AWS unveils on-site Nvidia clusters to lock in enterprise AI workloads

Amazon Web Services (AWS) has recently announced a product named "AI Factories". The system is designed to allow large corporations and governments to operate AWS's AI systems within their own data centers. AWS states the model requires customers to supply power and the data center facility, while AWS installs and manages the AI system, providing connections to other AWS cloud services.

The offering is intended to address concerns regarding data sovereignty, providing complete customer control over their data to prevent its exposure to competitors or foreign entities. Deploying an on-premise AI Factory means customer data does not travel to a model maker, and hardware is not shared.

The product name, AI Factories, matches the terminology used by Nvidia for its comprehensive hardware systems that run AI, including GPU chips and networking technology. AWS confirmed this AI Factory is a joint effort with Nvidia.

The AWS Factory will integrate AWS and Nvidia technology. Customers can choose between Nvidia's Blackwell GPUs or Amazon's new Trainium3 chip for their deployments. The system utilizes AWS networking, storage, databases, and security components and offers access to Amazon Bedrock for AI model selection and management, and AWS SageMaker AI for model training.

AWS is not the sole major cloud provider installing Nvidia AI Factories. Microsoft disclosed its initial AI Factories in October, which are being deployed in its global data centers to manage OpenAI workloads. At the time, Microsoft did not specify if these systems would be available for private clouds but highlighted its reliance on Nvidia AI Factory data center technology for its new "AI Superfactories," which are advanced data centers currently under construction in Wisconsin and Georgia.

Microsoft last month detailed data centers and cloud services to be established in local countries to address data sovereignty. Microsoft also offers "Azure Local," which is its own managed hardware solution for installation at customer sites.

The trend underscores an unexpected consequence of AI development, driving major cloud providers to significantly increase investment in corporate private data centers and hybrid clouds, mirroring earlier market strategies.