Apple Leveraged Google Hardware for Apple Intelligence Model Training

July 30, 2024 – Apple revealed in a technical paper released on Monday that the two artificial intelligence models supporting its AI system, Apple Intelligence, underwent pre-training on cloud chips designed by Google.

The paper, titled “Apple Intelligence Foundation Language Models,” provides an in-depth exploration of the model’s construction process and data sources. A section of the paper suggests that Apple may have utilized Google’s hardware during the initial stages. Specifically, it mentions that Apple’s Foundation Model (AFM) and its underlying server technology were originally built on “v4 and v5p cloud TPU clusters,” utilizing proprietary Apple software.

While some reports have speculated that Apple leased Google’s cloud clusters, the paper does not explicitly mention Google or NVIDIA. It is more likely that Apple directly purchased Google’s hardware and deployed it within its own data centers.

It’s important to note that despite the initial training on Google hardware, this does not alter Apple’s long-term plans. Reports indicate that Apple has already deployed hardware based on its own Apple Silicon chips within its data centers to handle Apple Intelligence queries. Additionally, the company has initiated a project codenamed “Project ACDC,” aimed at optimizing AI applications within the data centers.

In a bid to catch up with giants like Microsoft and Meta in the AI field, Apple plans to invest over $5 billion in upgrading its AI servers over the next two years. This includes purchasing tens of thousands of AI servers. Furthermore, Apple has acquired two companies, one in Canada and the other in France, both specializing in compressing the data required for AI queries.

Leave a Reply