Apple's recent publication of its artificial intelligence research, titled "Apple Intelligence Foundation Language Models," reveals an intriguing aspect of its development process. The highly technical paper details the origins and development of Apple's language model, known as the Apple Foundation Model (AFM). Buried within the paper is a noteworthy detail: Apple initially used Google hardware for training its AI.
According to the research, Apple's AFM and the supporting server technology were initially built on "v4 and v5p Cloud TPU clusters" using Apple software. This suggests that during the early stages of development, Apple leveraged Google's Tensor Processing Units (TPUs) to train its language models.
While the paper does not explicitly mention Google or Nvidia, a CNBC report on Monday suggested that Apple may have rented time on existing Google-hosted clusters. However, the research itself does not support this claim. It is more plausible that Apple purchased the Google-designed hardware and used it within its own data centers.
Despite the initial reliance on Google hardware, this does not have significant long-term implications. Apple is reportedly developing its own hardware, derived from Apple Silicon, to process Apple Intelligence queries. This initiative is part of a larger project, reportedly called "Project ACDC," aimed at optimizing AI applications within Apple's data centers.
Apple's commitment to artificial intelligence is substantial, with plans to invest over $5 billion in AI server enhancements over the next two years. This investment is aimed at acquiring tens of thousands of AI server units to match the technological capabilities of industry leaders like Microsoft and Meta.
Additionally, Apple has been expanding its AI capabilities through strategic acquisitions. The company has acquired firms in Canada and France that specialize in compressing data used in AI queries to data centers, further bolstering its AI infrastructure.