Apple’s Private Cloud Compute AI infrastructure employs a hybrid approach, utilizing its own AI models and servers running on Apple chips, alongside Google Cloud Compute resources. HPCWire highlighted that this strategy signifies a major advancement for Apple. This marks Apple’s significant return to server hardware following the discontinuation of Xserve in 2011.
AXLearn ML Platform
Apple’s AXLearn ML platform showcases this hybrid approach, leveraging both Apple’s servers and Google Cloud capabilities. The platform utilizes the Google JAX framework and XLA compiler to efficiently train models across various hardware and cloud platforms, including Google’s TPU AI accelerators and local GPU accelerators. Apple has developed two new proprietary AI models: a smaller model with 3 billion parameters for on-device AI applications and a larger LLM for server operations.
These models were trained using Google’s TensorFlow framework on Google’s TPU. AXLearn features an orchestrator that currently operates exclusively with Google Cloud. Nonetheless, Apple notes that this orchestrator could theoretically be expanded to other cloud platforms in the future. Inference operations are carried out solely on Apple’s servers, where a temporary instance is created and irretrievably deleted after processing each user request. This process employs end-to-end encryption for secure data transfer.
Collaboration with OpenAI
At WWDC 2024, Apple announced its collaboration with OpenAI to integrate the ChatGPT chatbot into the Siri voice assistant and other iOS tools, reminds NIX Solutions. To address potential data leakage risks on third-party platforms, Apple devices will seek user permission before sending data to ChatGPT. We’ll keep you updated on any further developments regarding this collaboration and its implications for user privacy and AI advancements.