Apple has developed a new AI infrastructure using resources from Google’s cloud infrastructure. Their new AI backend includes a local foundation model, Apple servers, and server processors based on their M-series chip designs. This private cloud computing infrastructure hosts Apple servers running on local Apple chips, summarized as Apple Intelligence. The machine learning framework, AXLearn, showcases a hybrid AI approach combining Apple servers and Google Cloud capabilities. Apple has two new AI models, one for on-device AI and a larger one for servers. These models, developed with TensorFlow, were trained on Google TPUs. AXLearn supports the Bastion orchestrator, currently only supported by Google Cloud, but theoretically extendable to other cloud providers. This new AI hardware signifies Apple’s return to server hardware after discontinuing their last dedicated server product in 2011. Emphasizing privacy, Apple’s hosted AI models delete user data after queries and offer features such as image and document generation. Apple also partners with ChatGPT for stronger answers, with user consent messages to ensure data privacy.
Article Source
https://www.hpcwire.com/2024/06/18/apple-using-google-cloud-infrastructure-to-train-and-serve-ai/
