Apple chooses Google's chips over NVIDIA for AI development
Apple has revealed that it uses Google's tensor processing unit (TPU) chips to train its artificial intelligence (AI) models. The information was disclosed in a research paper titled "Apple Intelligence Foundation Language Models." The paper indicates that Apple's Foundation Model (AFM) and the server technology supporting it were initially developed using "v4 and v5p Cloud TPU clusters" with Apple's software.
Apple sidelines NVIDIA's dominance in AI processors
Despite NVIDIA's dominance in the AI processors market with an approximate 80% share, Apple chose to depend on Google's cloud infrastructure. The tech giant used 2,048 of the TPUv5p chips to create the AI model for iPhones and other devices. For its server AI model, Apple utilized 8,192 TPUv4 processors. The company's engineers suggested in the paper that even larger and more complex models could be developed using Google's chips.
Apple's long-term AI plans remain unaffected
Apple's use of Google's hardware for initial training does not significantly affect its long-term plans. The company reportedly has its own hardware, derived from Apple Silicon, in its data centers to process Apple Intelligence queries. This is part of a project known as "Project ACDC," aimed at optimizing AI applications within its data centers.
Apple plans $5B investment in AI server enhancements
Apple plans to invest over $5 billion in AI server enhancements over the next two years. The objective is to match the technological capabilities of industry leaders like Microsoft and Meta by acquiring tens of thousands of AI server units, likely driven by Project ACDC. Additionally, Apple has acquired firms in Canada and France that specialize in compressing data used in AI queries sent to data centers.
