Waymo Miles Carry Passengers 54% of the Time, and the Source Sells the Fix

California data cited by The Robot Report shows that 54% of the 86 million miles its vehicles drove through 2025 carried a passenger. The other 46% went to repositioning, charging, cleaning and returning to depots, and that ratio is the argument in a September 26 piece on The Robot Report titled "From 14 cities to 15,000." Its author, Dan Keene, is the chief operating officer of a startup that sells the fix.

That does not make the figure wrong. It does mean the piece is a pitch, and the numbers need checking against what else is on the record.

The 54% figure and who is reporting it

The Robot Report piece says Waymo operates in 14 cities, has logged more than 220 million autonomous miles and provides more than 500,000 paid rides a week. Uber, by comparison, operates in more than 15,000 cities and completed 3.9 billion trips last quarter. Divide: 3.9 billion trips over 13 weeks is about 300 million a week, so Uber handles Waymo's weekly volume roughly every 17 minutes, which is the article's own figure and checks out.

TechCrunch's June 26 report identifies Keene as co-founder and COO of Aseon Labs, a Redwood City startup that raised a $10 million seed round led by Crane Venture Partners. Aseon's CEO is George Kalligeros, formerly a mechanical design engineer at Bentley Motors and Tesla. TechCrunch also reports that Aseon had no signed contracts at that point. The Robot Report article names Aseon as a company that can deploy its charging, cleaning and inspection stations "within a day."

The denominator matters here. The 54% figure covers California miles through 2025, not the whole 220 million, and I could not find the underlying filing in what I fetched. Treat it as the article's claim about a regulator's data, not a measured current utilisation rate.

Two hours a vehicle, and the grid queue behind it

The article puts a depot reset at about two hours per vehicle. At that rate a single bay turns roughly 12 cars in a 24-hour day, before any queueing. The scarcer input is the connection: the piece lists 6 months to energise a charging site, 1.9 years for a new circuit, 2.8 years for a substation upgrade and 8.9 years for a new substation. Those are the article's figures and it does not cite a utility for them in the text I read.

1. Energise a charging site0.5 years
2. New circuit1.9 years
3. Substation upgrade2.8 years
4. New substation8.9 years

Source: The Robot Report, Sept. 26, 2026, accessed 2026-09-30

Grid capacity is a real constraint on anything that plugs in at scale; Vermont's 110 MW virtual power plant shows one utility trying to add capacity from batteries instead of substations. A robotaxi depot draws far more power than a house, and years-long connection queues are the constraint that pods running on generators, as TechCrunch describes Aseon's, are meant to sidestep.

Fleet size against depot count

The Robotaxi Report newsletter's March 12 tally, by Jaan Juurikas, puts Waymo at about 3,000 vehicles, Baidu's Apollo Go at 1,000 to 2,000, Pony AI at 1,159 at the end of 2025, WeRide at 1,023 as of January 12, 2026, and Tesla at 392 or more. Paid rides globally ran near 1 million a week. Those fleets are small enough that a handful of depots per city covers them, which is the point where idle miles are a rounding error against engineering effort on the driving stack.

Operator	Fleet (as of date)	Source date
Waymo	about 3,000	March 12, 2026
Baidu Apollo Go	1,000 to 2,000	March 12, 2026
Pony AI	1,159	end of 2025
WeRide	1,023	January 12, 2026
Tesla	392 or more	March 12, 2026

The article's case is about the next order of magnitude. At 15,000 cities the depot count is a construction programme, and years per depot multiplied across thousands of depots is the bottleneck it describes.

What the pitch leaves out

Three gaps. First, a pod that resets a car in less than a depot does still has to be built in volume; Aseon planned five prototypes with its seed money and a team growing from six to roughly twelve. Second, the two-hour reset is one operator's average and no source shows what a pod achieves. Third, deadhead miles are only part of the 46%: some of that is repositioning to demand, which a pod does not remove.

Keene's argument, as TechCrunch quotes it, is that the vehicle has to be "in continuous operation during the entirety of the demand curve of the day" to match ride-hailing economics. That is plausible; it is also the sentence a supplier of utilisation would write.

What would change my read is a measured number: Waymo or another operator publishing passenger-mile share for 2026, or a pod pilot with a signed customer reporting reset time and cost per reset. Until then the 54% is a headline about a startup's market as much as about robotaxis.

For scale of the silicon side, see Sima.ai's $1.45 billion round for the physical-A.I. money pointed at vehicles and robots.