A massive power outage in San Francisco turned into an unplanned stress test for driverless cars, freezing Waymo’s robotaxis in intersections while human-driven Teslas kept threading through the dark. The contrast has quickly become a Rorschach test for the future of autonomous mobility, pitting ultra-cautious code against improvising humans. I see the blackout not as a simple win or loss for either side, but as a revealing snapshot of how different philosophies of automation behave when the city itself goes off script.
Waymo’s decision to halt its San Francisco service after the chaos, even as Tesla’s Full Self-Driving users kept rolling, crystallizes a deeper divide in how these companies think about risk, responsibility and public trust. The blackout exposed how dependent robotaxis are on urban infrastructure, how brittle “safety first” can look in the middle of a crisis, and how quickly public opinion can swing when a futuristic service becomes a very real traffic problem.
How a citywide blackout became a robotaxi stress test
The chain reaction started with a major power failure that plunged large parts of San Francisco into darkness and knocked out traffic signals across key corridors. Utility officials described “significant and extensive” damage from a fire in a substation, warning that repairs and safe restoration would take time and that the outage remained under investigation, a reminder that the city’s basic electrical backbone is still vulnerable to sudden shocks after the fire. At the same time, more than 20,000 residents were left without power, with local leaders publicly stressing that they were in touch with community groups to protect seniors and other vulnerable people as the blackout dragged on while over 20,000 customers.
In that environment, every intersection became a test of improvisation. Human drivers reverted to the unwritten rules of four-way stops, pedestrians negotiated crossings by eye contact and hand waves, and emergency crews tried to move through a city whose normal signaling logic had vanished. Into that mix rolled fleets of highly instrumented, software-driven vehicles that had been trained on a San Francisco full of working lights and predictable patterns. The blackout did not just cut electricity, it yanked away a key layer of structure that robotaxis quietly rely on, even when their marketing promises suggest they can handle anything.
Waymo’s robotaxis freeze in the dark
As the outage spread, multiple Waymo vehicles ended up stopped in live intersections, their hazard lights blinking while confused drivers tried to steer around them. Video from San Francisco showed driverless cars halted mid-intersection on a Saturday night, with one clip credited to Kevin Chen capturing a cluster of vehicles that appeared unable to decide how to proceed without functioning signals stopped mid-intersection. Another account described Waymo’s self-driving cars being put on pause by the company after a widespread power outage in San Francisco led to traffic signal failures that directly impacted how the vehicles behaved on the street after a widespread power outage.
From a safety engineering perspective, the behavior made sense: when the environment falls outside the parameters the system understands, the safest move is to stop and wait. But in the middle of a blackout, that conservative instinct translated into blocked lanes and stalled traffic, with some local news reports noting that the vehicles had to be manually retrieved after coming to a standstill after vehicles began to stall. What looks like prudence in a lab can feel like paralysis when you are sitting behind a frozen robotaxi with no way to communicate with it and no human in the driver’s seat to wave you through.
Waymo hits pause, then carefully restarts
Faced with that chaos, Waymo made the call to pause its robotaxi service across San Francisco, effectively pulling its vehicles out of circulation while the city sorted out the blackout. The company’s move was quickly noted on social media, where one update framed it bluntly: Waymo pauses robotaxi service in San Francisco after blackout chaos, while Musk says Tesla car service was unaffected after blackout chaos. In other words, the company that has built its brand on safety and caution leaned into that identity, even at the cost of a very public retreat.
By Sunday evening, Alphabet-owned Waymo had resumed its robotaxi service in the San Francisco Bay Area, but only after what it described as a temporary suspension to assess how the blackout had impacted vehicle behavior after a temporary suspension. Another account noted that Waymo, owned by Alphabet with tickers GOOG and GOOGL, restarted its robotaxis in the San Francisco Bay Area Sunday evening after pausing them in response to the blackout that had disrupted how the vehicles operated after the blackout impacted behavior. I read that sequence as a company trying to show regulators and residents that it will err on the side of caution, even if that means conceding that its system is not yet robust to every kind of urban disruption.
Tesla’s Full Self-Driving keeps moving
While Waymo’s cars were freezing and then disappearing from San Francisco streets, Tesla’s vehicles equipped with Full Self-Driving kept operating, at least according to the company’s own framing. In the same social media update that highlighted Waymo’s pause, Musk was quoted as saying that Tesla car service was unaffected, a pointed contrast that implied his system could handle the blackout conditions that had tripped up a rival while Tesla car service. Another report on the San Francisco blackout explicitly contrasted Waymo’s stalled robotaxis with Teslas that continued to navigate the darkened streets, framing the incident as a split-screen moment for competing visions of automated driving as Waymo robotaxis stalled.
It is important to note that Tesla’s system is driver-assistance, not a fully driverless service, and that every Tesla on the road still has a human legally responsible for the vehicle. That means when traffic lights go out, the person behind the wheel can fall back on their own judgment, treating intersections as four-way stops and improvising around stalled cars. In practice, that hybrid model looked more resilient during the blackout, because the human could override or supplement the software in ways a Waymo passenger simply cannot. The blackout did not prove that Tesla’s code is smarter than Waymo’s, but it did highlight how much flexibility remains in the human-machine partnership when things go sideways.
Two philosophies of autonomy collide
What unfolded in San Francisco was not just a tale of one company stopping and another continuing, it was a clash of philosophies about how autonomy should be deployed in real cities. Waymo’s robotaxis are designed to operate without a human driver at all, which means the system must be conservative enough to handle edge cases without relying on human intuition. That is why, when traffic lights failed and the environment no longer matched its training data, the vehicles defaulted to freezing in place, a behavior that some observers described as robotaxis that “froze on the streets” when the power outage knocked out signals, while human-driven cars kept adapting and navigating without panic as robotaxis froze on the streets.
Tesla, by contrast, has long treated autonomy as a spectrum, with its Full Self-Driving features marketed as advanced assistance that still expects a human to stay engaged. That approach can look riskier in normal conditions, because it leans on drivers to monitor a complex system that is doing much of the work. In a blackout, however, it meant that Teslas could keep moving through intersections where the rules had suddenly changed, because the human could reinterpret the scene in real time. The San Francisco incident did not settle the debate over which philosophy is safer overall, but it did show that a fully driverless model can be more brittle when the city’s infrastructure fails, while a human-in-the-loop system can absorb some of that shock.
Public perception and political pressure in San Francisco
For residents already skeptical of robotaxis, the sight of empty cars blocking intersections during a citywide emergency was a powerful image. One widely shared clip showed Waymo vehicles stopped in the middle of a junction in San Francisco on Saturday, with bystanders filming as traffic tried to weave around them and the caption highlighting that the company had halted service citywide around 8 p.m. because of the PG&E outage after Waymo halted service citywide. Another video-focused account emphasized that Waymo temporarily halted its service in San Francisco on Saturday after the widespread power outage hit the city and caused some of its vehicles to stall, according to local news reports, reinforcing the impression of a system that could not cope with the disruption after the widespread power outage hit.
Those images land in a city that has already been debating how many robotaxis it wants on its streets and under what conditions. When more than 20,000 people are without power and local officials are focused on keeping seniors safe, the idea that a private company’s vehicles are adding to the confusion at intersections is politically fraught as leaders look out for seniors. I expect regulators to seize on the blackout as a case study in how autonomous fleets should be required to respond when city infrastructure fails, from automatic geofenced slowdowns to clearer protocols for remote human intervention.
Infrastructure fragility meets software optimism
The blackout also underscored a tension that often gets glossed over in glossy autonomy demos: self-driving systems are only as resilient as the cities they operate in. The fire that caused “significant and extensive” damage at the substation did not just knock out lights, it exposed how much modern mobility, human and robotic, depends on a stable grid and functioning signals after significant and extensive damage. Waymo’s vehicles, for all their lidar and machine learning, still expect a certain baseline of order in the environment, and when that order disappears, their safest move is to stop. Tesla’s cars, for all their software, still rely on human drivers who learned to treat dark intersections as four-way stops long before over-the-air updates existed.
That gap between software optimism and infrastructure reality is likely to widen as more cities experiment with robotaxis without simultaneously hardening their grids, traffic systems and communication networks. In San Francisco, the blackout became a live demonstration of what happens when cutting-edge mobility meets aging electrical hardware and a utility still investigating the root cause of a major failure. The lesson I draw is that autonomy cannot be treated as a bolt-on upgrade to cities as they are; it has to be paired with investment in the mundane but critical systems that keep the lights, and the traffic signals, on.
What the blackout means for the next phase of robotaxis
For Waymo, the incident will likely accelerate work on how its vehicles interpret and respond to large-scale signal failures, from better detection of dark intersections to more nuanced rules for inching forward when other drivers clearly expect movement. The company has already shown it is willing to suspend service when conditions fall outside its comfort zone, as it did when it paused robotaxis in San Francisco and then resumed them in the Bay Area after reviewing how the blackout had affected behavior after it resumed its robotaxi service. I expect future updates to bake in more explicit “blackout modes” that treat failed signals as structured scenarios rather than total unknowns.
For Tesla, the episode is likely to be used as proof that its Full Self-Driving approach is robust, but regulators may see it differently. The fact that Teslas kept moving while robotaxis froze does not erase longstanding concerns about how drivers supervise advanced assistance systems, especially in chaotic conditions. One detailed account of the San Francisco blackout framed the moment as a split screen, with Waymo robotaxis stalled and Teslas continuing to navigate, and even referenced prompts like “Download App” and “Follow Us On” around the coverage as readers were urged to keep up with the unfolding story of how Waymo and Teslas handled the outage and its traffic as coverage urged readers to follow. The next phase of regulation will have to grapple with both sides of that picture: the brittleness of fully driverless fleets in edge cases and the risks of letting human drivers lean too heavily on software in emergencies.
More from MorningOverview