Tesla has officially unveiled its next-generation autonomous driving (FSD) chip, the AI5, which takes its AI hardware ambitions to the next level. This new chip, announced last month by Elon Musk and detailed in the company’s third-quarter financial presentation, will be used in both Tesla vehicles and its humanoid robot, Optimus.

Tesla Unveils AI5 Chip

With the AI5, Tesla delivers a 40x performance increase over its predecessor, the AI4. The company describes this chip as a “unique design” developed based on a deep understanding of hardware and software requirements.

Elon Musk stated that the performance increase isn’t limited to raw processing power; it’s also supported by memory capacity, data transfer bandwidth, and computational optimizations. Compared to its predecessor, the AI5 offers an 8x increase in raw processing power, a 9x increase in memory capacity, and a 5x increase in memory bandwidth.

This includes a significant leap in memory capacity, which is particularly critical for AI applications. The previous HW4 chip featured 16GB of RAM. Tesla plans to use up to 144GB of RAM with the AI5.

This higher capacity allows the AI5 to run complex AI models more efficiently and provides greater scalability in data centers. This dramatic performance increase is driven by optimizing SoftMax operations, which were a bottleneck in the AI4. These operations, which took 40 steps in the AI4, are now completed in just a few steps in the AI5.

The AI5 can also dynamically manage mixed-precision models and natively supports sparse tensor operations optimized for real-world workloads. By eliminating the traditional GPU and image processing units (ISP), the chip is poised to function as a standalone GPU, delivering the highest performance per watt, according to Musk.