Waymo's Robotaxi Growth Is Capped by the Grid, Not Software

Waymo operated in three U.S. cities in September 2024. Two years later it operates in 15, running roughly 4,000 robotaxis that together deliver about 500,000 paid rides a week, according to a Sept. 24 TechCrunch analysis of Waymo's fleet data. California and Texas together account for about 80 percent of that fleet, and the Texas count alone grew 49 percent in the three weeks before TechCrunch's tally was taken. The obvious question, given numbers like that, is what stops the next fivefold jump. The answer that gets the least attention is not the software, the sensors or a regulatory approval. It is the electrical grid underneath the depot.

That is the argument Dan Keene, co-founder of Aseon Labs, a company that sells software for speeding up grid interconnection, made in a September Robot Report column built around California utility timelines. Keene, whose company has a direct commercial interest in shortening these timelines, writes that energizing a charging site that already has spare capacity at the pole takes about 6 months. Adding a new electrical circuit to a site takes 1.9 years. Upgrading a substation takes 2.8 years. Building a new substation from nothing takes 8.9 years. Those are interconnection and construction schedules, not chip shortages or a shortfall in AI capability, and they apply per site: a fleet operator cannot make them move faster by writing better code.

The reset clock and the empty-mile problem

Keene's other number worth sitting with is the roughly 2 hours it takes today to cycle a robotaxi through cleaning, charging and inspection between shifts. Multiply that by a fleet in the thousands and the constraint stops being about how many vehicles a company can buy and becomes about how many charging bays, technicians and grid-connected amps sit behind one depot. About a third of Waymo's Texas fleet runs Ojai minivans, and the company expects roughly 5,100 more of them by the end of the year, TechCrunch reported, which is 5,100 more vehicles that need a bay, a charger and a technician on the same clock.

Waymo's own reporting touches a related cost. Through June 2026, the company had logged 271.3 million rider-only miles, according to Waymo's safety-impact page, and Keene's column cites California data showing 46 percent of an earlier 86-million-mile tranche carried no paying passenger at all. Empty miles still draw power and still need the same charging bays as paid ones, which means the capacity problem is bigger than the ride-volume numbers alone suggest.

Waymo footprint	Sept. 2024	Sept. 2026
U.S. cities served	3	15
Fleet size	not disclosed	about 4,000
Paid rides per week	not disclosed	about 500,000

None of this is disqualifying for the business. It is a capacity-planning problem with a knowable schedule, and Keene argues the operators best positioned are the ones treating grid interconnection as a procurement line item years ahead of a launch rather than a surprise discovered after vehicles arrive. Uber, which Keene's column notes handles Waymo's entire weekly ride volume roughly every 17 minutes across more than 15,000 cities worldwide, is the scale robotaxi operators are being measured against, fairly or not.

What the safety numbers do and do not say

The case for scaling faster rests partly on safety data Waymo publishes and updates itself on a schedule tied to federal reporting rules. Through June 2026, Waymo says its vehicles were involved in 95 percent fewer crashes causing serious injury or worse, and 93 percent fewer pedestrian-injury crashes, than human drivers over comparable mileage in the same operating areas. Those figures cover five metro areas where Waymo reports under the National Highway Traffic Safety Administration's Standing General Order: Los Angeles, the San Francisco Bay Area, Phoenix, Austin and Atlanta. They are Waymo's own count, not an independent audit, and the company does not publish the raw crash data behind the percentages on that page.

The 15-city figure TechCrunch reported and the five-city safety-reporting figure are not the same list; several of the newer markets are too recent to have accumulated enough miles for a comparable safety readout yet. The tension between an autonomous fleet grading its own performance and the need for independent verification is the same one that pushed OpenAI's Sam Altman and Anthropic's Dario Amodei to brief the United Nations Security Council in September on why AI capability was outrunning any single government's ability to oversee it. A fleet of vehicles operating without a driver, like the AI agent swarms detailed in September research, raises the same question of who checks the checker.

What would change this read

What would change this read is a genuine count of energized megawatts at Waymo's depots, not announced fleet size. Announced vehicle counts are a demand signal; energized capacity is the binding constraint, and right now only one side of that equation is public. Until an operator or a utility publishes the other side, the honest way to read a robotaxi fleet count is as a number capped by a slower one sitting a level below it: not how many cars a company can order, but how many megawatts a utility can connect.

1. Energize existing capacity0.5 years
2. Add new circuit1.9 years
3. Upgrade substation2.8 years
4. Build new substation8.9 years

Source: Dan Keene, Aseon Labs, analysis of California utility timelines published in The Robot Report, accessed 2026-09-26