The ambition for total autonomy is colliding with the stark realities of road safety. As Full Self-Driving (FSD) systems scale, the resulting surge in incidents is casting serious doubt on the reliability of existing algorithms. Central to this crisis is a fundamental disconnect between software aspirations and the physical constraints of the underlying hardware. The industry is now witnessing a moment where technological optimism hits the hard ceiling of hardware limitations.
Tesla’s pursuit of full autonomy is increasingly resembling a war of attrition against its own statistics. Despite efforts to keep road incidents out of the spotlight, regular filings with the National Highway Traffic Safety Administration (NHTSA) reveal a troubling trend. The frequency of accidents involving Full Self-Driving (FSD) is climbing steadily, with one month alone recording a staggering peak of 207 crashes.
To grasp the magnitude of the issue, one only needs to look at the retrospective data. Throughout the entirety of 2021, FSD-related accidents totaled fewer than 157; today, that figure is surpassed in a matter of weeks. In total, Tesla has logged 3,763 incidents across the United States—accounting for approximately 85% of all recorded accidents among companies deploying autonomous technologies nationwide.
This surge in accidents appears to be a logical consequence of aggressive expansion. The first half of this year saw 826 incidents, representing a 73% year-over-year increase in frequency. The primary drivers are the expanding active fleet and an increase in total mileage. FSD is becoming mainstream: last quarter, subscriptions for the feature accompanied 55% of all new vehicle sales. Furthermore, across four European countries where the technology has been approved, cumulative autopilot mileage since July has exceeded 50 million kilometers. The more data the system processes and the more miles it covers, the more frequently it encounters critical automation failures.
Yet, beneath the statistics lies a deeper, systemic crisis: hardware. For years, Tesla leadership maintained that the computing power of hardware released since 2016 would suffice for full autonomy via over-the-air software updates. It has now become evident that third-generation hardware (HW3) has hit its ceiling.
Acknowledging that HW3 cannot support a fully autonomous mode leaves the company facing a precarious choice. The ideal solution would be a mass upgrade of onboard computers for all pre-HW4 vehicles, but such an operation would impact roughly 4 million cars, incurring colossal financial costs. Consequently, owners of older models are left with truncated software versions that simply cannot unlock the full potential of the underlying algorithms.
The situation is further complicated by the possibility that even current HW4 hardware may be insufficient for the final leap toward autonomy. The company’s gaze is now fixed on the next generation of processors, slated for production by mid-next year. Even more promising is the AI5 chip, with mass production expected by the end of next year.
Notably, resource allocation priorities are shifting: the first batches of AI5 will be diverted not to vehicles, but to the Optimus humanoid robots. This implies that owners of legacy electric vehicles will find themselves at the back of the queue for modernization, rendering any tangible hardware upgrades a distant prospect.
Apple’s Comprehensive Overhaul of the Mac Lineup
