Insights into Tesla’s Next-Gen FSD Computer: AI5

As Tesla continues on its path towards the unified architecture of AI6 in the long term, the focus for now is on the upcoming HW5 / AI5 computer. This next-generation FSD computer is set to revolutionize autonomous driving with its significant performance boost over the current AI4 hardware.

Enhanced Performance

The AI5 computer is expected to deliver a massive leap in performance, estimated to be 3 to 5 times more powerful than the existing AI4. With a projected capability of 2,000 to 2,500 TOPS (trillion operations per second), the AI5 hardware surpasses even some of Nvidia’s high-end gaming GPUs in terms of processing power.

This enhanced performance will enable more complex neural networks, allowing Tesla vehicles to process data from cameras and sensors at a faster rate. It will also improve decision-making capabilities, making autonomous driving safer and more efficient.

Manufacturing and Release Date

Tesla has finalized the design of the AI5 chip and will be partnering with TSMC for production. Initial manufacturing will take place in Taiwan, with future production planned at TSMC’s new fab in Arizona. The release date for AI5 is expected to be late 2026 or early 2027.

First Use Case

While mass production of AI5 is still a year or more away, Tesla will be utilizing the chips for training and deployment in the Cybercab. The Cybercab, a driverless vehicle designed for Robotaxi service, will benefit greatly from the increased computational power of AI5.

Should You Wait for AI5?

For potential Tesla buyers considering waiting for AI5, it’s important to note that the current AI4 hardware is already highly capable and used in Tesla’s Robotaxi fleet. While AI5 promises significant improvements, it may take some time before these enhancements are fully realized in consumer vehicles.

Ultimately, AI5 represents a key step towards Tesla’s goal of achieving full autonomy. It will pave the way for future advancements such as the AI6 architecture and the development of homegrown Dojo chips.