Amazon’s AI Chips Power Apple Services

Apple has revealed its use of Amazon Web Services' (AWS) custom artificial intelligence (AI) chips for services such as search, and it is exploring the potential of Amazon’s Trainium2 chip for pretraining its proprietary AI models like Apple Intelligence. This announcement came during Apple’s rare public acknowledgment at the AWS Reinvent conference.

Strengthening Collaboration with AWS

Benoit Dupin, Apple’s Senior Director of Machine Learning and AI, highlighted the decade-long partnership between Apple and AWS, emphasizing the reliability and global reach of Amazon’s cloud infrastructure. “We have a strong relationship, and the infrastructure is both reliable and able to serve our customers worldwide,” Dupin stated. This collaboration marks a notable endorsement of AWS as it competes with Microsoft Azure and Google Cloud for AI-driven business.

Apple uses AWS for multiple services, including Siri, Apple Maps, and Apple Music. Specifically, the company has implemented Amazon’s Inferentia and Graviton chips, which have delivered a 40% improvement in efficiency for its search services. The efficiency gains and cost-effectiveness have been key drivers for this collaboration.

Adopting Trainium2 for AI Pretraining

Apple is now evaluating Amazon’s recently launched Trainium2 chip for pretraining its AI models. According to Dupin, early tests indicate up to a 50% improvement in efficiency compared to existing methods. This aligns with Apple’s vision to expand its AI capabilities, including its proprietary Apple Intelligence platform.

AWS CEO Matt Garman shared insights on this partnership, noting that Apple was an early adopter of Trainium chips and sought AWS’s expertise to build the infrastructure for its Generative AI initiatives.

Challenging Nvidia’s Dominance

The majority of AI training today relies on Nvidia’s graphics processors, which are known for their high costs. Cloud providers, including AWS, are developing alternatives to Nvidia’s GPUs, aiming for more cost-effective and efficient solutions. Apple’s adoption of custom chips like Trainium2 could signal a shift for other companies considering alternatives to Nvidia’s hardware.

AWS also announced plans to offer Nvidia’s Blackwell-based AI servers for rent, further broadening the options available to businesses investing in AI.

Apple’s Generative AI Progress

This fall, Apple released its first major generative AI product, Apple Intelligence. The platform includes features that summarize notifications, rewrite emails, and generate emojis. Apple plans to integrate these capabilities with OpenAI’s ChatGPT and introduce enhanced abilities for Siri, such as app control and natural language responses, in 2024.

Unlike other leading AI platforms, which rely heavily on Nvidia-based cloud clusters, Apple’s approach is hybrid. It processes as much data as possible on-device using its iPhone, iPad, and Mac chips, while offloading more complex queries to Apple-operated servers powered by its M-series chips.

The Future of AI Collaboration

Apple’s collaboration with AWS and its adoption of Trainium2 reflect a growing trend of companies exploring alternatives to traditional AI hardware solutions. As custom AI chips like Amazon’s gain traction, the industry could see a shift toward more efficient and cost-effective AI processing methods. Apple’s endorsement of these technologies further solidifies their potential to transform AI development and deployment.