AWS AI Factories: Bringing Full Cloud AI Infrastructure On-Prem for Data Sovereignty

According to Ian Buck, NVIDIA’s Vice President of Hyperscale and HPC, AWS unveiled its AI Factories service at re:Invent 2025, bringing AWS’s full AI infrastructure—including NVIDIA accelerated computing platforms and Trainium chips—directly into customers’ on-premises data centres to create high-performance, governance-aligned AI compute environments.

The offering functions as a Private AWS Region, letting customers use AWS-managed services and model catalogs in their own facilities while complying with data-localisation requirements. AWS is deepening its collaboration with NVIDIA, with AI Factories integrating the latest NVIDIA Grace Blackwell architecture and the forthcoming NVIDIA Vera Rubin platform, and Trainium4 chips set to support NVIDIA NVLink Fusion for tighter interoperability with NVIDIA GPUs and Graviton processors.

In addition, the launch includes the Amazon EC2 P6e-GB300 UltraServers powered by the NVIDIA GB300 NVL72 system, designed for massive inference workloads, with the first deployment for HUMAIN backed by Saudi Arabia’s Public Investment Fund to establish the country’s first “AI Zone” deploying up to 150,000 AI chips. HUMAIN has also partnerned with AMD and Qualcomm for AI chip supply and announced an AI PC with Qualcomm, while AWS will continue offering P6e-GB200 UltraServers and existing P6 instances.