Apple is leveraging custom artificial intelligence chips from Amazon Web Services (AWS) to process search queries and plans to use them for pre-training its Apple Intelligence AI models in the future.
Benoit Dupin, Apple’s director of machine learning and artificial intelligence, announced at AWS's re:Invent conference that the Trainium2 chip could provide up to a 50% increase in efficiency.
The collaboration between Apple and AWS spans several years, with Apple utilizing AWS's Graviton and Inferentia chips for services like Siri, Apple Maps, and Apple Music, achieving a 40% efficiency gain from these hardware resources.
However, the evaluation of the Trainium2 chip will only apply to pre-training models. User data processing and the operation of AI features will continue to rely on Apple's own chips and infrastructure, aiming to maintain privacy and security standards while enhancing efficiency.
Apple's interest in Amazon's chips aligns with its strategy to reduce costs and improve performance in training AI models. Previously known to use Google's Tensor chips, Apple aims to accelerate innovation by leveraging third-party technologies in its AI development process.