The New York Times counted Waymo’s edge cases. The more revealing number is buried in Alphabet’s segment disclosure, where revenue went backwards while the quarterly loss hit $1.8 billion.
Waymo runs close to 4,000 vehicles across 15 US metro areas and books roughly 500,000 paid trips a week. That trip figure has held steady across four public disclosures since late March, while the fleet kept growing and the city count went from 10 to 15. Each vehicle now completes about 125 trips a week, down from 167 in May 2025. Alphabet’s Other Bets segment, which houses Waymo, reported $382 million of revenue against a $1.8 billion operating loss in the second quarter of 2026. Waymo is buying map coverage, not ride volume, and the emergency-response rules now moving through Washington and Sacramento would let officials switch that coverage off during the busiest hours of the year.
On the evening of July 4, several Waymos sat dead in San Francisco traffic with flat batteries. Tow trucks were dispatched. The tow trucks could not get through, because the streets were jammed, and the streets were jammed in part because the Waymos were sitting in them. Nobody cleared the vehicles for two hours.
That loop is the whole story. The failure blocked its own remedy, and no amount of software shipped from Mountain View could shorten the wait. What ended the incident was a truck, a driver, and a road.
What Happened
The New York Times published a survey of Waymo’s growing pile of “edge cases” on August 10, reported by Emmy Martin from San Francisco. The July 4 fireworks show drew more than 100,000 people to the waterfront. Waymos blocked intersections, some drivers sat behind them for over three hours, several robotaxis were towed after their batteries died, and at least two rolled over lit fireworks. One caught fire with a passenger inside.
Waymo had prepared. The company put an employee in San Francisco’s Emergency Operations Center, restricted where riders could hail cars, and staffed a team that could take manual control of stuck vehicles. Reports of stalled Waymos started reaching the center more than two hours before the 9:30 p.m. show began.
Carnegie Mellon’s Philip Koopman, who studies autonomous vehicles, told the Times that the technology is “prone to overconfidence and spectacularly stupid failure” when it meets something new, and that Waymo’s fix has been to hand the car to a remote human rather than solve the underlying problem. Waymo spokeswoman Sandy Karp said the recalls were proactive and that the rate of edge cases is falling as the company grows.
Three federal recalls have landed since December, including one for cars driving into flooded roads. Waymo issued three in the previous 22 months.
The Backstory
Waymo’s expansion pace changed in 2025 and never changed back.
In May 2025, the company operated more than 1,500 vehicles across San Francisco, Los Angeles, Phoenix and Austin, running more than 250,000 paid trips a week. By late March 2026 it had reached 500,000 weekly trips across 10 cities. In February, it raised $16 billion at a $126 billion valuation, the largest round any autonomous vehicle company has ever closed. Co-CEO Tekedra Mawakana set a target of one million weekly rides by the end of the year.
Franklin Trujillo, who runs Waymo’s city-expansion team, gave the Times the argument that justifies the valuation: software written for one city increasingly works in the next one without retraining for the locale. Denver, Las Vegas, San Diego and Tampa all went fully driverless inside a single month.
Grant him the point. The driving software probably does generalize. The trouble is that the software was never the expensive part of a new metro.
The Plan
Waymo’s stated plan is to keep adding cities and to make the vehicles handle novel situations without having seen them before. Trujillo cites a safety record showing 94% fewer serious-injury crashes than human drivers.
The plan Alphabet is funding looks different when you read the segment line. Other Bets brought in $411 million in the first quarter of 2026 and $382 million in the second, a sequential decline of 7.1%. Year over year, second-quarter revenue rose 2.4%. Over the same stretch, Waymo doubled its trip count versus 2025, added five metro areas, and took its fleet past 3,000 and toward 4,000.
The operating loss went the other way, widening from $1.24 billion to $1.8 billion in the June quarter. Alphabet’s own 10-Q attributes the first-quarter increase in the Other Bets loss primarily to higher employee compensation. Waymo’s losses are a payroll story, in a segment that just lost Verily’s payroll to deconsolidation.
Do the division. In the second quarter of 2025, Other Bets lost $3.32 for every dollar of revenue. A year later, it lost $4.71.
The Business Model Angle
Ride-hailing has one structural feature that decides whether a fleet earns its cost of capital: demand is spiky. Friday nights, airport rushes, marathons, concerts, New Year’s Eve. Uber and Lyft built surge pricing because those hours carry a disproportionate share of the year’s margin, and human drivers can be bribed into showing up for them.
Waymo cannot surge its supply. It owns every car, so peak capacity is a permanent capital commitment rather than a variable one. That was already a hard economic hand. The legislation now in motion makes it harder.
Representative Kevin Mullin unveiled the AV Emergency Response Coordination Act in San Francisco on July 28. Alongside a 24-hour hotline for public officials with a 30-second response standard, the bill would create federal “geofence notices” letting authorities compel autonomous vehicles out of an area for up to 72 hours, with two-minute compliance. Mayor Daniel Lurie backed it the same day, after asking California’s transportation secretary to write operational standards for driverless cars during major events.
Read that as a business rule rather than a safety rule. The instrument targets exactly the events where ride-hail demand peaks. A fleet that gets ordered out of every marathon, festival and fireworks show is an off-peak network wearing peak-capacity capital costs.
California already moved. Rules effective July 1 let police cite AV manufacturers directly for moving violations, which converts what used to be somebody else’s traffic problem into a line item with an invoice attached. Waymo’s own ground economics are visible in gig-work postings: DoorDash drivers in Atlanta have been offered $11.25 to close a robotaxi door, Honk workers in Los Angeles up to $24 for the same task, and as much as $80 to help tow a disabled vehicle. None of those prices fall when the perception stack improves.
The Risk
The bear case here has a real steelman, and it deserves stating.
Utilization dropping while you enter new markets is normal and often correct. A car in month one of Tampa should do less work than a car in year six of Phoenix, because the service area is small and nobody has downloaded the app yet. Waymo may be buying incumbency in 15 metros ahead of a federal rulebook that will make entry more expensive later, which is a defensible land grab. Amazon ran that play for two decades.
The Other Bets read has limits too. Alphabet aggregates Waymo with Wing, X and others, and Verily’s deconsolidation in the first quarter removed revenue that had nothing to do with robotaxis. Nobody outside Alphabet can extract a Waymo P&L from that segment, and Alphabet has no obligation to publish one while the unit stays under the reporting threshold. Watch the fourth quarter, when GFiber’s exit is expected to close and Other Bets becomes the closest thing to a Waymo proxy investors have ever had.
The harder risk is the one Berkeley’s Matthew Raifman put to the Times: tens of thousands of people converging on a festival is a normal weekend in a big city, not an exotic event. NHTSA administrator Jonathan Morrison made the same point in his July 8 letter to developers, writing that “Emergency scenes are not rare or extreme ‘edge cases'” and calling the inability to handle them a functional insufficiency. At a San Francisco hearing in March, Supervisor Alan Wong put the cost transfer plainly: “Our first responders should not be AAA.”
If regulators keep reclassifying edge cases as baseline requirements, the recall count stops being the metric that matters. Permission to operate becomes the metric, and permission gets renegotiated locally, one supervisor and one fire chief at a time.
Quick Questions
Is Waymo profitable? No, and nobody outside Alphabet knows how far from it. Other Bets lost $1.8 billion on $382 million of revenue last quarter, but that segment includes several businesses.
Are rides actually flat? Waymo discloses in round numbers and infrequently. It reported passing 500,000 weekly trips in late March, repeated the figure at both quarterly earnings calls, and the Times still describes it as roughly half a million. Treat “flat” as the honest read of four disclosures, not a confirmed print.
Does the 25% utilization drop mean demand is weak? Not on its own. New metros dilute the fleet average by design. It does mean the one-million-rides target needs demand growth that has not shown up yet in anything Alphabet has said out loud.
Who pays when a Waymo blocks a fire truck? Since July 1 in California, the manufacturer can be cited directly. Everywhere else, the answer is still being written.
What is the next test? Fleet Week in October, when the Blue Angels pull crowds back to the San Francisco waterfront. Supervisor Stephen Sherrill has called a hearing for September 3 to coordinate city agencies first.
The Business Model Analyst Take
Waymo has solved the problem the market gave it credit for and has not solved the one that actually gates the P&L. The Driver generalizes across cities. The tow contract, the depot lease, the charging queue, the roadside crew and the relationship with a fire chief do not travel, and they arrive at full cost on day one in every new metro whether or not anyone in Tampa has hailed a car yet.
Reading a 25% drop in trips per vehicle as failure would be lazy. Reading it as free would be lazier. Waymo is spending roughly $2 billion a quarter to hold ground in 15 cities before the federal rulebook exists, and the draft rulebook contains a switch that removes it from the calendar’s most valuable days. That is a coherent bet on incumbency. It is not a bet on unit economics, and the two get confused constantly because both look like growth from outside.
The number to watch is not the recall count. It is trips per vehicle per week. When that line turns back up while the city count keeps climbing, the flywheel is real. Until then, Waymo is running a chain rollout with a software multiple attached.
