Amazon is shaking up the enterprise AI world with its new on-premises AI Factories, a solution built for companies and governments that want full control over their data without giving up the power of cloud-grade AI.

With this model, customers provide their own power and data center space, while AWS installs and manages the entire AI system, a plug-and-play AI super-infrastructure right inside the customer’s walls. This setup helps organizations maintain data sovereignty, ensuring their sensitive information never leaves their premises or sits on shared hardware.

What makes this even more interesting is Amazon’s collaboration with Nvidia, the company already known for its own “AI Factories.” AWS’ version blends Nvidia’s latest GPUs (including Blackwell) with Amazon’s own Trainium3 chips, plus AWS networking, storage, security, and access to services like Bedrock and SageMaker.

Amazon isn’t alone. Microsoft is deploying Nvidia-powered AI Factories across its global data centers to run massive OpenAI workloads though not all are available as private on-prem systems yet. Still, Microsoft has its own local-cloud options, including Azure Local, to address sovereignty concerns.

What’s clear is this:

AI is pushing big cloud providers back into hybrid and on-prem solutions. The future of AI isn’t just in the cloud,  it’s in high-power systems that live closer to where sensitive data is created.

For startups watching these moves, one thing is certain: the new AI race is all about control, speed, and custom infrastructure.