Amazon announced a new product on Tuesday called “AI Factories.” This service allows large corporations and governments to run Amazon’s AI systems within their own data centers. As described by AWS, customers provide the power and the data center facility, while AWS installs the AI system, manages it, and can connect it to other AWS cloud services.
The product is designed to cater to organizations concerned with data sovereignty, meaning they require absolute control over their data to prevent it from falling into the hands of competitors or foreign adversaries. An on-premises AI Factory means customer data is not sent to a model maker, and the hardware is not shared.
The product name may sound familiar because Nvidia uses the same term for its hardware systems packed with AI tools, from GPU chips to networking technology. This AWS AI Factory is indeed a collaboration with Nvidia.
The AWS Factory will use a combination of AWS and Nvidia technology. Deploying companies can choose between Nvidia’s latest Blackwell GPUs or Amazon’s new Trainium3 chip. The system utilizes AWS’s homegrown networking, storage, databases, and security, and can integrate with Amazon Bedrock for AI model selection and AWS SageMaker for model building and training.
AWS is not the only major cloud provider installing Nvidia AI Factories. In October, Microsoft showcased the first of many AI Factories being rolled out into its global data centers to run OpenAI workloads. At that time, Microsoft did not announce these systems for private clouds. Instead, it highlighted its use of Nvidia AI Factory technology to build and connect its new “AI Superfactories,” which are state-of-the-art data centers in Wisconsin and Georgia.
Last month, Microsoft also outlined plans for data centers and cloud services built in local countries to address data sovereignty concerns. Its options include “Azure Local,” which is Microsoft’s own managed hardware that can be installed on customer sites.
It is somewhat ironic that AI is causing the biggest cloud providers to invest heavily in corporate private data centers and hybrid clouds, reminiscent of the trends from over a decade ago.