Robotaxi Utilization Is the Real Constraint on China's Liste

Robotaxi utilization has become the number that decides whether autonomous fleets ever pay for themselves, and on that measure China's listed operators remain well behind Alphabet's Waymo. Bloomberg Intelligence expects Pony AI and WeRide to record sustained operating losses through 2028 because their domestic fleets carry too few paying passengers per vehicle to cover what it costs to keep those vehicles running.

The assessment, published this week, turns on an amortization question. Level-4 hardware, remote assistance staff and high-definition mapping are fixed costs that only pay back when each car logs enough paid miles. Waymo clears that bar with robotaxi utilization above 58%. Pony AI and WeRide, both listed in the United States, do not.

That single ratio explains more about the competitive standing of the two fleets than any comparison of sensor suites, city counts or press coverage.

Why Robotaxi Utilization Decides the Race

Waymo's scale shows what high utilization looks like when a fleet matures. The Alphabet unit runs roughly 4,000 robotaxis and averages 500,000 paid rides a week. Its Texas fleet grew 49% in three weeks to reach 1,102 vehicles. The company also widened who may ride: Nashville became the second market, after Phoenix, where riders aged 13 to 17 can book a robotaxi on their own.

Volume of that kind spreads fixed costs across a much larger base. A car completing more paid trips each day repays its sensors, its onboard compute and its share of the remote-assistance desk faster, and it does so without discounting the fare.

Vehicle cost is the second lever. Waymo is on track to take delivery of 5,000 Zeekr vehicles by the end of the year and is ramping production of its Ojai platform in Arizona. A cheaper car lowers the fixed cost each paid mile has to carry, which lowers the utilization threshold a fleet needs before it breaks even. Chinese operators hold a structural advantage on hardware cost, but cheaper hardware only widens the margin on a trip that is already being taken.

China's Operators Add Projects Faster Than They Add Revenue

Pony AI, WeRide and Baidu's Apollo Go have moved more projects from testing into some stage of commercialization than their American counterparts, according to a BloombergNEF analysis. Project count is not revenue. Every added city brings mapping, depot and remote-assistance overhead before it brings enough riders to cover that overhead.

The cost stack that utilization has to absorb is long:

Level-4 sensor and compute hardware on every vehicle
Remote assistance staff who monitor cars and intervene when software cannot
HD mapping, refreshed as streets change
Depot space, cleaning and sensor maintenance
Insurance and compliance work tied to local permits

Each line is fixed or nearly fixed. None shrinks when a car finishes a shift with empty seats. The distance between Waymo's utilization rate and the rates Bloomberg Intelligence attributes to Pony AI and WeRide therefore shows up directly in gross margin, and the losses projected through 2028 are the arithmetic result.

The remote-assistance ratio is the clearest example of why volume matters. If one supervisor can oversee a fixed number of vehicles, that cost is identical for a car running twenty paid trips a day and one running five. Utilization converts a fixed supervisory cost into a per-ride number, and per-ride cost is what a fare has to beat.

Depreciation behaves the same way. Level-4 vehicles carry sensor and compute packages that lift the purchase price well above a conventional car, and that premium is written down over the vehicle's service life regardless of how many fares it collects. A fleet running below Waymo's utilization absorbs the same depreciation over a smaller revenue base.

Regulation Drew the Headlines; the Economics Did Not Move

China's permit freeze drew far more attention than the utilization data. Regulators stopped issuing new autonomous vehicle licenses after more than 100 Apollo Go vehicles halted at once on Wuhan streets on 31 March, stranding riders. The pause was presented as an industry-wide safety review, and permits began flowing again by July.

The freeze cost the sector months of expansion. It did not change the underlying economics. An operator that regains its permits still has to fill seats at close to Waymo's rate before per-mile revenue exceeds per-mile cost, and the suspension added a second drag by keeping vehicles and capital idle through the review.

China's policy environment carries its own brake. The country has the world's largest workforce and a recognized lead in autonomous driving, and those two facts pull in opposite directions: faster robotaxi deployment displaces driving jobs that Beijing has an interest in protecting. That tension shapes how quickly permits and city approvals arrive, which in turn shapes how quickly a fleet can reach the density that utilization requires.

The Trade-Offs Operators Face

Cutting fares to fill seats raises robotaxi utilization on paper while pushing break-even further away, because the fixed costs do not move. Rides per car can climb even as revenue per car falls, which is the trap the metric can hide.

Shrinking the service area is the second option, and the cheapest to execute. Geofencing already confines robotaxis to mapped roads, so tightening a boundary concentrates demand on streets where riders already are. The same boundary caps the addressable market and leaves the door-to-door promise of a taxi unmet.

Adding vehicles is the third route and the most expensive. A larger fleet lifts total rides and can improve dispatch efficiency, but it multiplies the fixed-cost base before demand exists to absorb it. Waymo can fund that bet because existing utilization covers the marginal car. Operators below that threshold cannot.

Fleet growth rates make the gap harder to close. Waymo added roughly 360 vehicles in Texas in three weeks, a pace that compounds its utilization advantage because each new car joins a network that already has demand. A Chinese operator scaling at a similar rate would raise its fixed-cost base first and its ridership second, which lengthens the period before the two lines cross.

Waymo's own recent decisions show the trade-off in reverse. The company pulled its robotaxis off freeways and has identified at least 13 instances of its vehicles entering highway sections closed for construction. Freeway trips are long, which lifts miles per vehicle, and they tend to be higher-value. Withdrawing from them protects safety and removes some of the easiest utilization the fleet had.

International expansion brings a different problem. Riders in Croatia, Singapore, Dubai, Abu Dhabi and Riyadh can already hail self-driving cabs, with London under consideration, but every new market opens at low density. A fleet needs local demand concentrated enough to keep cars busy, and early deployments rarely have it.

Waymo's position also rests on Alphabet's balance sheet. A fleet operator that can absorb losses while utilization climbs has time to reach density; one that depends on public markets for its next funding round does not. Pony AI and WeRide are both publicly traded, so their losses are visible every quarter and their expansion is bounded by what investors will finance.

Why This Matters

The robotaxi utilization gap reframes the China-versus-Waymo contest. Licensing, safety reviews and fleet counts dominate headlines, but the variable that decides which companies survive the next three years is how many paid seats each car fills per day. For investors holding Pony AI, WeRide or Baidu exposure, the metric to track is rides per vehicle rather than new city announcements, and Waymo's benchmark above 58% is the number every rival now gets measured against.

Bloomberg Intelligence's projection through 2028 sets a concrete horizon. Operators that close the gap will be able to fund the next fleet without returning to capital markets; the ones that cannot will keep posting losses while their vehicles wait for riders.

Photo by Sam on Unsplash

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.