Apple made a surprising appearance Tuesday at Amazon Web Services’ annual re:Invent conference, revealing extensive use of AWS’s custom chips and infrastructure for its AI development, including the recently launched Apple Intelligence features .
Key points:
- Apple uses AWS’s custom AI chips for its new Apple Intelligence features, achieving efficiency gains of more than 40%
- The iPhone maker plans to evaluate AWS’s new Trainium 2 chips, hoping for up to a 50% improvement in training speed.
- Apple’s approval comes as AWS competes with Microsoft Azure and Google Cloud for AI infrastructure dominance.
Benoit Dupin, Apple’s senior director of machine learning and AI, took the stage in Las Vegas to detail how the usually secretive iPhone maker leverages AWS technology. “We have a strong relationship and the infrastructure is both reliable and capable of serving our customers around the world,” Dupin said.
This approval marks an important moment for AWS in its competition with Microsoft Azure and Google Cloud for dominance in AI infrastructure. Apple has used AWS for over a decade for services like Siri, Apple Maps, and Apple Music, but this public acknowledgment of their AI collaboration is unprecedented.
Apple has achieved notable efficiency gains using custom silicon from AWS. The company reported more than 40% improvement in efficiency for machine learning workloads using AWS’s Graviton and Inferentia chips compared to traditional x86 instances. Looking ahead, Apple is evaluating AWS’s recently announced Trainium 2 chips, with early testing suggesting improvements of up to 50% in model training efficiency.
AWS CEO Matt Garman highlighted Apple’s role as an early adopter of its AI infrastructure. “Apple came to us and said, ‘How can you help us with our generative AI capabilities, we need infrastructure to be able to build,’ and they had this vision to build Apple Intelligence,” he said. explained Garman.
As Apple continues to prioritize on-device AI processing for its products, this partnership suggests a hybrid approach to AI development. The company processes simpler queries on iPhone, iPad or Mac chips, but relies on cloud infrastructure for more complex AI tasks and model training.
The move comes as cloud providers scramble to develop alternatives to Nvidia’s dominant AI chips. Apple’s approval of AWS’s custom silicon could signal to other companies that non-Nvidia approaches may be viable for AI development and deployment.
AWS also announced that Trainium 3, its next generation of AI chips, will debut next year, demonstrating its commitment to competing in the AI infrastructure space. The cloud provider is also expected to announce new details on leasing NVIDIA Blackwell-based AI servers.
This public endorsement from Apple, known for its selective partnerships and secretive nature, represents a significant victory for AWS as it competes for AI infrastructure spending. However, Apple has relationships with other cloud providers, including Google Cloud, which it previously used for training iPhone AI services.