Apple is turning to Amazon Web Services (AWS) to enhance the training of its AI models, including potential updates for its Apple Intelligence platform. The announcement was made at the annual AWS Reinvent conference, where Apple’s senior director of machine learning and AI, Benoit Dupin, shared details about this collaboration.
Why Amazon Chips?
Apple has been using AWS for over a decade to power services like Siri, Apple Maps, and Apple Music. Recently, Apple integrated Amazon’s Trainium and Graviton chips to handle search queries, resulting in 40% greater efficiency.
Dupin also revealed Apple is evaluating Amazon's Trainium2 chip, which could deliver up to 50% improved efficiency in pretraining AI models. This improvement could reduce costs or allow for more extensive training within the same budget, making it a promising avenue for future Apple Intelligence models.
Training vs. Processing
Apple’s use of AWS chips focuses solely on training AI models, a resource-intensive step that does not involve user data. For actual query processing and customer-facing tasks, Apple continues to use its proprietary hardware with on-device and Private Cloud Compute capabilities, maintaining its commitment to privacy and security.
Expanding Beyond AWS
This move echoes a similar strategy Apple employed in July 2024, when it used Google-designed hardware to build the Apple Foundation Model. While details on the arrangement remain unclear, it underscores Apple’s readiness to adopt third-party solutions for efficiency gains in AI training.
What It Means for Consumers
The collaboration with AWS chips is unlikely to impact Apple's privacy-centric reputation. By outsourcing training tasks while keeping real-time data processing in-house, Apple ensures user security remains intact.
As Apple continues to refine its AI models using external hardware, the focus stays on delivering improved services and innovative features to its users.
