Amazon is launching AWS AI Factories, allowing customers to run AWS AI infrastructure within their own data centres. This offering combines AWS Trainium accelerators and Nvidia GPUs, along with networking, storage, and AI services. AI Factories aims to speed up AI buildouts, potentially reducing the time required by months or years.
AWS is integrating Nvidia's NVLink Fusion into its Trainium4 chips to simplify deployment and systems management. The collaboration allows AWS to provide access to advanced AI services, enabling the deployment and training of large models while maintaining control of data. New servers based on the Trainium3 chip are also being rolled out, featuring 144 chips per server and offering increased computing power with reduced energy consumption.
AI Factories provides access to the Nvidia accelerated computing platform, Nvidia AI software, and GPU-accelerated applications within customer data centres. The AWS Nitro System, along with networking and Amazon EC2 UltraClusters, supports the Nvidia Grace Blackwell and Vera Rubin platforms. This setup operates as a private AWS Region within customer facilities, providing compute, storage, database, and AI services.
