How AVs will go mainstream

2025 was an important year of progress for autonomous vehicles (AVs) - from Waymo expanding into new cities, to more companies going driverless (Zoox, Tesla), to the first commercial deployment of autonomous trucks (Aurora). In cities like San Francisco, driverless cars became a part of everyday life, with Waymo even surpassing Lyft in market share by gross bookings.

However, we are still in the early innings of the AV story. In this post, I consider what it will take for AVs to reach mainstream scale, with a focus on the US. I define this milestone as having a viable presence in most major metropolitan areas. AV adoption will not just be driven by general autonomy improvements, but will be determined by three critical unlocks:

1. technical breakthroughs around highway and winter weather
2. operational scaling enablement
3. public trust & acceptance

These unlocks build upon each other. Technical progress without operational scale won’t drive adoption. Operational scale without trust won’t endure. And trust without reliability won’t convert into utilization.

I believe we will achieve this level of mainstream adoption in the next 3-5 years, based on recent milestones and ongoing industry progress. While there are real risks that could slow this timeline (safety incidents, regulatory setbacks, capital constraints), steady expansion, increasing regulatory familiarity, and continued capital deployment suggest the trajectory should broadly remain intact.

Note: This piece is largely based on my personal experience launching driverless ridehail services in the US at Cruise.

Technical advancements

The last three years have shown that L4 AV systems have matured enough to be deployed in urban environments, with driverless ridehail services now operating in multiple US cities. But there are two key technical challenges that need to be overcome to achieve mainstream adoption: highway driving and winter weather management.

Highway driving is important for two reasons. First, it will unlock airport trips in most metro areas, which account for 15% of ridehail bookings (which Uber shared in its 2024 annual report). This is a key use case that allows riders to see AVs as a reliable option, while increasing rider retention. Second, enabling highways will make AVs more viable outside of core downtown areas. In cities like Los Angeles and Atlanta, highways are essential to serving a majority of trips that riders want to take.

Winter weather driving is needed to reach many of the largest ridehail markets in the US - New York, Boston, Chicago, and Philadelphia. But can AV companies just deploy fleets during the non-winter months in these cities? There are two challenges with this approach:

• Rider retention: If a service isn't reliably available year-round, non-enthusiast riders will eventually stop considering them for everyday use, defaulting to services they expect will always be available. Being offline for multiple weeks or months will prevent AV services from becoming part of a rider's default transportation options.
• Fleet utilization: Take, for example, New York City, where a viable service would likely require at least 1,000 vehicles. Having these vehicles sit in a parking lot for several weeks in response to weather is economically infeasible. Moving these vehicles to warmer‑weather cities would also be costly and a logistical challenge.

Current status: In 2025 we saw initial highway deployments at Waymo, though it remains in limited access. This will likely be a slow, gradual rollout given the risk profile. Aurora's deployment of driverless trucks in Texas also indicates good progress on this front.

The winter weather story is a bit different. Ultimately, it is both a software and hardware problem. On the software side, sensor-fusion systems will need to account for environmental changes (falling snow, ground color changes). We've already seen systems that can adapt to some inclement conditions like heavier rain and fog in San Francisco, so there should be a path forward. On the hardware side, new sensor stacks will be needed, including features like sensor self-cleaning (e.g., physical wipers, heating elements). Waymo's Gen 6 hardware system - which has entered initial testing and deployment - will be the first to include these capabilities.

Operational scaling

To support the scale definition mentioned above, tens of thousands of vehicles will need to be deployed. Achieving this will require both adequate manufacturing capabilities and support infrastructure.

Manufacturing at scale is difficult in the auto industry, even before considering the added complexity of autonomy. The first challenge is building the know-how. This was the core thesis of GM purchasing Cruise and is where Tesla has a big advantage. As others like Waymo and Zoox try to scale, they will need to learn on the job and/or partner with existing OEMs to overcome this hurdle. The other piece of this is AV vehicle cost. Reaching mainstream levels of scale will likely require total vehicle costs to drop to ~$100k, and we're not there yet.

Infrastructure to support operations at scale has its own challenges. The two biggest are parking spaces and charging stations. Finding parking for thousands of vehicles in a city is difficult. Operators will prefer larger structures that batch vehicles to optimize maintenance operations (e.g., cleaning, servicing) and covered lots to prevent vehicle damage. Even better would be structures that have large volumes of fast chargers to support the fleet.

Current status: The largest driverless fleets today are a couple thousand vehicles. Outside of Tesla and traditional OEMs, AV companies still need to prove they can build vehicles at scale. I expect companies to leverage existing manufacturing know-how to accelerate scaling (like Waymo is doing with the IONIQ 5 and purpose-built Ojai). However, I expect the ongoing bottleneck for deployments to be infrastructure buildouts given the limited requisite infrastructure that exists today. It takes 12-18+ months to get new projects in a city built out, from permitting to construction.

Public trust & acceptance

We can have the most technically sound autonomy systems with cheap cars easily deployed in cities, but none of this matters if communities don't accept them on their streets. I believe this is the long pole in the tent for AV adoption, and yet it often gets the least attention.

Regulator trust is needed to get AVs on the roads. While permitting requirements and structures vary by city and region, regulators will always have the ability to pull vehicles off the road. Building this trust starts with education on the technology and safety profiles. However, we'll need to go further to make communications consistent and trustworthy. I believe we need something akin to the NHTSA 5-star safety rating system for L4/5 autonomous systems. Similar to the benchmarks and evals that have become prominent with GenAI, we need a common set of driving evals.

Community trust is needed to keep AVs on the roads. While regulators might approve initial deployments, community sentiment and response will determine the longevity of the deployment. This isn't about people being comfortable getting into the vehicles (more on that below) but rather about being comfortable having them on the roads. Pedestrians and bicyclists must feel safe sharing the roads. Parents must feel comfortable with their young children playing in neighborhood streets with AVs deployed. First responders must trust the vehicles to behave appropriately in emergency situations. In this early stage of the industry, building trust won't be easy or "scalable"; it will be a process of deeper community engagement that will differ across cities.

Rider trust is important to drive utilization of fleets. This sits at the bottom of the trust funnel, a by-product of regulator and community trust. Education is also important at this level, but now from the perspective of the rider experience. For the time being, transparency around how the vehicle behaves (e.g., 3D visualizations of the surroundings, thought bubbles explaining key events on the road and vehicle intent) helps build familiarity and trust.

Current status: Regulator trust continues to grow, with more jurisdictions building frameworks for AV testing and deployment. However, this isn't a linear process, as we've seen recently with progress then pullback in New York. At the national level, we still lack a common framework to discuss safety profiles, which will be critical as more autonomy providers emerge and deployments span multiple regions. Community and rider trust is very strong in certain regions like San Francisco, but other regions still have a way to go. I think this is going to require patience and deep engagement.

While this is a slow climb, I'm optimistic we'll get there. A quick story - in Fall 2023 I was in Houston for the public launch of Cruise's service. While there, I was at a local community fair talking with residents about the service. An older lady walked up to me and said she would never get into an AV. And yet, after a few minutes of open and honest conversation with good intentions on both sides, she was willing to try a ride later that night. This exchange highlights the importance of open and deep engagement with all stakeholders - because that's the responsibility that comes with introducing such a pivotal technology to society.

Summary: Key gaps and timeline predictions

Technical

• Highway driving, unlocks airports and drives usage & retention
• Winter-weather management, driven by improvements to software and hardware
• Timeline prediction: 2 years given Waymo's phased rollout and Gen 6 hardware

Operational

• Scaled AV manufacturing, likely with support from existing OEMs
• Buildout of support infrastructure (parking, charging); seek support from local governments to accelerate timelines
• Timeline prediction: Manufacturing scale in 3 years (based on Waymo's trajectory); infrastructure buildout will match deployment timelines by starting now in key cities

Public Trust

• Ongoing regulatory buildout of frameworks to support AV testing and deployment
• Standardized framework for safety performance (like the NHTSA 5-star safety rating)
• Deep, local community engagement that meets residents where they are at; don't get fooled by acceptance in San Francisco
• Timeline prediction: 3-5 years; this will require sustained efforts and ultimately determine the pace of mainstream adoption