Waymo has now driven 220.6 million miles with nobody behind the wheel, and its own analysis puts the result at 94% fewer serious-injury-or-worse crashes than human drivers on the same roads. Translated into absolute terms that is an estimated 47 serious or fatal injury crashes that did not happen, plus 707 injury crashes of any severity. The commercial side crossed a threshold in the same window: roughly 500,000 paid rides a week across 10 US cities, targeting a million by year end. The figures below come from Waymo’s Safety Impact data hub, California DMV testing disclosures, and crash reports filed under NHTSA’s Standing General Order.
TL;DR
- Waymo has driven 220.6 million rider-only miles through March 2026 (Waymo)
- 94% fewer serious-injury-or-worse crashes than the human benchmark (Waymo)
- 82% fewer airbag-deployment crashes and 82% fewer any-injury crashes (Waymo)
- An estimated 47 serious or fatal injury crashes avoided (Waymo)
- An estimated 707 injury crashes of any severity avoided (Waymo)
- 93% fewer pedestrian injury crashes, an estimated 76 avoided (Waymo)
- Phoenix alone accounts for 80.551 million rider-only miles (Waymo)
- Roughly 500,000 paid robotaxi rides per week across 10 US cities (Waymo, July 2026)
- Target of more than 1 million weekly rides by end of 2026 (Waymo)
- California AVs logged over 9 million test miles in the year to November 30, 2025 (California DMV)
- Disengagements fell to under 9,300, from 28,800 the prior year (California DMV)
- Waymo reported 63,415 miles per disengagement across 5.2 million California miles (California DMV)
- California scrapped annual disengagement reporting for failure-based monthly and quarterly filings (California DMV)
1. The Safety Record at Scale
Autonomous vehicle safety claims were untestable for years because the mileage was too small. That is no longer true. Across 220.6 million rider-only miles, Waymo reports 94% fewer crashes causing serious or fatal injury, 82% fewer crashes where an airbag deployed, and 82% fewer crashes causing any reported injury than human drivers would have produced over the same distance in the same places. The absolute counts matter more than the percentages for anyone weighing policy: 47, 305, and 707 crashes respectively.
| Metric | Value | Source |
|---|---|---|
| Rider-only miles through March 2026 | 220.6 million | Waymo |
| Reduction in serious-or-fatal-injury crashes | 94% | Waymo |
| Estimated serious-or-fatal crashes avoided | 47 | Waymo |
| Reduction in airbag-deployment crashes | 82% | Waymo |
| Estimated airbag-deployment crashes avoided | 305 | Waymo |
| Reduction in any-injury-reported crashes | 82% | Waymo |
| Estimated injury crashes avoided | 707 | Waymo |
| Underreporting correction applied to human benchmark | 32% | Waymo, per NHTSA research |
| Methodology framework | RAVE checklist | Waymo |
The 32% correction is the methodological hinge: human crash rates are drawn from police reports, which systematically undercount minor collisions, so Waymo inflates the human benchmark rather than comparing against raw police data. That choice is disclosed and defensible, but it is a choice. Source: Waymo Safety Impact.
2. Who Actually Gets Protected
Aggregate crash reduction hides the question that matters most politically, which is what happens to people outside the vehicle. Waymo reports 93% fewer injury crashes involving pedestrians, an estimated 76 avoided, alongside 84% reductions for both cyclists and motorcyclists. Vulnerable road users are where human driving performs worst and where automation’s advantage in reaction time and attention should theoretically be largest, and the data is consistent with that expectation.
| Metric | Value | Source |
|---|---|---|
| Reduction in pedestrian injury crashes | 93% | Waymo |
| Estimated pedestrian injury crashes avoided | 76 | Waymo |
| Reduction in cyclist injury crashes | 84% | Waymo |
| Estimated cyclist injury crashes avoided | 48 | Waymo |
| Reduction in motorcyclist injury crashes | 84% | Waymo |
| Estimated motorcyclist injury crashes avoided | 32 | Waymo |
| Combined vulnerable road user crashes avoided | 156 | Derived from Waymo figures |
| Cities with sufficient data for comparison | 5 | Waymo |
The combined figure is simple addition on Waymo’s three published estimates and is not itself a disclosed metric. Source: Waymo’s safety data hub.
3. Where the Miles Are
Safety percentages are only meaningful against the road types that produced them, and Waymo’s mileage is heavily concentrated. Phoenix accounts for 80.551 million rider-only miles, more than a third of the total, with the San Francisco Bay Area at 67.078 million second. Phoenix and San Francisco are close to opposite driving environments, which strengthens the generalizability of the combined result more than a single-city dataset would.
| Metric | Value | Source |
|---|---|---|
| Phoenix rider-only miles | 80.551 million | Waymo |
| San Francisco Bay Area rider-only miles | 67.078 million | Waymo |
| Los Angeles rider-only miles | 51.816 million | Waymo |
| Austin rider-only miles | 15.789 million | Waymo |
| Atlanta rider-only miles | 5.379 million | Waymo |
| Phoenix share of total miles | approx. 37% | Derived from Waymo figures |
| Top two cities combined share | approx. 67% | Derived from Waymo figures |
| Data cutoff for these figures | March 2026 | Waymo |
Austin and Atlanta are new enough that their individual results carry wide confidence intervals, which is why Waymo reports five cities rather than every market it serves. Source: Waymo safety data update.
4. Commercial Scale Arrived
2026 is the year robotaxis stopped being a pilot. Waymo reported roughly 500,000 paid rides per week across 10 US cities as of July 2026, up from about 400,000 earlier in the year, against a stated target of more than 1 million weekly rides by year end. At that run rate the service would be generating roughly USD 1 billion annualized, which would make it the first autonomous vehicle business at genuine commercial scale.
| Metric | Value | Source |
|---|---|---|
| Paid rides per week, July 2026 | approx. 500,000 | Waymo |
| Paid rides per week, earlier in 2026 | approx. 400,000 | Waymo |
| Stated target by end of 2026 | more than 1 million weekly | Waymo |
| US cities with paid service | 10 | Waymo |
| Cities named | Phoenix, San Francisco, Los Angeles, Austin, Atlanta, Miami, Dallas, Houston, San Antonio, Orlando | Waymo |
| Planned expansion by end of 2026 | more than 20 cities, including London and Tokyo | Waymo |
| Robotaxis with 5th-generation system, December 2025 | 3,067 | NHTSA filing |
| Implied annualized revenue at 1 million weekly rides | approx. USD 1 billion | Industry estimate |
The revenue figure is an outside estimate derived from ride volume and typical fares, not a Waymo disclosure, and Alphabet does not break out Waymo revenue. Broader in-car interface data sits in our automotive voice assistant statistics. Sources: Forbes on Waymo’s 1 million ride target and TechCrunch on ridership growth.
5. What California’s Data Shows
California remains the only jurisdiction publishing standardized cross-operator testing data, which makes it the closest thing to an independent scoreboard. Autonomous vehicles logged more than 9 million test miles in the year to November 30, 2025, with fewer than 9,300 disengagements against 28,800 the year before, a roughly three-fold improvement. Waymo accounted for over 68% of all vehicle miles travelled.
| Metric | Value | Source |
|---|---|---|
| Total AV test miles, Dec 1 2024 to Nov 30 2025 | more than 9 million | California DMV |
| Total disengagements | fewer than 9,300 | California DMV |
| Disengagements the prior year | 28,800 | California DMV |
| Waymo share of vehicle miles travelled | over 68% | California DMV |
| Zoox share of drivered miles | nearly 25% | California DMV |
| Waymo California miles | 5.2 million | California DMV |
| Waymo miles per disengagement | 63,415 | California DMV |
| New reporting cadence, driverless testing | monthly | California DMV |
| New reporting cadence, deployment permits | quarterly | California DMV |
The regulatory change matters more than any single number here: by replacing annual disengagement reports with failure-based filings, California has ended the continuity of its own longest-running AV dataset. Regulatory context sits in our AI regulation statistics. Source: California DMV, AV permit holders logged more than 9 million test miles.
6. The Federal Picture and a Comparison Worth Resisting
Every operator running an automated driving system on US public roads must report crashes to NHTSA under a Standing General Order, which produces the only cross-company federal dataset. Tesla has filed 18 ADS incidents since its Austin launch against roughly 697 for Waymo, and that comparison is close to meaningless without exposure data. Waymo drives orders of magnitude more autonomous miles; a lower absolute count reflects a smaller denominator, not a demonstrated safety advantage.
| Metric | Value | Source |
|---|---|---|
| Tesla ADS incidents reported since Austin launch | 18 | NHTSA filings |
| Waymo ADS incidents reported | approx. 697 | NHTSA filings |
| Tesla crash narratives initially redacted | all of them | NHTSA filings |
| Tesla narratives later unredacted | 17 | NHTSA filings |
| Other operators filing unredacted narratives | Waymo, Zoox, Avride, May Mobility | NHTSA filings |
| First Tesla report categorised as remote operator collision | 2026 | NHTSA filings |
| Prior remote intervention incidents | July 2025 and January 2026 | NHTSA filings |
| Reporting cadence under the order | near real time | NHTSA |
The transparency gap is the more defensible finding in this section: Tesla was the only ADS operator to mark every crash narrative as confidential business information, a designation it later removed. Enterprise deployment context for the underlying models sits in our enterprise AI adoption statistics. Source: NHTSA Standing General Order on Crash Reporting.
Summary: Self-Driving Cars by the Numbers
| Metric | Value | Source |
|---|---|---|
| Waymo rider-only miles | 220.6 million | Waymo |
| Reduction in serious-or-fatal-injury crashes | 94% | Waymo |
| Serious or fatal injury crashes avoided | 47 | Waymo |
| Reduction in airbag-deployment crashes | 82% | Waymo |
| Airbag-deployment crashes avoided | 305 | Waymo |
| Reduction in any-injury crashes | 82% | Waymo |
| Injury crashes avoided | 707 | Waymo |
| Reduction in pedestrian injury crashes | 93% | Waymo |
| Pedestrian injury crashes avoided | 76 | Waymo |
| Reduction in cyclist injury crashes | 84% | Waymo |
| Reduction in motorcyclist injury crashes | 84% | Waymo |
| Phoenix rider-only miles | 80.551 million | Waymo |
| San Francisco Bay Area miles | 67.078 million | Waymo |
| Los Angeles miles | 51.816 million | Waymo |
| Paid rides per week, July 2026 | approx. 500,000 | Waymo |
| Cities with paid service | 10 | Waymo |
| Target weekly rides by end of 2026 | more than 1 million | Waymo |
| Robotaxis with 5th-generation system | 3,067 | NHTSA filing |
| California AV test miles, year to Nov 30 2025 | more than 9 million | California DMV |
| California disengagements | fewer than 9,300 | California DMV |
| Waymo miles per disengagement in California | 63,415 | California DMV |
| Tesla ADS incidents since Austin launch | 18 | NHTSA filings |
Methodology and Sources
- Mileage, crash reduction percentages, absolute crashes avoided, vulnerable road user figures, and per-city mileage come from Waymo’s Safety Impact data hub, reflecting data through March 2026 (Waymo Safety Impact, data hub background, June 2026 update, prior mileage milestone).
- California testing miles, disengagement counts, operator shares, and the change in reporting requirements come from the California DMV (press release, permit holder listings).
- Crash reporting obligations and cross-operator filing behaviour come from NHTSA’s Standing General Order programme (NHTSA).
- Ride volume, city counts, and expansion targets come from Waymo statements as reported contemporaneously (Forbes, TechCrunch).
- Data watch: every safety figure here is Waymo’s own analysis of its own performance. The methodology is published and follows the RAVE checklist, and the comparison inflates the human benchmark by 32% to account for police underreporting, but this is not independent verification and no regulator has audited it. The results apply only to the five metropolitan areas with sufficient mileage and should not be generalised to highways, rural roads, or severe weather. Absolute crash counts filed under NHTSA’s Standing General Order cannot be compared across operators without exposure data, and NHTSA does not publish per-operator autonomous mileage, so no like-for-like federal rate exists. California’s shift away from annual disengagement reporting means the historical disengagement series ends with the year to November 30, 2025. The USD 1 billion revenue figure is an external estimate; Alphabet does not disclose Waymo revenue.
- Last updated: August 1, 2026. We update this roundup quarterly as Waymo refreshes its Safety Impact hub and new NHTSA and state filings are published.