Amazon Web Services (AWS) has introduced AWS AI Factories, a new service that provides enterprises and government organisations with dedicated AI infrastructure installed directly in their own data centres.

The new offering is designed to address the needs of organisations with sovereignty and compliance requirements by enabling them to develop and deploy AI applications using AWS technology on-premises.

AWS AI Factories include AI accelerators such as Nvidia AI computing and Trainium chips, as well as AWS networking, storage, databases, and security features.

Customers can access AWS AI services, including Amazon Bedrock and SageMaker, within their own facilities.

By deploying this infrastructure in existing data centres, AWS aims to help organisations avoid the complexity and capital expenditure associated with building large-scale AI systems independently.

The service functions as a private AWS Region, providing secure and low-latency access to compute, storage, and AI services.

This setup allows organisations to use their current data centre resources and meet requirements for data sovereignty and regulation.

AWS manages the infrastructure and offers access to foundation models and AI tools without the need for separate contracts with model providers.

AWS and Nvidia have expanded their partnership to support these deployments.

The collaboration allows AWS customers to use Nvdia’s accelerated computing platform, AI software, and graphics processing unit (GPU) optimised applications within their own data centres.

AWS infrastructure, including the Nitro System and Elastic Fabric Adapter networking, supports the latest Nvidia Grace Blackwell and future Vera Rubin platforms.

AWS also plans to support Nvidia NVLink Fusion technology in upcoming Trainium4 and Graviton chips, as well as in the Nitro System.

AWS AI Factories are built to meet AWS security standards and are intended to support government workloads at all classification levels.

AWS reported that the service will provide governments with the availability, reliability, security, and control needed to advance AI technologies.

In Saudi Arabia, AWS and Nvidia are working with HUMAIN to create an “AI Zone” that will feature up to 150,000 AI chips, including GB300 GPUs, AWS AI infrastructure, and AWS AI services within a HUMAIN-operated data centre.

Nvidia hyperscale and HPC general manager and vice president Ian Buck said: “Large-scale AI requires a full-stack approach—from advanced GPUs and networking to software and services that optimise every layer of the data centre. Together with AWS, we’re delivering all of this directly into customers’ environments.”

Separately, AWS has announced the general availability of Amazon EC2 Trn3 UltraServers, powered by the Trainium3 chip built on 3nm technology.

Each Trn3 UltraServer contains up to 144 Trainium3 chips, delivering up to 4.4 times more compute performance than the previous generation.

The Trainium3 chip features design improvements, including optimised interconnects and enhanced memory systems, and is said to deliver 40% better energy efficiency than earlier models.