Apple has revealed that they used Google’s custom chips to train its AI models that underpin Apple Intelligence. In a research paper published by the hardware company on Monday, they shared that two key parts of its AI system were pretrained on processors designed by Google. The move is rare considering the dependence that Big Tech companies have had on AI chipmaker Nvidia.
The company stated that it used two different versions of Google’s homegrown tensor processing units (TPUs) that are organised in large clusters of chips. To train the AI model for iPhones and other devices, Apple used 2,048 of the TPUv5p chips and for its server AI model, Apple used 8,192 TPUv4 processors.
Besides, the iPhone maker also released a preview of its awaited Apple Intelligence system.
Currently, Nvidia owns around 80% of the AI chip market with expensive graphics processing units (GPUs) that have been in high demand since the AI boom.
(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)
More recently, several tech companies including Google, Meta, Microsoft and Amazon have all resorted to making their own AI chips given the cost of AI infrastructure training massive AI models requires.
Google’s TPUs can be accessed through their Google Cloud Platform meaning users have to also build the software on Google Cloud if they want to use the chips.
The research paper also said that it would be possible to build even larger and more complex models with Google chips aside from the ones mentioned.
Apple has announced a host of new AI tools and features at its WWDC conference in June including the inclusion of OpenAI’s ChatGPT onto its software. They will be rolling out selected Apple Intelligence features for its beta users within this week.
