When It Comes to Tesla’s Unsupervised Full Self-Driving, Here’s Why Reality Bites

Reality bit about 24 minutes into Tesla’s 2026 Q1 earnings call, when Elon Musk finally revealed why true Full Self-Driving, first offered to Tesla customers in October 2016, had yet to materialize. “I wish it were otherwise,” Musk said. “But Hardware 3 simply does not have the capability to achieve unsupervised FSD.”

It was an extraordinary admission, as Hardware 3 was the computer system Tesla launched in 2019 to deliver full autonomous driving capability. “We did think at one point it would have that [capability],” Musk said, “but relative to Hardware 4, it has only one-eighth the memory bandwidth, [and] memory bandwidth is one of the key elements needed for unsupervised FSD.”

The reality: Tesla had for years been selling a feature it couldn’t make work. The bite: Was it fraud? Or simply a fiasco? Either way, Tesla customers who bought Hardware 3–equipped cars and paid for FSD have a right to be aggrieved; the lawsuits are going to come thick and fast. Tesla is almost certainly going to be on the hook for hefty compensation payouts. How hefty? Well, Volkswagen’s Dieselgate affair perhaps provides some useful context: The German automaker has reportedly paid out $9.5 billion to U.S. customers after it was found it had sold them diesel-powered cars it knew did not actually meet U.S. emissions standards. Counting criminal and civil penalties, plus mandated emissions-mitigation costs and investments in EV infrastructure, Dieselgate is said to have cost Volkswagen upward of $25 billion.

Musk did offer a remedy. Of sorts. “For customers that have bought FSD,” he said, “what we’re offering is essentially a discounted trade-in for cars that have Hardware 4.” Musk said Tesla would also offer the ability to upgrade Hardware 3–equipped cars to Hardware 4. But there’s an expensive sting in the tail. “You also need to replace the cameras to go to Hardware 4,” Musk admitted. “To do this efficiently, we’re going to have to set up micro factories or small factories in major metropolitan areas to do it efficiently. I think over time it’s going to make sense for us to convert all Hardware 3 cars to Hardware 4, because that’s what enables them to enter the Robotaxi fleet and have unsupervised FSD.”

Given that Tesla built about 4 million Hardware 3–equipped cars from April 2019 through early 2023 and some early estimates suggest the conversion could cost $1,500–$2,000 per car, that’s a potentially huge hit to the company’s bottom line over the coming years. And Tesla isn’t exactly raking in money at the moment: Although 17 percent up on the same period last year, the company’s reported net income of $477 million for the first quarter of 2026 was the second worst quarterly profit number in five years. What’s more, it had 50,000 unsold cars in stock as the quarter ended.

Beyond that, though, is a bigger issue: Is Hardware 4 really going to be able to deliver genuine unsupervised Full Self-Driving? Since 2021, Musk has been convinced full autonomous driving can be achieved purely by using cameras and AI, to the point that radar and ultrasonic-sensor systems were no longer fitted to Tesla vehicles. And although some Hardware 4–equipped Model S and Model X cars were fitted with a different type of radar in 2023, Musk remains convinced the vision-only approach is the only way to achieve Level 4 or 5 autonomous driving capability at scale. He has long argued that because humans drive using only their eyes (biological cameras) and brain (a neural network), autonomous driving vehicles should be able to achieve full autonomy using only cameras and AI.

Most other automakers that are developing autonomous vehicles disagree. “Our philosophy is to have redundancies,” said Gregor Kugelmann, the senior development manager of Mercedes-Benz’s Drive Pilot, which in 2022 became the world’s first legally approved Level 3 autonomous driving system. Drive Pilot uses long-range radar, long-range lidar, and cameras to deliver Level 3 autonomous driving capability when certain operational conditions are met. It also uses acoustic data—a sensor that measures noise from spray in the front wheelwell determines how wet the road is, and the voice command microphones in the cabin pick up the sound of sirens from emergency vehicles approaching from behind—as well as data from the ultrasonic parking sensors.

Although Kugelmann, or any of the Mercedes engineers working on Drive Pilot, won’t say it outright, the subtext here is clear: Tesla’s strategy of relying solely on cameras and AI, with a little help from a front-facing radar unit, to enable a vehicle to safely drive autonomously is a strategy with a critical redundancy issue. To prove the point, Kugelmann’s team has developed tests for Benz’s autonomous drive technologies: driving behind a spray system that replicates a tropical downpour, driving toward a modified searchlight that replicates low-sun backlighting, scenarios that make objects, pedestrians, and animals virtually impossible to see, even for humans, let alone cameras.

“Each of Drive Pilot’s sensor systems have pros and cons,” Kugelmann said. “The combination of radar, lidar, and stereo camera technologies gives us the best and the safest sensor set needed for Level 3 autonomous driving.”

Hardware 4 might have the brain power Musk says it needs to deliver unsupervised FSD. But will it be the best, safest autonomous-driving technology on the market? And if it’s not, do you really want to sit in a car with no steering wheel or brake pedals and be driven by it?