Amazon Brings Its AI Tools to Your Data Center Instead of the Cloud

Amazon Web Services has announced a new offering called AI Factories, which lets companies and government agencies run AWS artificial intelligence systems in their own buildings instead of sending everything to the cloud. This is a significant shift for a company that has built its entire business around cloud computing.

The AI Factories program is built with NVIDIA, the graphics chip company. Together they will provide NVIDIA's latest high-powered processors — called Grace Blackwell chips — installed and managed by AWS right on your premises.

Why Now?

This move reflects a real tension in modern AI deployment. Many organizations need AI's power but face strict rules about where their data can physically sit. Banks handle sensitive financial records. Government agencies work with classified information. Factories need split-second response times that the internet can't reliably provide. For these organizations, sending data to a distant cloud feels risky or impractical, even if the cloud is technically more efficient.

It's worth flagging that this shift echoes a pattern we saw before. When cloud computing first emerged in the early 2010s, many large enterprises resisted moving critical systems entirely to public clouds. They preferred hybrid arrangements — some stuff on-premises, some in the cloud — to keep local control. AI deployment appears to be following the same curve, with vendors now recognizing that a "cloud only" approach may simply shut them out of entire sectors.

How It Works

Instead of uploading sensitive data to AWS, organizations install Grace Blackwell GPU systems in their own data center. AWS manages and updates the hardware, but the AI models run inside your building. You get the advanced AI capabilities AWS offers in its public cloud, but your data never leaves your premises.

The technical reality is complex. These GPU systems demand a lot of power — they generate heat, require specialized cooling, and need robust networking. Most data centers will need upgrades before they can support this equipment. And while the AI computing happens on-site, most organizations will still want to connect back to AWS services for training new models, processing other data, or integrating with wider business systems. That requires careful planning to avoid slowdowns.

The Bigger Picture

Amazon itself is spending almost $150 billion over 15 years on global data center expansion. The company now operates more than 900 data center facilities across more than 50 countries. It has announced new facilities in Saudi Arabia (launching 2026 with an investment exceeding $5.3 billion) and the UAE, which makes this on-premises offer look like part of a broader strategy: reach customers wherever they are, whether that's in a cloud data center or down the hall from their own IT team.

Other big cloud providers — Microsoft, Google — are making similar moves. The market is signaling that there is no single right answer for where AI should run. Some workloads belong in the cloud. Some belong on-premises. Most large organizations will likely use both.

The real test for AWS will be whether it can deliver the ease of use and integration that made cloud computing popular in the first place, while also addressing the control and compliance concerns that keep certain workloads locked on-site. For organizations with the expertise and the business need to manage a hybrid setup, AI Factories opens a practical path forward — serious AI capabilities without surrendering data governance.