On May 4, 2026, Tesla quietly updated its safety page with a number that most people would scroll past: 10,010,684,206. Ten billion miles. It’s the kind of figure that sounds impressive in a press release but means almost nothing to the average person trying to understand when they can actually take a nap while their car drives them to work.
Here’s why this number matters and more importantly, why it might not matter as much as Tesla wants you to think.
The Goalpost That Keeps Moving
Let’s rewind to January 8, 2026. Tesla had just failed to deliver on Elon Musk’s promise of unsupervised Full Self-Driving by the end of 2025. Again. In what’s becoming a familiar pattern, Musk took to X (formerly Twitter) with a new explanation: “Roughly 10 billion miles of training data is needed to achieve safe unsupervised self-driving.”
Ten billion. Not the six billion he’d mentioned earlier. Not the “sometime in 2025” he’d promised before that. Ten billion miles the threshold that, once crossed, would supposedly unlock true autonomous driving.
Four months later, Tesla crossed it. The fleet is now adding roughly 29 million miles per day, nearly double the rate from the start of 2026. At this pace, the next billion miles will arrive in under four weeks.
So… are we there yet?
Not even close.
What 10 Billion Miles Actually Represents
Before we get into what this milestone doesn’t mean, let’s acknowledge what it does. Ten billion miles of real-world driving data is genuinely unprecedented. No other company not Waymo, not Cruise, not anyone has collected data at this scale.
To put it in perspective:
- About 3.76 billion of those miles were driven on city streets the hardest environment for any autonomous system
- The fleet is now accounting for roughly 0.25% of every mile driven in North America each year
- That’s about 0.05% of all miles driven globally, and both figures are climbing fast
The compression of time between billion-mile markers tells its own story. The most recent billion from 9 billion to 10 billion took just 31 days. The previous billion took 43 days. The one before that took 53 days. The trend is unmistakable: Tesla’s data collection is accelerating dramatically.
This acceleration isn’t random. More people are buying FSD (Supervised), more users are actually using it, and Tesla’s software updates are making people more comfortable keeping it engaged. It’s a virtuous cycle or it would be, if data volume alone could solve the problem.
The Uncomfortable Truth About Training Data
Here’s something that doesn’t get talked about enough in all the breathless coverage of Tesla’s mileage milestones: there’s no scientific consensus that any specific number of miles equals “ready for autonomy.”
Think about it. When Musk said six billion miles were needed, was that based on rigorous analysis of edge case frequency distributions? Or was it a number that sounded big enough to justify another delay? When he revised it to ten billion, did Tesla’s AI team suddenly discover new mathematical requirements? Or did they just need a new goalpost after missing the old one?
I’m not saying the data is useless it’s incredibly valuable. Every mile teaches Tesla’s neural networks something. But the relationship between data volume and capability isn’t linear. The hardest problems in autonomous driving aren’t about common scenarios that happen millions of times. They’re about the weird, rare edge cases that might occur once in a billion miles or once in ten billion and kill someone.
Ask yourself: How many times does an autonomous system need to see a mattress fly off a truck in front of it to handle it perfectly every single time? How many instances of sun glare hitting the camera at precisely the wrong angle? How many scenarios where a pedestrian is wearing a costume that makes them look like a traffic sign?
The tail end of edge cases is, for all practical purposes, infinite. And this is where Tesla’s camera-only approach runs into physics problems that more data can’t solve.
Why Cameras Alone Might Not Be Enough
Tesla’s bet on a vision-only system is either genius or reckless, depending on who you ask. Elon Musk has been famously dismissive of LiDAR, calling it “a fool’s errand” and declaring that “anyone relying on LiDAR is doomed.”
His argument makes intuitive sense: humans drive with just two eyes (cameras), so why would cars need anything else? If the neural network is good enough (with enough training data from billions of miles), it should be able to replicate human vision and decision-making.
Except there’s a problem. Several, actually.
Problem 1: Cameras Can’t See What They Can’t See
Recent NHTSA investigations have highlighted a concerning pattern. Tesla’s FSD system has been involved in crashes where it failed to detect or respond appropriately to reduced visibility conditions sun glare, fog, airborne dust. In many of these crashes, according to NHTSA, “FSD also lost track of or never detected a lead vehicle in its path.”
This isn’t a training data problem. It’s a sensor limitation problem. When you’re driving straight into the sun and your camera is overwhelmed by glare, no amount of previous training data will help you see the stopped car ahead. A human might squint, slow down, or use peripheral vision. A camera just… doesn’t see.
LiDAR doesn’t have this problem. It literally shoots lasers and measures how long they take to bounce back. Fog, glare, darkness these don’t affect LiDAR the same way they affect cameras. This is why companies like Waymo use camera + LiDAR + radar. It’s not because they’re wasteful or behind on AI. It’s because redundancy saves lives.
Problem 2: The Austin Robotaxi Reality Check
In June 2025, Tesla launched a limited robotaxi service in Austin, Texas. Modified Model Y vehicles running FSD-Unsupervised software, with onboard safety monitors just in case. Through February 2026, this fleet logged about 800,000 miles and reported 14 crashes to NHTSA.
Let’s do the math: one crash every 57,000 miles. According to analysis from Bloomberg and Electrek, that’s roughly four times the average human-driven crash rate in similar urban conditions.
Wait. Four times worse than humans? How does that square with Tesla’s safety page claiming one major collision per 5.3 million FSD miles compared to one per 660,000 miles for the average US driver?
The answer lies in methodology, reporting standards, and what counts as a “major collision.” But here’s the thing: when you’re operating a robotaxi service actual unsupervised driving where passengers pay for rides crashes get reported to NHTSA under Standing General Order in near real-time. There’s no wiggle room in the numbers.
Compare that to Waymo, which has logged over 200 million autonomous miles with a reported safety record showing 90% fewer serious injury-causing crashes and 82% fewer airbag deployments than human drivers. The gap isn’t small. It’s enormous.
The Liability Question Nobody Wants to Answer
Here’s the dirty secret about “Full Self-Driving (Supervised)”: Tesla takes zero responsibility when it crashes.
Read that again. Zero.
When you enable FSD in your Tesla, you’re still legally the driver. You’re still responsible for monitoring the system. You’re still liable if something goes wrong. Tesla’s terms are crystal clear: this is a Level 2 driver assistance system, not autonomous driving.
Meanwhile, Waymo operates true Level 4 autonomy in ten cities. No human behind the wheel. No steering wheel at all in some vehicles. When a Waymo crashes, Waymo is liable. The company accepts legal responsibility for the driving decisions.
That’s the real threshold for unsupervised autonomy not ten billion miles, not twenty billion miles. It’s the moment a company is willing to say: “We’ll take over your insurance. We’re responsible now. You can sleep.”
During Tesla’s Q1 2026 earnings call, Musk pushed the consumer unsupervised FSD timeline to “no earlier than Q4 2026, starting in select states.” Even that came with caveats. Heavy caveats.
Because here’s what accepting liability means: every crash becomes Tesla’s legal and financial problem. Every injury lawsuit, every wrongful death claim Tesla would be the defendant, not the driver. With ten billion miles of data, how confident is Tesla in that proposition?
Not confident enough to actually do it, apparently.
The NHTSA Investigations Nobody Talks About
While Tesla celebrates mileage milestones, federal regulators are asking harder questions.
On March 18, 2026, NHTSA escalated its investigation into Tesla FSD from a Preliminary Evaluation to a formal Engineering Analysis the last step before the agency can demand a recall. The investigation now covers approximately 3.2 million Tesla vehicles.
The concerns? Tesla’s camera-based system allegedly fails to detect or warn drivers appropriately under degraded visibility conditions. NHTSA has documented nine FSD crashes including one fatality and at least two injury crashes where the system couldn’t handle common roadway conditions like sun glare or airborne dust.
There’s a separate investigation too opened in October 2025 after 58 complaints described FSD running red lights and crossing into wrong-way lanes. The agency cataloged 80 instances across the complaints.
Then there’s the August 2025 Miami federal jury verdict: $243 million awarded in a 2019 Autopilot crash, with Tesla assigned 33% of the fault. The jury found that Tesla’s system design was a substantial factor in the crash, even though the driver was distracted.
These aren’t outliers. They’re patterns.
What Comes After 10 Billion Miles?
The honest answer? More miles. Lots more miles.
At the current rate of 29 million miles per day, Tesla will hit 11 billion miles in about a month. By the end of 2026, the fleet could log 11 to 12 billion additional miles bringing the total to over 20 billion.
Will that be enough for true unsupervised driving?
Based on Musk’s track record with FSD timelines, I wouldn’t bet on it. He’s been promising autonomous Teslas for a decade now. Every year, the goalpost moves. More data needed. Different approach required. Regulatory approval pending. Next year for sure.
Here’s what actually needs to happen before unsupervised FSD becomes reality:
1. Solve the Sensor Limitation Problem
Either Tesla’s cameras need to work flawlessly in all visibility conditions (which might be physically impossible), or Tesla needs to admit that LiDAR or radar redundancy makes sense. Ego is expensive when it’s measured in lives.
2. Handle the Edge Cases
Ten billion miles sounds like a lot until you realize that rare events are still rare at that scale. A scenario that happens once per billion miles will occur ten times in Tesla’s dataset not enough to guarantee perfect handling every time. The statistical distribution of dangerous edge cases suggests Tesla might need hundreds of billions of miles, not tens of billions.
3. Accept Legal Liability
This is the real test. When Tesla is willing to take full legal and financial responsibility for FSD crashes to truly make it unsupervised and Level 4 that’s when you’ll know they believe the technology is ready. Everything else is marketing.
4. Pass Regulatory Scrutiny
NHTSA isn’t going away. The investigations will continue, the data will be analyzed, and if the safety metrics don’t hold up to independent scrutiny, regulators will step in. This isn’t some hypothetical future concern it’s happening right now.
The Waymo Comparison Tesla Doesn’t Want You to Make
Let’s talk about what actual Level 4 autonomy looks like in 2026.
Waymo is now operating driverless robotaxis in ten cities. They’re targeting one million weekly rides with their 6th-generation system. Their vehicles have no steering wheel, no pedals, and no human safety driver. Passengers get in, tell the car where they want to go, and the car drives them there.
More importantly, Waymo’s safety data across over 200 million autonomous miles shows substantially fewer crashes than human drivers. And Waymo accepts legal liability for the driving.
This isn’t theoretical. It’s operational, at scale, right now.
Tesla’s response? Keep collecting more supervised miles while promising that unsupervised is “coming soon.” It’s the same song we’ve been hearing since 2016 when Musk first claimed that Tesla would achieve full autonomy “in two years.”
The difference in approaches is stark:
Waymo: Cameras + LiDAR + radar. Geo-fenced operating areas. Extensive HD mapping. Slow, methodical expansion. Legal liability accepted. Level 4 today.
Tesla: Cameras only. Works anywhere (in theory). No HD maps needed. Rapid data collection. No liability accepted. Still Level 2. Level 4 “coming soon” (for the tenth year in a row).
One approach is conservative and operational. The other is ambitious and aspirational. Which one would you trust your life to?
Why This Matters Beyond Tesla
Tesla’s ten billion mile milestone and the hype surrounding it matters because it shapes public perception of autonomous driving technology.
When people hear “ten billion miles” and “Full Self-Driving,” many assume the technology is nearly ready. They might get complacent behind the wheel, trusting the system more than they should. They might not understand that “supervised” means “you’re still the driver and still legally responsible.”
The National Transportation Safety Board has repeatedly warned about this complacency problem. Drivers misunderstand the capabilities of driver assistance systems, over-rely on them, and stop paying attention. People have literally died because they trusted their Tesla to drive itself.
Meanwhile, the terminology itself is problematic. “Full Self-Driving” for a system that absolutely, definitively does not drive itself fully is… let’s call it optimistic marketing. It’s like selling a plane ticket and calling it “Full Flying” but specifying in the fine print that you still need to pilot the aircraft yourself.
The Real Timeline (Probably)
Based on everything we know the technical challenges, the regulatory scrutiny, the liability questions, and Musk’s historical accuracy on FSD predictions here’s my realistic assessment:
Q4 2026: Unlikely to see true unsupervised FSD for consumers. Musk’s “no earlier than Q4 2026” timeline already builds in wiggle room, and his track record suggests further delays.
2027: Possible limited rollout in select geofenced areas similar to Waymo’s approach but years behind. Maybe some states, specific routes, certain conditions. Not the “works everywhere” dream.
2028-2030: If Tesla adds additional sensors (swallowing pride and admitting LiDAR has merit) and fundamentally improves the system’s reliability, we might see broader deployment. But this assumes regulators are satisfied and liability issues are resolved.
The Wildcard: A serious crash involving FSD that leads to massive liability, regulatory crackdown, or public backlash could set the timeline back years. One high-profile tragedy is all it takes.
What Should You Actually Expect?
If you own a Tesla or are thinking about buying one, here’s what you should know:
FSD (Supervised) is getting better. The system today is measurably more capable than it was a year ago. The intervention rate is dropping. The smoothness is improving. For highway driving in good conditions, it’s genuinely impressive.
But it’s not self-driving. Not remotely. You need to pay attention every second. Your hands should hover near the wheel. Your foot should be ready to hit the brake. You are the driver, legally and practically.
The ten billion mile milestone is marketing. It’s a big number that sounds impressive but doesn’t represent a technical breakthrough or a fundamental capability shift. It’s data valuable data but data alone doesn’t solve the core challenges of edge case handling and sensor limitations.
True autonomy remains years away. Unless Tesla makes significant architectural changes (different sensors, different approach), we’re looking at 2028 at the earliest for anything resembling real unsupervised operation. And that’s an optimistic estimate.
Competitors are already there. If you want to experience true Level 4 autonomy today, you can ride in a Waymo in San Francisco, Los Angeles, Phoenix, or several other cities. It’s not perfect, but it’s operational and legally responsible for its driving.
The Bottom Line
Tesla hitting ten billion FSD miles is a milestone worth noting. It demonstrates the power of fleet learning and the advantage of having millions of vehicles collecting real-world data. No other company can match this data volume, and that’s genuinely valuable.
But let’s be clear about what this milestone doesn’t mean:
It doesn’t mean unsupervised driving is imminent.
It doesn’t prove cameras alone are sufficient.
It doesn’t address the liability question.
It doesn’t satisfy regulatory concerns.
And it definitely doesn’t mean you should trust FSD to drive you home while you take a nap.
The hard part isn’t collecting ten billion miles it’s solving the problems that ten billion miles can’t solve. Physics limitations of cameras. Infinite edge case distributions. Legal liability frameworks. Regulatory approval processes. Public trust after crashes.
These are human problems, not just data problems. And they don’t have easy answers.
Musk’s timeline has proven unreliable for a decade. The “ten billion miles” threshold was set after missing the previous deadline, and there’s no reason to believe this one is more credible than the others. The goalpost will move again. It always does.
So when will Tesla achieve true unsupervised, Level 4 autonomy that doesn’t require a human driver?
The honest answer is: we don’t know. And more importantly, neither does Tesla otherwise they’d be willing to accept liability for it.
Until that day comes, FSD (Supervised) remains what it’s always been: an advanced driver assistance system that makes highway cruising easier but still requires constant human supervision. A billion more miles won’t change that fundamental reality.
The technology is impressive. The data collection is unprecedented. But impressive data collection doesn’t equal autonomous driving. And until Tesla is willing to say “we’re liable now, you can sleep,” all those billions of miles are just… miles.
Now, if you’ll excuse me, I need to keep both hands on the wheel.
Leave a Reply