While Elon Musk has a track record of missing self-driving deadlines, Tesla’s latest bet on custom-made computer chips shows the company’s commitment to controlling its own technology. The AI5 chip, which was supposed to arrive in late 2026, won’t hit mass production until mid-2027.
Tesla’s AI5 chip, originally slated for late 2026, now won’t reach mass production until mid-2027, delaying autonomous vehicle plans by a full year.
That’s a full year later than originally planned, and it’s throwing Tesla’s self-driving roadmap into question. Tesla designed the AI5 chip to power its next generation of self-driving vehicles, including the Cybercab. The company planned to make two versions with help from manufacturing partners TSMC and Samsung.
The AI5 was supposed to be a major upgrade from existing hardware. Now, the delay means those plans have to be rescheduled. This isn’t Musk’s first missed deadline on autonomous vehicles. Back in 2015, he predicted “complete autonomy” by 2018.
In 2016, he pushed that to 2017. By 2020, he said the software would be “feature complete” by year’s end. Those promises didn’t come through. In 2021, he expressed “extreme confidence” in achieving full autonomy for customers that year. That didn’t happen either. Throughout these delays, Tesla’s FSD has remained regulated as a Level 2 ADAS system requiring continuous driver supervision.
Tesla’s been building its own computing power for years. The company developed custom chips designed for neural networks and self-driving software. In July 2025, Tesla announced a massive supercompute cluster called “Cortex” built with roughly 100,000 Nvidia GPUs in Texas. Early Cybercabs will likely deploy AI4 chip hardware due to the persistent AI5 production bottlenecks. Tesla also shifted focus from Dojo to Cortex as its primary training infrastructure for artificial intelligence applications.
The company also built Dojo, its own supercomputer designed to store video and train self-driving software. Tesla’s been shipping vehicles with HW4 hardware since January 2023, which represents the current SAE Level 2 automation standard that requires continuous driver supervision. The company’s fleet now includes over one million connected vehicles with self-driving sensors.
Current FSD software version 14.2.1 continues running on existing hardware while engineers work on AI5 integration. The delay matters because Tesla wants to use AI5 chips in the Cybercab, its driverless taxi vehicle. Without the chip ready on time, those plans have shifted.
Hardware 6, an even more powerful chip, is already in development. It’s expected to double AI5’s performance within 10 to 12 months after AI5 shipping begins. Tesla’s betting that custom chips will finally deliver on autonomous driving promises that’ve been delayed for years.