gap between them is where the most interesting questions live.
The technology is no longer the primary bottleneck. The incident frameworks, liability law, public trust, cold-weather performance, and regulatory infrastructure are. Each one is a genuine constraint that will determine the pace — and the ceiling — of AV adoption.
The case for optimism is real and data-backed
The AV story has been littered with missed deadlines and spectacular failures — Cruise's suspension after a San Francisco incident, Tesla's robotaxi promises still largely unrealised, Uber's own costly $2 billion retreat from self-driving in 2020. But a different story has been quietly emerging in the operational data.
Waymo, the clearest bellwether, more than tripled its annual ride volume in 2025 alone, reaching 15 million trips. Across 127 million fully autonomous miles, it reported a 90% reduction in serious injury crashes compared to human drivers. That is not a press release statistic. It is a statistical argument that the technology can already outperform us at the thing we assumed we did better.
Goldman Sachs projects robotaxi costs falling from over $3 per mile today to approximately $1 per mile by 2040. At that price point, the economics of personal car ownership in cities fundamentally shift. A car sitting idle 95% of the time becomes increasingly difficult to justify when a safe, reliable autonomous alternative costs less than a coffee per mile.
"If 2024 was the year AVs proved they could work, 2025 was the year they proved they could scale. 2026 is the year they are going global."
ITquette Autonomous Lab analysis — March 2026The Uber ecosystem is particularly instructive. Rather than competing with AV technology companies, Uber has become the platform that every AV company needs to reach riders. Waymo, Wayve, Rivian, Zoox, Nvidia, Avride, WeRide — all are either on the Uber platform or in discussions to join it. That is the architecture of a mature industry forming around a dominant distribution layer, not a collection of competing moonshots.
But "replace every car" is a different question entirely
Total replacement — every privately owned, human-driven vehicle removed from global roads — is a far more radical proposition than "robotaxis are a growing and profitable business." Even the most bullish analysts stop well short of claiming full replacement in any near-term window.
Goldman Sachs' best-case scenario has AVs accounting for roughly 60% of new light vehicle sales globally by 2040. That means 40% of new sales in the optimistic case are still non-autonomous. And the existing global fleet of approximately 1.5 billion human-driven vehicles turns over slowly — typically 12 to 15 years per vehicle. Full fleet replacement, at that pace, would not complete until well past 2060.
There are six genuine structural barriers that even the most sophisticated AV companies have not yet fully solved:
Liability and legal frameworks
When an AV injures someone, who pays — the manufacturer, the software developer, the fleet operator, or the vehicle owner? Existing tort law was written for human negligence, not algorithmic decision-making. Courts in 2025 and 2026 are still actively drawing the lines, and a clear federal liability framework in the US remains unresolved.
The edge case problem
AVs excel in known environments. They struggle with genuinely novel scenarios — a child chasing a ball between parked cars, a downed power line across a lane, hand signals from a construction worker. Nvidia's Alpamayo 1.5 model is attacking this problem through language-prompted scenario generation, but it remains one of the deepest unsolved challenges in the field.
Weather and infrastructure dependency
Waymo's earliest markets were Phoenix, San Francisco, and LA — sunny, well-mapped, and relatively predictable. Its 2026 expansion into Detroit, Washington DC, and Philadelphia marks its first serious cold-weather deployments. Snow, ice, degraded lane markings, and poor infrastructure remain meaningful technical barriers, particularly in emerging markets.
Public trust has not arrived yet
Only 25% of consumers say they are comfortable riding in an AV today. A full 41% cite lack of trust in the technology as their primary barrier, with 44% unconvinced AVs are safer than human drivers — despite the mounting statistical evidence to the contrary. Trust is built in years, not product launches.
Regulatory patchwork
AV regulation varies dramatically by state in the US and by country globally. Japan's receptiveness to AVs is an outlier, not a rule. The markets where AVs could have the greatest impact — dense, chaotic, under-mapped cities across Africa, Southeast Asia, and South Asia — are also the markets where current AV technology and regulatory frameworks struggle most.
Cybersecurity and network risk
A car is a software platform on wheels. Over 70% of consumers in recent surveys cite hacking risk as a concern. As AVs become networked at fleet scale, a fleet-wide software compromise creates risks that have no precedent in traditional vehicle safety — and the incident management frameworks to handle that scenario are still nascent.
The milestones that will tell us whether the optimists are right
Rather than guessing a date, the more useful question is: what would have to be true for AVs to genuinely displace personal car ownership? Here are the milestones that matter.
The proof of concept. Fully autonomous, commercially operating, no safety driver. The first time paying passengers rode in a vehicle with no human at the wheel.
Over 450K weekly rides, 15M annual trips. The "can this scale?" question answered — at least in favourable urban environments with good weather and well-mapped roads.
Waymo entering Detroit, DC, London. Uber, Wayve, and Nissan launching in Tokyo. Nvidia and Uber announcing 28 cities across four continents. If AVs perform in snow, ice, and non-American traffic cultures, the geographic constraint effectively disappears.
Goldman Sachs forecasts robotaxi cost falling below $1 per mile by 2030 versus roughly $2 or more for current rideshare. If this holds, the economic case for personal car ownership in cities weakens substantially.
Without a clear federal compensation framework, liability uncertainty will continue to deter deployment at scale. A US federal AV liability bill would be one of the most important enabling milestones in the entire AV story.
Goldman Sachs and S&P Global project up to 10% of new vehicle sales reaching Level 3 by 2030. China is expected to lead, potentially reaching 90% of new sales at L3 or above by 2040.
Goldman's bull case: roughly 60% of global new car sales autonomous by 2040. But this is new sales — the existing fleet of human-driven cars will still constitute the majority on roads well beyond 2050.
The verdict: optimism yes, complete replacement not so fast
The case for long-run optimism is strong and backed by data. AVs are already statistically safer than humans in known environments. Costs are falling on a trajectory that makes them competitive with car ownership in cities by the 2030s. The deployment model pioneered in Phoenix is going global. If you are investing, regulating, or building in the transport space, treating AVs as inevitable in some form is the only defensible position.
But "will replace all cars on the road" conflates the trajectory with the destination. The existing vehicle fleet turns over slowly. Trust is built in a generation, not a product cycle. Liability law is unresolved. The most consequential markets — the chaotic, under-mapped cities of the Global South — remain genuinely unsolved. And the incident management and safety governance infrastructure that will determine whether AV operators keep their operating licences as scrutiny increases is still being built.
The honest answer is: AVs will replace most new vehicle sales in urban markets within 20 to 25 years. Every car on every road? That is a 2060s conversation at the earliest — and it depends on solving problems that are operational and regulatory, not just technological.
The five things to watch as litmus tests: Tokyo's operational reality by year-end 2026. Waymo's winter performance data from Detroit and DC. US federal liability legislation. Consumer trust trend lines in the 2027 S&P Global survey. And whether robotaxi cost per mile actually hits $1 by 2030. Those five data points will tell us more about the pace of full replacement than any number of press releases.
Related insights
Incidents are sequences, not moments
Most operational breakdowns happen after the event. Trust improves when the lifecycle is consistent and auditable.
Trust is cumulative for cities and regulators
Stakeholders ask what changed afterward and how recurrence is prevented. Evidence readiness makes those conversations calmer.
UL 4600 is a safety argument that must be maintained
Safety maturity shows up in governance and traceability — not just in initial testing results.
Follow the research. Reach out if it resonates.
Published openly through this site and LinkedIn. If something connects with a challenge you are working through, we welcome the conversation.