Two deadlines are now sitting on Waymo’s desk, and neither of them is about self-driving software.
The first came from Washington. On July 8, National Highway Traffic Safety Administration Administrator Jonathan Morrison sent a letter to every autonomous vehicle developer on the Department of Transportation’s Standing General Order list, telling them that driverless cars blocking ambulances and driving into active emergency scenes is not a rounding error. It is, in his words, a functional insufficiency. Companies have until the end of July to bring the agency their solutions.
The second deadline has no date attached, which is what makes it dangerous. Uber and Waymo quietly ended their Phoenix robotaxi partnership on June 29. Austin and Atlanta are still running. Nobody in the industry seriously believes those survive indefinitely.
Put together, they describe a company that has won the hardest technical race in mobility and is now discovering that the technical race was the easy part.
Key Takeaways
- NHTSA gave AV developers until the end of July 2026 to present fixes for a documented pattern of robotaxis interfering with police, fire, and EMS. The letter names no company, but Waymo runs the largest US fleet and has the longest incident list.
- The July 4 Presidio gridlock in San Francisco, where more than a dozen Waymos drained their batteries and had to be towed, turned an abstract regulatory concern into a viral video problem.
- Uber and Waymo ended their Phoenix pilot on June 29 after roughly three years. Waymo pulled the vehicles back onto its own app. Uber says it is signing a different AV partner in the city.
- The two stories are the same story: Waymo’s business model requires it to own the customer relationship and the city relationship simultaneously, and it is currently losing ground on both.
- Waymo operates around 4,000 vehicles and roughly 500,000 paid rides a week across ten cities. Uber completed 13.5 billion trips in 2025. That asymmetry defines the negotiation.
What NHTSA Actually Demanded
Morrison’s letter, issued as a public call to action, says the agency has documented a clear pattern of driverless AVs interfering with law enforcement and other first responders, including vehicles driving directly into active emergency scenes, blocking the paths of ambulances and firefighters, and failing to recognize basic safety cues like flashing lights, flares, smoke, fire, and traffic cones.
The framing is the part worth reading twice. Morrison called the inability to detect and respond to these situations a functional insufficiency, and stated that emergency scenes are not rare or extreme edge cases. He also put it more bluntly: an AV that cannot safely interact with first responders is a danger to the general public.
That language matters more than any fine would. “Edge case” is the load-bearing word in every AV company’s safety narrative. It is how you tell investors that the remaining 1% of scenarios is a long tail you will grind down with more miles and more data. NHTSA has just told the industry that the regulator does not accept the category.
The agency doesn’t explicitly name a company, but the details point at robotaxi operators like Waymo, which declined to comment. NHTSA said it will schedule meetings with AV developers by the end of the month to hear their solutions.
For a fuller picture of Waymo’s regulatory position, see our earlier analysis of Waymo’s $16 billion raise and its New York roadblock.
The July 4 Breakdown That Gave the Letter Teeth
Regulators write letters all the time. What gives this one force is that four days earlier, San Francisco produced a two-hour video essay proving the point.
Dozens of Waymo vehicles became stranded in heavy post-fireworks traffic in the Presidio on July 4. Some stalled in roadways, others required towing, and one carrying passengers drove over a lit firework. Video verified by NBC News showed at least a dozen stationary Waymos, many of them Jaguar I-PACE models, lined up on a street after the Golden Gate Bridge fireworks display. One rider spent nearly two hours in standstill traffic, and described people getting out of their cars to yell at vehicles that had no driver to yell at. A separate unoccupied Waymo caught fire after driving over a firework.
Waymo said extreme congestion in northern San Francisco disrupted normal operations, that some vehicles ran out of charge while idling and had to be towed, and that it had preemptively embedded a staffer in the city’s Emergency Operations Center.
Read that last part carefully. Waymo did the responsible thing. It staffed the emergency center in advance. And the fleet still died in the road.
The failure rhymes with December 2025, when a PG&E outage killed traffic signals and stranded hundreds of Waymos at intersections, forcing the company to send staff or tow trucks to retrieve 64 cars, with first responders taking the wheel in two cases. And this week, San Francisco supervisor Bilal Mahmood said he plans to open an inquiry into how AVs affected transit and emergency response on July 4.
Why This Is an Operating Model Problem, Not a Software Bug
The instinct is to file this under perception and planning. Teach the model what a flare looks like. Ship the recall. Move on.
That reading is too generous.
What the December blackout and the July 4 gridlock share is not a perception failure. It is a fallback failure. In both cases the cars behaved conservatively and correctly by their own logic, then had nowhere to go, and the human system that was supposed to rescue them could not scale to the number of vehicles that needed rescuing at once. A human driver in gridlock makes a judgment call, mounts a curb, takes an alley, follows a cop’s hand signal. A Waymo waits. Multiply that by a hundred cars in the same square mile and the fleet stops being a service and starts being an obstruction.
That is a ratio problem, not a model problem. Every robotaxi network runs on a hidden denominator: remote operators and roadside crews per vehicle. Scale the numerator faster than the denominator, and the fleet’s worst day gets worse in a straight line. This is the cost curve nobody puts in the deck, because the entire investment thesis for autonomy is that the denominator eventually goes to zero.
NHTSA’s letter is, whether it intends to be or not, a demand that the denominator go up.
How Waymo’s Year Went Sideways
The pattern only reads as a pattern when you line it up.
The Second Deadline: Uber and Waymo Are Decoupling
On June 29, Waymo robotaxis disappeared from Uber’s app in Phoenix, ending a nearly three-year partnership. Uber said the two companies ended the deployment because it hit its contracted end date, and that it is readying a separate AV partnership in the city without naming the partner. Waymo said the vehicles have been folded back into its own Phoenix fleet.
Both companies performed the ritual. Uber called Phoenix an intentionally limited pilot of just over a dozen dedicated vehicles, and said the lessons helped it scale Austin and Atlanta, where hundreds of Waymo AVs are available exclusively on Uber. Waymo called it a productive pilot that paved the way for future expansions.
Now the tell. The Phoenix vehicles will stay in the city and make autonomous deliveries via DoorDash, which competes directly with Uber Eats. And Waymo plans to offer rides through Lyft in Nashville later this year, non-exclusively.
That is not a partnership winding down. That is a supplier walking its inventory across the street.
What Uber Does Next
Here is the structural conflict, stated plainly. Uber wants to be the app you use to hail any car. Waymo wants you to know you are riding in a Waymo. Those two ambitions are compatible only while one side needs the other.
Uber’s case is demand. It completed 13.5 billion trips in 2025, a network Waymo cannot replicate at any speed. Waymo’s case is supply. It runs roughly 500,000 paid rides a week across ten cities. In a market where autonomous supply is the scarce input, the supplier holds the pricing power, and Uber knows it. That is why Uber has spent two years signing what its own executives describe as dozens of AV partners rather than betting the company on one.
The strategic question for Uber is whether it can commoditize autonomy fast enough. If a dozen credible AV stacks reach commercial viability, Uber’s aggregator model survives intact and Waymo becomes one supplier among many. If Waymo stays two years ahead, Uber’s app becomes a customer acquisition channel that Waymo can switch off market by market, exactly as it just did in Phoenix. Our breakdown of the Uber business model and the Uber SWOT analysis go deeper on that dependency.
The uncomfortable read for Uber: the NHTSA letter is arguably good news for it. Anything that slows Waymo’s city-by-city expansion buys Uber time to build a portfolio.
Tesla, Zoox, and a Rulebook Being Rewritten
The federal picture is not uniformly hostile. The same administration issuing safety ultimatums is also loosening design rules.
The updated 2026 Regulatory Plan and Unified Agenda contains a long list of proposed changes to Federal Motor Vehicle Safety Standards, the rules that govern vehicle design and equipment. Those changes could benefit companies like Tesla and Zoox that are building vehicles without steering wheels or pedals.
So the regulatory posture is split: permissive on hardware, aggressive on behavior. Build whatever shape you like. Just do not block the fire truck.
That split favors whoever is best at operational discipline rather than whoever is best at engineering theater. It is also a warning for Tesla, which is scaling a robotaxi fleet into cities this year with a very different sensing philosophy and an existing pile of NHTSA scrutiny. Our Tesla SWOT analysis and Tesla in charts lay out how much of Tesla’s current valuation rests on the robotaxi line item specifically.
The Unit Economics Nobody Wants to Discuss
Strip the story down to a P&L and here is what changed in one week.
Waymo now has to fund three things it was hoping to avoid funding at this stage. Denser remote operations coverage, so a hundred simultaneous stalls do not become an eight-hour civic incident. Deeper municipal relations staffing in every market, because SF just demonstrated that a single bad night triggers a supervisor inquiry. And a compliance workstream aimed at a federal regulator that has publicly rejected the “edge case” defense.
Meanwhile it is losing the cheapest customer acquisition channel in ride-hailing, one city at a time, and replacing it with its own app and its own marketing spend.
Every one of those is a fixed cost that scales with cities rather than with rides. That is the opposite of the operating leverage the robotaxi thesis promises. Goldman Sachs Research has projected the global robotaxi market could reach roughly $415 billion by 2035, including $48 billion in the US. The number is only reachable if per-city fixed costs fall. Right now they are rising.
What to Watch
End of July. Whether NHTSA’s meetings produce voluntary commitments or something with enforcement behind it. A functional insufficiency finding is a short walk from a defect investigation.
Austin and Atlanta. Those are the last exclusive Uber-Waymo markets. Watch for either side to start hedging publicly.
Uber’s Phoenix replacement. The unnamed partner tells you which AV stack Uber thinks is closest to commercial parity with Waymo.
Nashville. Waymo on Lyft, non-exclusive, is the template for what Waymo actually wants: aggregators as a channel, not a gatekeeper.
Waymo’s next incident. Not if. The question is whether the response looks like a company that has built an ops organization or one that has built a demo.
FAQ
What did NHTSA demand from robotaxi companies? NHTSA Administrator Jonathan Morrison gave AV developers until the end of July 2026 to address a documented trend of driverless cars interfering with law enforcement and other first responders, and said the agency will meet with developers to hear their proposed solutions.
Is the letter aimed at Waymo? It names no company. But Waymo operates the largest robotaxi fleet in the US and has had repeated run-ins with first responders, including at least six incidents through March 2026 in which first responders had to take control of vehicles.
Why did Uber and Waymo end their Phoenix partnership? Both companies say the pilot reached its contracted end date. Waymo integrated the vehicles back into its own Phoenix fleet, and Uber said it is preparing a separate AV partnership in the city.
Are Waymo rides still available on Uber? Yes, in Austin and Atlanta. In nine other cities, Waymo rides come primarily through its own app.
How big is Waymo’s fleet? Around 4,000 automated vehicles in the US, delivering roughly 500,000 paid rides a week across ten cities.
The Business Model Analyst Take
The consensus read is that Waymo won. It has the fleet, the miles, the $16 billion, and a two-year technical lead over anyone credible. Both of this week’s stories get filed as friction on the way to inevitability.
We think that read is lazy.
Waymo’s business model has a load-bearing assumption inside it: that autonomy converts a variable cost, the driver, into a fixed cost, the software, and that the fixed cost then amortizes toward zero across an ever larger fleet. Everything about the valuation depends on that conversion being clean.
What December and July 4 revealed is that the driver was never just a variable cost. The driver was also the operational shock absorber, the local knowledge layer, and the entity that keeps a fire truck moving when the map is wrong. Waymo has not eliminated that function. It has centralized it into remote ops teams and roadside crews, and centralized functions have a nasty property: they fail correlated. A hundred human drivers in gridlock produce a hundred independent improvisations. A hundred Waymos in gridlock produce one queue at one call center.
That is the real content of Morrison’s letter, even if he did not phrase it that way. He is telling the industry that its rescue capacity has to scale with its fleet, which converts a chunk of the promised software margin back into headcount.
And this is precisely the wrong moment for Waymo to be losing Uber. Distribution independence is strategically correct and financially expensive. Owning the rider means owning acquisition cost at exactly the moment your ops cost is being repriced upward by a regulator. Waymo is choosing to fight a margin war on two fronts because the alternative, permanent dependence on a company that wants to commoditize it, is worse. That is a defensible choice. It is not a free one.
The company to watch here is not Uber. It is Tesla. If Tesla’s camera-only stack turns out to be the one that scales cheaply while Waymo’s sensor-heavy fleet drowns in per-city fixed costs, the winner of the robotaxi race will not be the company with the best technology. It will be the company with the lowest cost of failure. Right now nobody has proven they have that.
Autonomy was supposed to be the end of operations. It turns out to be the beginning of a different, harder one.
