Waymo vs Tesla FSD: Real Crash Data Compared

Photo by Jp Valery on Unsplash

What's on the Table

2.52. That's the number of crashes per million fully autonomous miles Waymo logged across 7.14 million miles driven in 2023, according to Waymo's own Safety Report. It's a specific number in an industry that mostly trades in vibes and marketing footage, and it's a useful place to start because Waymo and Tesla aren't actually competing in the same category — even though they're constantly compared as if they are. According to AI Fallback's original reporting, the two companies represent fundamentally different bets on how self-driving actually gets built: one geofenced and driverless today, the other camera-only and supervised, sold into millions of driveways.

Waymo's spec sheet is straightforward: lidar, radar, and cameras layered for redundancy, running in defined zones in Phoenix, San Francisco, Los Angeles, and Austin. As of late 2024, the fleet had logged over 25 million autonomous miles and was carrying more than 150,000 paid rides per week by Q4 2024 — real commercial volume, not a pilot program. Tesla's spec sheet is almost the opposite philosophy: no lidar at all, just cameras and a neural network, running in essentially every Tesla sold, but requiring a human hands-on-the-wheel (or at least eyes-on-the-road) the entire time. Tesla shifted FSD version 12 to end-to-end neural networks in late 2023, dropping the hand-coded driving rules that governed earlier versions — a bet that vision-only AI, trained on billions of fleet miles, can eventually solve driving the way a human brain does.

Side-by-Side: How They Differ

The safety data is where this comparison gets genuinely interesting, and where the two companies' numbers stop being directly comparable. Swiss Re's 2023 institute analysis, which underwrites Waymo's insurance, found Waymo's property-damage crash rate at 0.41 per million miles against an estimated 3.26 for human drivers in similar scenarios:

Chart: Swiss Re Institute comparative analysis, 2023.

Tesla doesn't publish an equivalent per-mile figure, which is part of the problem when journalists try to line these companies up head-to-head. What exists instead is NHTSA's Standing General Order database, a government-mandated reporting system that logged 736 Tesla Autopilot/FSD crashes since June 2021, including 17 fatalities. Reuters and other outlets covering that dataset have generally framed it as raw exposure — millions of Teslas driving nationwide versus Waymo's controlled fleet in four metro areas — and that's exactly the divergence worth naming rather than glossing over. Carnegie Mellon's Phil Koopman argues Waymo's redundant sensor layering \"provides multiple layers of safety that camera-only systems cannot match in challenging conditions,\" while George Mason's Missy Cummings makes the more structural point that Tesla's Level 2 system \"fundamentally requires different safety validation than true autonomy\" — meaning it's not just the technology that differs, it's the entire regulatory and testing framework each is held to. California DMV disengagement data for 2023 put Waymo at 17,311 miles per disengagement, a figure with no consumer-side Tesla equivalent since FSD disengagements aren't publicly reported the same way. Some analysts contend that gap alone makes direct crash-rate comparisons statistically invalid; others counter that Waymo's numbers, however narrow in scope, still demonstrate what deliberate, sensor-redundant engineering buys you. Both readings have merit, and neither fully settles the argument.

The AI Angle

Both systems are, underneath everything, computer vision AI projects — but they're solving different problems. Waymo trains its models on fused lidar-camera-radar data with heavy simulation testing before anything touches public roads. Tesla's approach trains transformer neural networks on billions of miles of raw camera footage pulled from its actual customer fleet, effectively using its installed base as a continuous data-collection network. It's a scale advantage no other automaker has, and it's the core of Tesla's bet that vision alone, given enough data, can match or exceed sensor-fusion systems. Whether that bet pays off is still unresolved — Chinese entrants like Baidu Apollo and Pony.ai are now deploying lower-cost robotaxis that pressure both companies' economics, and GM shut down Cruise entirely in October 2024 after its own safety incidents, a reminder that this field consolidates fast when the numbers don't hold up.

Which Fits Your Situation

For a buyer actually cross-shopping these, the real-world calculus splits cleanly. If you want to be driven — not to drive — Waymo is available today in four U.S. metros with no lease, no subscription, no ownership commitment; you pay per ride, and the 5-year TCO question simply doesn't apply because you never own the hardware. If you already own or are buying a Tesla, FSD is a software subscription layered onto a car you're financing, insuring, and depreciating regardless — so the marginal cost-per-mile math is really about whether the subscription fee buys you meaningfully less fatigue on your daily commute, not about replacing a driver. Our read: the more likely near-term outcome is coexistence rather than a winner-take-all outcome — Waymo scaling geofenced robotaxi fleets city by city while Tesla keeps pushing supervised FSD toward wider capability in millions of existing vehicles, with regulators (NHTSA has open investigations into both) as the actual pace-setter for how fast either model can expand.

Frequently Asked Questions

Disclaimer: This article is for informational purposes only and does not constitute financial or purchasing advice. Research based on publicly available sources current as of July 22, 2026.