Apple plans to use M2 Ultra chips in the cloud for AI

Illustration: The Verge

Apple plans to commencement its foray into generative AI by offloading analyzable queries to M2 Ultra chips moving successful information centers earlier moving to its much precocious M4 chips.

Bloomberg reports that Apple plans to enactment its M2 Ultra connected unreality servers to tally much analyzable AI queries, portion elemental tasks are processed connected devices. The Wall Street Journal previously reported that Apple wanted to marque customized chips to bring to information centers to guarantee information and privateness successful a task the work says is called Project ACDC, oregon Apple Chips successful Data Center. But the institution present believes its existing processors already person capable information and privateness components.

The chips volition beryllium deployed to Apple’s information centers and yet to servers tally by 3rd parties. Apple runs its ain servers crossed the United States and has been moving connected a new halfway successful Waukee, Iowa, which it archetypal announced successful 2017.

While Apple has not moved arsenic accelerated connected generative AI arsenic competitors similar Google, Meta, and Microsoft, the institution has been putting retired probe connected the technology. In December, Apple’s instrumentality learning probe squad released MLX, a instrumentality learning model that tin marque AI models tally efficiently connected Apple silicon. The institution has besides been releasing different probe astir AI models that hint astatine what AI could look similar connected its devices and however existing products, similar Siri, may get an upgrade.

Apple enactment a large accent connected AI show successful its announcement of the caller M4 chip, saying its caller neural motor is “an outrageously almighty spot for AI.”