Apple Leverages Amazon AWS Custom AI Chips to Boost Search and Train 'Apple Intelligence' Models

Apple Leverages Amazon AWS Custom AI Chips to Boost Search and Train 'Apple Intelligence' ModelsApple is actively using Amazon Web Services (AWS) custom artificial intelligence (AI) chips to enhance the performance of its search engine and other services, and is evaluating Amazon's latest AI chips for pre-training large language models like "Apple Intelligence," according to a revelation at Amazon's annual AWS re:Invent conference on Tuesday. This rare public acknowledgment of Apple's reliance on AWS marks a significant deepening of collaboration between the two tech giants

Apple Leverages Amazon AWS Custom AI Chips to Boost Search and Train 'Apple Intelligence' Models

Apple is actively using Amazon Web Services (AWS) custom artificial intelligence (AI) chips to enhance the performance of its search engine and other services, and is evaluating Amazon's latest AI chips for pre-training large language models like "Apple Intelligence," according to a revelation at Amazon's annual AWS re:Invent conference on Tuesday. This rare public acknowledgment of Apple's reliance on AWS marks a significant deepening of collaboration between the two tech giants.

Benoit Dupin, Apple's senior director of machine learning and AI, personally showcased Apple's utilization of AWS cloud services at the conference. He highlighted the "rock-solid" partnership between Apple and Amazon, emphasizing the reliability, stability, and scalability of AWS's infrastructure to meet the global demands of Apple's customers. Apple's public endorsement of Amazon and its chips provides strong backing for the cloud giant, particularly amid intense competition in the AI realm with Microsoft Azure and Google Cloud.

Dupin's presentation revealed Apple's long-standing dependence on AWS, noting that Apple has leveraged AWS to support services like Siri, Apple Maps, and Apple Music for over a decade. Specifically, Apple has deployed Amazon's Inferentia and Graviton chips to optimize its search services, achieving up to a 40% efficiency improvement. This demonstrates the cost-effectiveness of Amazon's custom chips in AI inference.

However, the collaboration extends beyond search optimization. Dupin also stated that Apple plans to use Amazon's latest Trainium 2 chips for pre-training its proprietary models. This indicates that Amazon's chips excel not only in inference but also in training new AI applications. With Amazon announcing the general availability of Trainium 2 chips for rent, Dupin expressed optimism, projecting a potential 50% efficiency boost in the early evaluation phase of Trainium 2.

Matt Garman, AWS CEO, confirmed in an interview that Apple is an early adopter and testing partner for AWS Trainium chips. He noted that Apple proactively sought out AWS, outlining its strategic needs in generative AInamely, building the infrastructure for "Apple Intelligence."

Apple's ambitions for "Apple Intelligence" are clearly evident. Earlier this year, an Apple research paper revealed the company's use of Google Cloud's TPU chips for training AI services on iPhones, including "Apple Intelligence." While AI training has traditionally relied on expensive NVIDIA GPUs, the emergence of cost-effective alternatives from cloud providers and startups, along with more efficient processing pathways, makes Apple's adoption of Amazon's custom chips a significant development, showcasing a viable non-NVIDIA training solution.

It's noteworthy that AWS is expected to unveil further details at the conference regarding its offering of AI server rentals based on NVIDIA's next-generation Blackwell chips. This underscores the fierce competition in the AI cloud space and the ongoing demand for high-performance computing resources.

This fall, Apple launched its first major generative AI product, "Apple Intelligence," encompassing services like summarizing notifications, rewriting emails, and generating new emojis. Apple announced that "Apple Intelligence" will integrate with OpenAI's ChatGPT later this month, and Siri's capabilities will be significantly enhanced next year, including app control and more natural conversations.

Unlike leading chatbots like OpenAI's ChatGPT, which rely on large NVIDIA-based cloud server clusters, Apple's AI strategy prioritizes performing as much processing as possible on the chips within iPhones, iPads, or Macs, sending only complex queries to its own M-series chips on Apple-operated servers. This reflects Apple's unique approach to balancing cloud and on-device computing.

Apple's choice of AWS custom AI chips and its public acknowledgment of this partnership not only demonstrate its confidence in Amazon's technology but also reflect Apple's strategic direction of continuous innovation in the AI field. Apple's reliance on AWS, along with its positive evaluation of Amazon's custom chips, points towards a deeper future collaboration between the two companies in AI, opening up new possibilities for the field. Apple's AI strategy, incorporating both reliance on existing cloud providers and continued investment in its own chip development, allows it to balance technological advancement and cost control, ultimately delivering a better AI experience for its users. This collaboration with Amazon represents a significant step for Apple in expanding its technological partnerships and seeking broader support in its AI endeavors. Apple's path of AI innovation deserves continued observation.