Tesla 'not about to replace Nvidia as EV maker develops chip

Tesla not about to replace Nvidia as EV maker develops chips for cars, robots, Musk says

Teslas forthcoming AI chip called the AI5 will be manufactured by Samsung in Texas, as well as by TSMC in Arizona.

Both foundries will make the chips because Tesla is shooting for excess production.

If we have too many AI5 chips for the cars and and robots, we can always put them in the data center, CEO Elon Musk said.

Teslas forthcoming artificial intelligence chip that its calling the AI5 will be manufactured by Samsung in Texas as well as by TSMC in Arizona, CEO Elon Musk told investors on Wednesday.

On the electric vehicle makers third-quarter earnings call, Musk said Tesla is shooting for excess production, and that any chips not used in cars or robots can be used in his companys data centers.

Our explicit goal is to have an oversupply of AI5 chips, Musk said.

Tesla formerly used Nvidias Drive chips in its vehicles, but dropped them in favor of its own processors in 2019. Musk said that Tesla will continue to use Nvidias graphics processing units, which dominate the AI market, for training its models.

Were not about to replace Nvidia, to be clear, but we do use both in combination, Musk said. Tesla announced on Wednesday that it has computing capacity equivalent to 81,000 of Nvidias H100 chips.

Musks comments offer a fresh glimpse into Teslas AI chip strategy after ex-Apple engineer Peter Bannon left the company earlier this year. Bannon previously helmed chip design for Tesla and was leading the development of Dojo, a supercomputer meant to help the company improve its driverless technology.

The AI5 chip was first announced in 2024 at Teslas shareholder meeting. Its the latest version of the car companys Autopilot hardware that processes signals needed for self-driving

Musk said at the time that the AI5 chip would be made by TSMC and that its successor, the AI6 chip, would be made by Samsung. On Wednesday, he said the AI5 chip would be made by both foundries at their U.S. facilities. He said it would fit on a half reticle, meaning the chip is half the size of full reticle AI chip designs from Nvidia and Advanced Micro Devices.

Hyperscalers like Google, Amazon and Microsoft are all investing in their own AI chips in order to offer an alternative to Nvidia, and because some AI experts say custom chips may be more cost-efficient or faster for certain tasks.

Aside from Tesla, only Apple currently designs its own silicon and uses it both in its products and in data centers for AI services.

Musk said that because Tesla is the only client for the AI5, its chip designers were able to focus on the companys needs and to eliminate older parts of the chip that slowed it down.

Tesla only has to satisfy requirements from one customer, Musk said. That that makes the design job radically easier and means we can delete a lot of complexity from the chip.

He said Teslas design team had removed the chips legacy GPUs, signal processors and other parts, and predicted that the chip could have the best performance per dollar for AI maybe by a factor of 10.

Nvidia has done an amazing job of dealing with almost an impossibly difficult set of requirements, he said. But in our case, we are waiting for radical simplicity.

Musks AI startup, xAI, which does business with Tesla, has become a major Nvidia customer. The company is currently building a massive supercomputing facility in Memphis, Tennessee, its second in the area, based around high-end Nvidia Grace Blackwell chips.