Report: Apple deploying servers with custom chips to power upcoming iOS update – Business

Apple Inc. is reportedly equipping its data centers with internally designed chips, including a processor originally developed for Macs, to support the rollout of an upcoming iOS update.

Bloomberg detailed the initiative today, citing people familiar with the matter. The chip rollout is said to be part of a project known internally as Apple Chips in Data Centers, or ACDC for short. It’s believed that the infrastructure upgrade has so far cost the iPhone maker hundreds of millions of dollars.

At its upcoming WWDC developer conference in June, Apple is expected to introduce a new version of iOS with an array of generative artificial intelligence features. Some of those capabilities will run locally on iPhones’ processors. The remaining features, in turn, will reportedly be powered by AI models deployed inside Apple data centers.

To ensure those neural networks will run efficiently, the company is reportedly equipping its data centers with servers that feature internally-developed chips. The effort is currently said to revolve around the M2 Ultra processor that Apple introduced for high-end Macs last year. It’s a system-on-chip with 134 billion transistors, about 50 billion more than an Nvidia Corp. H100 graphics card.

The M2 Ultra includes a compute module known as the Neural Engine that is specifically designed to run AI models. According to Apple, the module features 32 cores that can perform 31.6 trillion computations per second.

In the future, the company is also expected deploy servers equipped with versions of the M4, an iPad system-on-chip it debuted this week. The M4 features 28 billion transistors, or 102 billion less than the M2 Ultra, yet it features a faster Neural Engine. The module is capable of performing 38 trillion computations per second when running AI models, or about 60 times more than Apple’s first Neural Engine.

The M4’s fast AI performance partly stems from the fact that it’s based on a newer manufacturing process. Apple is producing the chip using a three-nanometer node, whereas the M2 Ultra is powered by five-nanometer technology. 

One of the new generative AI features expected to roll out for iOS is an enhanced version of the built-in Spotlight search bar. According to Bloomberg’s sources, the tool will gain the ability to answer complex questions using large language models. Additionally, it will reportedly be capable of performing actions in the apps installed on the user’s iPhone.

Apple Music and the Keynote presentation creation app are likewise expected to receive generative AI features. Moreover, Apple’s machine learning push is said to extend beyond iOS. The company will reportedly add an AI coding assistant to Xcode, a macOS application that developers use to write software for its operating systems.

In the long term, Apple doesn’t plan to rely solely on internal data centers to power its devices’ generative AI features. The company is expected to offload some workloads to “outside facilities.” Apple has taken a similar hybrid cloud approach with several existing services including its iCloud file storage platform. 

Image: Apple