Now loading...
Amazon Web Services has introduced a new service called AWS AI Factory, designed to deliver artificial intelligence infrastructure straight to customers’ own data centers.
The announcement came during the company’s Re:Invent 2025 event in Las Vegas. This offering allows AWS to deploy AI resources, such as Nvidia graphics processing units, its own Trainium processors, along with networking, storage solutions and databases, directly into client facilities.
These setups are customized and run solely for the individual customer, helping governments and major enterprises expand their AI initiatives while adhering to regulatory requirements and data sovereignty standards.
Functioning similarly to a private AWS cloud region, the AI factories provide entry to AWS-managed tools, including foundational AI models, all while letting customers decide exactly where their data gets handled and kept.
There’s an option for a combined Nvidia-AWS AI Factory setup, which unlocks Nvidia’s hardware, comprehensive AI software stack and computing platform. It incorporates AWS’s Nitro System, Elastic Fabric Adapter for high-speed networking at petabit levels, and Amazon EC2 UltraClusters to back Nvidia’s Grace Blackwell and upcoming Vera Rubin systems. Looking ahead, AWS intends to align its next Trainium4 processors with Nvidia’s NVLink Fusion technology.
“Scaling up AI demands a comprehensive strategy, spanning top-tier GPUs and networking to software and services that fine-tune every aspect of the data center. Partnering with AWS, we’re bringing this complete package right into customers’ setups,” stated Ian Buck, Nvidia’s vice president and general manager for hyperscale and high-performance computing.
“Merging Nvidia’s newest Grace Blackwell and Vera Rubin designs with AWS’s robust, secure infrastructure and AI tools enables organizations to deploy potent AI systems much faster, letting them concentrate on breakthroughs rather than piecing together components.”
This initiative expands on AWS’s Project Rainier, an earlier AI factory built for Anthropic using Trainium2 chips, and mirrors the approach in AWS’s collaboration with Humain in Saudi Arabia.
Just last month, AWS and Humain broadened their alliance to roll out approximately 150,000 AI processors, featuring Nvidia GB300 units and Trainium chips.
“The AI factory AWS is establishing in our fresh AI Zone marks the start of a multi-gigawatt expansion for Humain and AWS. This setup was crafted from the ground up to meet surging local and worldwide needs for AI processing power,” remarked Tareq Amin, chief executive of Humain.
“Our collaboration stands out due to its vast scope and creative methods of teamwork. We selected AWS for their expertise in large-scale builds, dependable enterprise performance, wide array of AI features and strong regional dedication. With our joint push into global markets, we’re fostering an environment that will define how AI concepts are developed, launched and grown for the entire globe.”
The Re:Invent gathering also unveiled Trainium3 UltraServers and shared more on AWS’s forthcoming Trainium4 processors.
Recently, AWS revealed intentions to invest $50 billion to boost AI and high-performance computing resources tailored for the U.S. government.
