Apple used Google’s chips to train 2 AI models for Apple Intelligence

Apple’s (NASDAQ:AAPL) used chips designed by Alphabet’s (GOOG) (GOOGL) Google rather than that of Nvidia’s (NVDA) to develop two vital parts of its AI software infrastructure for its upcoming suite of AI features, according to an Apple research paper.

In the document, Apple detailed how two of its models —Apple Foundation Model, or AFM-on-device, a nearly 3 billion parameter language model, and AFM-server, a larger server-based language model — have been built and adapted to perform specialized tasks efficiently, accurately, and responsibly.

At its Worldwide Developers Conference 2024, Apple introduced Apple Intelligence, a personal intelligence system integrated into iOS 18, iPadOS 18, and macOS Sequoia.

Apple noted in the research paper that Apple Intelligence consists of multiple generative models which are fast and specialized for its users’ everyday tasks.

“The AFM models are pre-trained on v4 and v5p Cloud TPU clusters with the AXLearn framework, a JAX based deep learning library designed for the public cloud,” said Apple in the paper.

Tensor Processing Units, or TPUs, are Google’s custom-developed application-specific integrated circuits, or ASICs, used to accelerate machine learning workloads.

AFM-on-device was trained on one slice of 2048 TPUv5p chips, while it trained AFM-server from scratch for 6.3T tokens on 8192 TPUv4 chips.

Apple did not openly say that it did not use any Nvidia chips, however, the iPhone maker’s description of the hardware and software infrastructure of its AI features did not have any mention of Nvidia hardware, as per a report from Reuters.

Nvidia does not design TPUs but builds graphics processing units, or GPUs, which used widely for AI applications. Nvidia dominates about 80% of the AI chip market, while other players include Google and Amazon (AMZN).

Google sells access to TPUs via Google Cloud and customers must build software through Google’s cloud platform in order to use the chips. Meanwhile, Nvidia sells its chips and systems as standalone products, the report added.

Apple (AAPL) plans to give software developers access to Apple Intelligence for early testing as soon as this week through iOS 18.1 and iPadOS 18.1 betas, to fix bugs and ensure a smooth launch. However, the AI tool will be rolled out to customers by October, a few weeks after the release of the new iPhone and iPad software in September.