 "AWS’s new servers are set to launch next year, alongside NVIDIA's latest offerings, as both companies race to meet booming AI demands. For Apple, this collaboration with Amazon reflects a pragmatic approach to stay competitive while adhering to its commitment to privacy and innovation. Image Credit : Reuters")
Apple has confirmed it will utilise Amazon Web Services’ (AWS) advanced AI chips to power the training of its upcoming Apple Intelligence models. This marks a significant step for Apple as it collaborates with Amazon’s cloud services to enhance its machine learning capabilities, leveraging the efficiency of Amazon’s custom-built Trainium chips.
The announcement, made during AWS’s Reinvent 2024 conference, highlights Apple’s growing reliance on external providers for tasks requiring immense computational power. While the move might surprise those familiar with Apple’s privacy-focused approach, it aligns with the company’s strategy to maximise efficiency while maintaining data security.
Efficiency gains with Amazon’s Trainium Chips
Apple has been using Amazon’s Trainium and Graviton chips to improve search functionality, achieving a 40 per cent boost in performance. Now, the company is evaluating Amazon’s Trainium2 chips for pre-training new AI models. Early results indicate up to a 50 per cent increase in efficiency, which could lower costs or enable more extensive model training.
The chips are part of AWS’s Trn2 UltraServers, a next-generation supercomputing solution designed for demanding AI workloads. The servers will feature Amazon’s Trainium2 chips, interconnected to handle massive generative AI tasks at a lower cost compared to NVIDIA’s current offerings. AWS claims that some AI models trained on Trainium2 chips can achieve a 40 per cent cost reduction compared to NVIDIA chips.
Balancing privacy and outsourcing
Apple’s use of AWS’s AI chips is focused solely on pre-training models — a resource-intensive process that doesn’t involve user data. By outsourcing this step, Apple avoids the expense of building its own high-performance computing infrastructure while ensuring privacy standards remain intact. On-device data processing and other privacy-sensitive operations continue to rely on Apple’s proprietary hardware.
This isn’t Apple’s first collaboration with external providers. Earlier this year, reports revealed that Apple used Google-designed hardware to train its Foundation Model. By partnering with AWS and exploring Trainium2 chips, Apple is diversifying its AI development strategy to keep pace with growing demand.
The race against NVIDIA
Amazon’s push with Trainium chips comes as NVIDIA dominates the AI chip market with a 70 per cent share. However, AWS is positioning its Trainium2 and upcoming Trainium3 chips as cost-effective alternatives for high-performance AI workloads. With AWS claiming superior scalability and reduced training costs, the competition between these tech giants is heating up.
AWS’s new servers are set to launch next year, alongside NVIDIA’s latest offerings, as both companies race to meet booming AI demands. For Apple, this collaboration with Amazon reflects a pragmatic approach to stay competitive while adhering to its commitment to privacy and innovation.