Support CleanTechnica’s work through a Substack subscription or on Stripe.
Elon Musk made a statement last week that was really hard to understand, because it was seemingly way outside of his normal statements on self-driving capability and Tesla robotaxis. I still don’t really know what to make of it, but let’s dive in. And then after looking at some of Musk’s predictions, I’ll circle around to different opinions from the public, or at least people following autonomous driving progress.
Well, one key point that may indicate why Musk’s comment was so outside the norm is that it was focused on NVIDIA’s self-driving ambitions, not Tesla’s. Whereas Tesla is always seemingly two weeks away from some magical breakthrough, Musk was basically saying that NVIDIA was years away from one.
Here’s the quick backstory in bullet-list format:
NVIDIA announced at CES that it was working on autonomous driving now and would partner with automakers on it. (Actually, I wrote about that in October, but it got a lot more attention after some announcements at CES.)
Tesla fans (read: stockholders) defensively argued that this wasn’t a threat because NVIDIA would be way behind on real-world training.
In response to a long tweet about this, Musk wrote: “Roughly 10 billion miles of training data is needed to achieve safe unsupervised self-driving. Reality has a super long tail of complexity.”
Per Tesla’s most recent shareholder letter, the company had logged nearly 6 billion FSD miles by the end of the 3rd quarter of 2025. According to it’s sort-of-live tracker here, the number is now at nearly 7.3 billion.
Before I come back to those numbers and claims, let’s remember Musk’s claim from about half a year ago. In July of 2025, Musk said that the Tesla Robotaxi network would cover “probably half the population of the US by the end of the year.” How much did it end up covering? Zero percent (0%). So, as has been the case for a decade, you have to really take Musk’s statements with a grain of salt on this topic.
Getting back to the 10 billion mile claim, if it took 4 months to gain 1.5 billion more FSD miles, you could presume it might take another 8 months to get to 10 billion miles. Okay, assume the miles driven speed up, but then again, you need time after reaching 10 billion miles to make use of the supposedly adequate data and update the system. So, anyway, say 6–12 months. But, then again, one has to ask if Musk is being overly optimistic again.
Here are some issues to consider:
Why 10 billion, such a nice round multiple of 10? Is there something genuinely behind this, or does it just sound good?
10 billion miles of what kind of driving? I’ve had “Full Self Driving” for 6 and a half years. It was useless at the beginning of that time and got better gradually, which made it more useless and requiring less intervention over time. (Though, admittedly, I still don’t prefer it to driving the car myself — I’d rather control the car, accelerate, and navigate as I feel most comfortable rather than be a passenger in a car that is driving in a way I don’t like.) Were the constant failures and problems of 2024 FSD part of those 10 billion miles? Or are we talking about 10 billion miles of limited intervention?
Is this 10 billion mile figure for the USA/world as a whole, or is it for places without snow — like California and Texas most of the time — and for when the morning or evening sun isn’t blinding the cameras?
Is 10 billion miles the point where a company, like Tesla, can take over liability for crashes and no longer leave legal liability to the owners/drivers of the cars?
Does the mileage count depend at all on the sensors being used?
Waymo says it has driven over 20 billion miles in simulation, and that presumably means it has driven far, far less than that in the real world, yet Waymo is operating actual robotaxis without human drivers or supervisors in multiple cities. And many of its trained miles were logged when hardware was much worse and software was at a much earlier stage of development. About a month ago, Waymo reported, “With well over 100 million fully autonomous miles driven, we are making streets safer where we operate — achieving a more than ten-fold reduction in crashes with serious injuries compared to human drivers.” Hmm….
And could NVIDIA not find a way to reach self-driving capability faster than Waymo did?
I think the biggest question from Elon Musk’s comment is really what it means for how far away Tesla is from unleashing true driverless robotaxis. Who knows? The guy makes all kinds of nonsense claims on social media every day, and I think there has been no one more wrong in their Tesla FSD predictions over the last decade than him. That said, I can tell that FSD is world better right now than it was a year or two ago, and I don’t have version 14 on my car since I have an old car without adequate hardware for that (another story I’ll try to publish today). Here are different popular opinions on Tesla’s proximity to actual driverless cars:
Hardcore Tesla fans (i.e., TSLA shareholders): Tesla is basically ready for this now. Any day, Tesla will flip a switch and vehicle owners around the country who have FSD will be able to sleep while the car drives, watch movies, play video games, or scroll on X. They will also be able to send those cars out ferrying passengers around and make money on the robotaxi service.
Mid-core Tesla fans (probably still shareholders): Tesla isn’t there yet, but it will be within a few months, or a year at the most.
Mid-core Tesla skeptics: Tesla is still years away from this due to that long tail of edge cases, but it will get there eventually.
Hardcore Tesla skeptics: Tesla’s hardware is inadequate for this and it will never work, at least not until Tesla implements a serious hardware overhaul and changes its approach to the software.
I’ve seen all of these arguments made a million times (hyperbole, I think), and I don’t know what I genuinely think. Frankly, I don’t have enough data or insight into the data to know. But I do find it “interesting” that Elon Musk completely missed his very strong prediction that half the US population would be within a Tesla Robotaxi service area by now, and that he seems to be saying Tesla needs 3 billion more miles of training data to release unsupervised robotaxis.
That also reminds me of his COVID-19 prediction from March 19, 2020:
That was one of the most absurd statistical predictions I’ve ever seen. I could not comprehend how he thought this was true. And ever since then, I have really, really questioned Musk’s understanding of statistics. Of course, there were other factors at play as well, but it was really incomprehensible that he could have that opinion in March 2020. However, it did help explain how he would get his Tesla FSD predictions so wrong for a decade.
Sign up for CleanTechnica’s Weekly Substack for Zach and Scott’s in-depth analyses and high level summaries, sign up for our daily newsletter, and follow us on Google News!
Advertisement
Have a tip for CleanTechnica? Want to advertise? Want to suggest a guest for our CleanTech Talk podcast? Contact us here.
Sign up for our daily newsletter for 15 new cleantech stories a day. Or sign up for our weekly one on top stories of the week if daily is too frequent.
[embedded content]
CleanTechnica uses affiliate links. See our policy here.
CleanTechnica’s Comment Policy
Share this story!