Apple used Google chips to train its AI

It turns out that Apple’s belated entry into the AI game, aka Apple Intelligence, was at least partially trained up using Google chips.

In its latest Machine Learning Research paper (via CNBC), titled Apple Intelligence Foundation Language Models, Apple has revealed some of the work that went into developing its forthcoming AI tools. The interesting thing is that rather than developing and using its own custom silicon for its AI, or even turning to market leader Nvidia, Apple hit up its age-old frenemy Google.

Cutting through the dry academic talk, the paper essentially reveals that Apple rented a whole bunch of “Cloud TPU clusters” – with TPU representing Google’s Tensor Processing Unit – in its AI training. This consisted of 2,048 TPUv5p chips for its on-device model, and 8,192 TPUv4 chips for its Apple Foundation Model server.

If you were being mischievous, you could say that the key feature of the iPhone 16 family – Apple Intelligence – will run on Pixel power, though Apple’s own silicon will come into play when it comes to actually running its pre-trained AI model.

Apple Intelligence recently arrived in the iOS 18.1 beta, meaning you can get a taste of the company’s AI work ahead of the full rollout in September. You’ll need at least an iPhone 15 Pro or an M1-powered iPad to run it, though.

Once you download the beta, you’ll be able to test Apple’s new AI writing tools and a redesigned version of Siri, which comes with a more personal style. Other Apple Intelligence features, such as Image Playground and Genmoji image generation, will come later.