Nobody Has Built the Car Wash for Self-Driving Cars
Waymo can handle the driving. Cleaning, charging, and inspecting its vehicles still requires a trip across town, an industrial lot nobody can find, and years of permitting. That's the real bottleneck.
Key points
- Waymo's own California data shows only 54 percent of its 86 million reported miles through 2025 carried a paying passenger, with empty depot runs accounting for a large share of the gap.
- Getting a new electrical substation approved and built to power a charging depot takes an average of 8.9 years per site, according to figures cited by The Robot Report.
- Waymo currently operates in 14 cities; Uber, its largest distribution partner, operates in more than 15,000.
- Data centre developers, including firms like Microsoft and Amazon, are outbidding industrial buyers for the powered land that robotaxi depots need.
- Y Combinator-backed startup Aseon Labs says its modular servicing station arrives on a flatbed and can go live within a day, though the company is at an early stage.
Waymo works. The company has logged more than 220 million autonomous miles and gives roughly half a million paid rides every week. But the question that doesn't get asked loudly enough is simpler and more embarrassing: what happens when a passenger spills coffee on the back seat?
Right now, the answer is a 15-mile empty drive to a centralised depot, a two-hour reset, then back out. Every depot run is a paid ride that never happened.
Why does this matter to ordinary riders?
Empty miles and long reset times push up the cost of every fare and limit how quickly services expand. If you're waiting for cheap, widely available robotaxis, the cleaning bay is a bigger obstacle than the driving software.
The arithmetic is harsh. Getting a standard charging site energised in California takes six months on a fast track. A new power circuit averages 1.9 years. A substation upgrade runs 2.8 years. Building a new substation from scratch: 8.9 years. Per city. Uber is in more than 15,000 cities. At that pace, full coverage is measured in centuries.
Land is the other chokepoint. The industrial-zoned, grid-connected parcels that a depot needs are exactly what data centre developers want. AI companies with deep pockets are routinely outbidding fleet operators for powered plots in growth markets. When we reported on London's first Wayve robotaxis joining Uber on 3 September, the constraint wasn't the software; it was that only 15 vehicles were out there.
What does the industry look like now?
The self-driving business has split into specialists. Wayve and Mobileye sell the software brain. Car manufacturers, including Nissan and Volkswagen, install those brains in their own vehicles. Uber funnels riders to more than 30 autonomous partners. The result is dozens of different fleets running in the same cities, all needing charging, cleaning, and inspection several times a day, none sharing a depot.
Waymo solved this by doing everything itself early on, but it's since handed depot operations to third parties: Moove, Avomo, and Flexdrive now run its sites. The bespoke-depot model that worked across 14 cities simply doesn't scale.
A better comparison is shipping containers, which didn't make ships faster but made every port interchangeable, unlocking global trade. Robotaxis need something similar: a standard servicing unit that drops into a parking bay, connects to available power, and serves any vehicle on any network.
Aseon Labs, whose co-founder Dan Keene wrote the piece first published by The Robot Report, is betting on exactly that model: a robotic charging, cleaning, and inspection station that arrives on a flatbed. It's early, and the company won't be alone for long. Nuro was conspicuously absent from Keene's specialist list, which is worth watching as the field consolidates.
Cars can drive themselves. The pit still doesn't exist at scale, and that's the actual frontier.
