Apple has confirmed that it uses Amazon Web Services (AWS) custom AI chips for essential functions like search. The company is also evaluating Amazon’s latest Trainium2 chip for pretraining its proprietary models, including Apple Intelligence.
At the AWS Reinvent conference on December 3, 2023, Apple’s senior director of machine learning and AI, Benoit Dupin, discussed Apple’s reliance on AWS for over a decade. He highlighted how AWS supports services like Siri, Apple Maps, and Apple Music. Dupin also revealed that Amazon’s Inferentia and Graviton chips have improved efficiency by 40%.
Apple is now testing the Trainium2 chip for pretraining AI models. Dupin noted that early evaluations indicate the chip could deliver up to a 50% efficiency improvement in pretraining, making it a promising alternative to traditional GPU-based training methods.
Apple’s collaboration with AWS highlights a growing partnership. AWS CEO Matt Garman shared that Apple had been an early adopter and beta tester for the Trainium chips. Garman also noted Apple’s vision for building its generative AI infrastructure using AWS technology.
Generative AI and Apple’s Approach
In fall 2023, Apple launched its first major generative AI product, Apple Intelligence. This service enables tasks like summarizing notifications, rewriting emails, and creating new emojis. Later this month, Apple Intelligence will integrate with OpenAI’s ChatGPT, with plans to enhance Siri in 2024, giving it more natural language capabilities and app control features.
Unlike many companies that rely on Nvidia GPUs for AI training, Apple uses a hybrid approach. Its devices, like iPhones and Macs, handle basic AI processing locally. Complex queries are sent to Apple-operated servers equipped with its proprietary M-series chips.
Industry Implications
Apple’s use of Amazon’s chips marks a significant endorsement of AWS’s capabilities as a cloud provider. It also shows the viability of alternatives to Nvidia GPUs for AI training. AWS’s Trainium2 chip, now available for rent, positions the company as a key competitor in the race for cost-effective AI solutions.
This collaboration could inspire other companies to explore non-Nvidia solutions for AI innovation. As AWS competes with Microsoft Azure and Google Cloud, partnerships like this solidify its role in the AI ecosystem, ensuring its place at the forefront of cloud-based AI development.
