In October 2020, Waymo put the first fully driverless commercial rides on public roads in Phoenix. No safety driver. No one watching from the front seat. Passengers got in, went where they needed to go, and got out.
That moment is the line that matters. Not the first autonomous mile in a test environment. The first paying customer in a production vehicle with no human backup.
By end of 2025: 450,000 paid rides per week. 14 million trips in the year, three times the 2024 figure. More than 100 million autonomous miles. Annualised revenue approaching $350 million. Weekly rides grew 10x in under 20 months.
Waymo has proven the technology works. It has not yet proven the business does. That’s the more honest and more interesting story.
What the Safety Data Actually Shows
Across 56.7 million autonomous miles and multiple peer-reviewed studies, the numbers are consistent.
73–84% fewer airbag or injury crashes compared to human drivers. 92% fewer pedestrian injury crashes. 85% fewer cyclist injury crashes. Roughly 90% fewer serious injury incidents — 0.02 per million miles versus 0.22 for human drivers. In comparable conditions, the human crash rate is 6.7 times higher.
These are not promotional figures. They come from external studies, and the magnitude is large enough that the direction won’t reverse with more data. On the hardest, most consequential test — does this system hurt fewer people than human drivers — the answer is clearly yes.
This part of the Waymo story tends to get buried under coverage of the financial losses. It shouldn’t. The technology performs where it most needs to.
The Honest Counterbalance
Also true: three software recalls in 2024–2025.
The most serious came in December 2025 — an NHTSA investigation and voluntary recall after 19 documented school bus violations in Austin, including one incident where a student was still in the road. Some of these occurred after Waymo had stated the fix was already deployed.
It’s worth being precise about what this shows. It doesn’t contradict the aggregate safety data. Across 100 million miles, Waymo still outperforms human drivers. What it reveals is a structural challenge that doesn’t go away with more compute: the question is no longer whether the model can drive — that’s answered. The harder question is whether the system has encountered enough rare environmental combinations to behave safely in the long tail. A school bus at an unexpected angle, a cyclist in an unusual position, a road condition that exists in production but wasn’t in the training distribution. You don’t know what you haven’t seen.
This is a deployment and validation problem as much as a model problem. In open-world physical deployment, edge cases you didn’t test become incidents — not because the model is wrong in aggregate, but because the long tail of real environments is effectively infinite. The error rate per million miles can be very low. The absolute number of incidents still rises as the fleet scales. Maintaining safety performance per mile while operating at 10x the previous fleet size requires continuous validation at every tier of the system — not a launch gate to pass and move on from. That’s the ongoing engineering challenge, and it doesn’t have an end date.
The Unit Economics
Here’s where the story gets harder.
$30 billion invested since 2009. Alphabet’s Other Bets segment lost $3.6 billion in Q4 2025 alone. The company is not profitable. At current ride volumes, annualised revenue of $350 million sits against a capital base with a very long payback horizon.
At $15–17 per ride — about 15% below Uber and Lyft — Waymo is pricing for market development, not margin. The bet is that volume and operational efficiency will eventually close the economics. Alphabet’s CFO has said Waymo should “meaningfully contribute” to financials by 2027. A $16 billion Series D in February 2026 at a $126 billion valuation signals institutional capital still believes the trajectory.
That claim is probably credible in the narrow sense — efficiency gains from Gen 6’s cost reduction, improved utilisation as the fleet scales, infrastructure leverage from city expansion. But history suggests investors hear these timelines as platform revolutions, while operators experience them as years of fleet permitting, insurance framework development, process integration, and incremental unit economics improvement. “Meaningfully contribute to financials” and “established profitable business” are different claims, separated by a long distance. 2027 is when the former might arrive. The latter is a harder timeline to name.
Gen 6: The Cost Curve Inflection
Waymo’s Gen 5 vehicle — a Jaguar I-PACE with their sensor stack — cost roughly $150,000 per unit. At that price, fleet economics are structurally difficult regardless of ride revenue.
Gen 6, the Ojai, deployed in February 2026 on a Zeekr van platform. Cost: $50,000–55,000. That’s a 3x reduction. It runs 42% fewer sensors and delivers better performance. Magna is manufacturing at scale, targeting tens of thousands annually.
This is the actual inflection. At $150,000 per vehicle, no ride pricing makes the math work at realistic utilisation rates. At $50,000, the capital model for a large fleet looks fundamentally different. At the company’s stated target of one million weekly rides in 2026, annualised revenue approaches $780 million to $1 billion — the range where the numbers start resembling a business rather than a research programme.
No credible competitor is at commercial scale. Tesla has 44 vehicles in Austin running supervised FSD — 60 times smaller fleet, and a fundamentally different product category. The comparison deserves more precision than it usually gets: Tesla’s supervised FSD is a consumer assistance product operating across open public roads with a human ready to intervene at any moment. Waymo is deploying operational autonomy inside constrained commercial workflows — specific cities, mapped routes, defined operational parameters, no human backup. These aren’t two points on the same spectrum. They’re different design contracts with different validation requirements and different paths to commercial viability. The broader lesson applies beyond vehicles: industrial AI tends to succeed earlier when the environment, routes, and operational assumptions are tightly controlled. The more open and variable the operating context, the longer the validation tail. Comparing the two as if they’re racing toward the same finish line obscures the actual competitive question for each.
GM Cruise is rebuilding after their 2023 incident. Amazon Zoox is early stage. Waymo’s race is not primarily against competitors. It’s against their own cost curve: can they reduce per-vehicle cost and grow fleet size fast enough before a better-capitalised operator arrives with a comparable system?
What This Tells Enterprise Leaders
Waymo is the most advanced available case study for physical-world AI deployment where failure has consequences. Three things stand out.
The validation bar is higher than software, and it doesn’t end at launch. One hundred million miles of safety data and three software recalls are not contradictory. In open-world physical deployment, edge cases you didn’t test become incidents. The validation process is ongoing, not a gate to pass before commercial release.
Capital requirements are front-loaded and long. $30 billion before profitability is not a universal template, but the general shape applies broadly: massive upfront investment in data, hardware, and validation, with unit economics that only close at scale. The companies that succeed are the ones that can sustain the investment curve long enough for the cost structure to improve.
The cost reduction is the real news. Gen 6’s 3x hardware cost reduction matters more strategically than any weekly ride milestone. In capital-intensive AI deployment, the moment per-unit economics become viable is when the business model becomes real. Watch the cost of the enabling hardware, not the revenue headline.
Waymo has crossed the hardest threshold: from experiment to paid product at scale, with safety data that holds up to external scrutiny. Whether the business closes by 2027 as Alphabet expects is the open question. The technology answer is largely in. The economics answer isn’t.
That’s a more interesting position than the hype says, and a more promising one than the skeptics allow.