AWS Expands AI Portfolio with Factories and New Nova Models

AWS has announced AI Factories, an offering that deploys dedicated infrastructure into customers’ existing data centres to address sovereignty and compliance requirements for organisations scaling AI projects.

The service combines AI accelerators including Nvidia computing platforms and Trainium chips with AWS networking, storage, databases, security infrastructure and AI services including Amazon Bedrock and Amazon SageMaker. The infrastructure operates as a private AWS Region within customer facilities, providing access to compute, storage, database and AI services.

“With this launch, we're enabling customers to deploy dedicated AI infrastructure from AWS in their own data centres,” says Matt Garman, CEO of AWS. “We also give them access to leading, purpose-built compute, including the very latest training UltraServers and access to services like SageMaker.”

AWS and Nvidia extend 15-year partnership

The collaboration between AWS and Nvidia extends a relationship that began 15 years ago with the launch of the world’s first GPU cloud instance. “AWS is by far the best place to run Nvidia GPUs,” says Matt. “We were the first cloud provider committed to use the cloud for GPUs and we’ve been collaborating together for over 15 years. If you talk to anyone who’s run large GPU clusters, they’ll tell you that AWS is by far the most stable for running GPU workloads.”

AI Factories provides AWS customers with access to the Nvidia accelerated computing platform, Nvidia AI software and GPU accelerated applications within customer data centres. The AWS Nitro System, Elastic Fabric Adapter petabit scale networking and Amazon EC2 UltraClusters support the Nvidia Grace Blackwell and next-generation Vera Rubin platforms, while AWS will support Nvidia NVLink Fusion high-speed chip interconnect technology in its next-generation Trainium4 chips.

“Large-scale AI requires a full-stack approach, from advanced GPUs and networking to software and services that optimise every layer of the data centre,” says Ian Buck, VP and General Manager of Hyperscale and HPC at Nvidia. “By combining Nvidia’s latest Grace Blackwell and Vera Rubin architectures with AWS’ secure, high-performance infrastructure and AI software stack, AWS AI Factories allow organisations to stand up powerful AI capabilities in a fraction of the time and focus entirely on innovation instead of integration.”

AWS and Nvidia are also collaborating with Saudi Arabia-based Humain to build an AI Zone featuring up to 150,000 AI chips including GB300 GPUs, dedicated AWS infrastructure and AWS AI services within a Humain data centre.

“The AI factory AWS is building in our new AI Zone represents the beginning of a multi-gigawatt journey for HUMAIN and AWS,” says Tareq Amin, CEO of Humain. “What truly sets this partnership apart is the scale of our ambition and the innovation in how we work together. We chose AWS because of their experience building infrastructure at scale, enterprise grade reliability, breadth of AI capabilities and depth of commitment to the region.”

Amazon Bedrock expands with new Nova 2 models and 18 open weight options

Amazon has also released four Nova 2 models alongside Nova Forge – a service for organisations to build custom model variants – and Nova Act for creating AI agents. Amazon Bedrock is now powering AI in production for more than 100,000 companies globally. The Nova 2 family includes Lite, Pro, Sonic and Omni variants designed for different applications across reasoning, multimodal processing, conversational AI and code generation.

Nova 2 Lite processes text, images and videos to generate text, with adjustable reasoning depth for balancing intelligence with speed and cost. Nova 2 Pro handles text, images, video and speech for applications including agentic coding, long-range planning and problem-solving. Both models include web grounding and code execution capabilities.

Nova 2 Omni processes text, images, video and speech inputs while generating text and images within a single model. The system handles up to 750,000 words, hours of audio, videos and hundred-page documents, with organisations including Cisco, Siemens, Sumo Logic and Trellix using Nova 2 models for applications ranging from threat detection to video understanding.

AWS has added 18 open weight models to Amazon Bedrock, including two new model sets from Mistral AI available first on the platform. Mistral Large 3 provides long-context, multimodal, and instruction reliability capabilities, while Ministral 3 offers compact, general-purpose and multimodal AI functionality. The expansion also sees includes models from Google’s Gemma 3, MiniMax’s M2, Nvidia’s Nemotron and OpenAI’s GPT OSS Safeguard available on Bedrock.

Nova models enable custom training and browser automation

Nova Forge provides organisations with access to pre-trained, mid-trained and post-trained Nova model checkpoints, allowing customers to integrate proprietary data with Amazon Nova-curated datasets throughout the training process. The platform offers reinforcement learning environments for training AI using synthetic scenarios, synthetic data-based distillation for creating smaller models and a responsible AI toolkit for implementing safety controls.

Organisations including Booking.com, Cosine AI, Nimbus Therapeutics, Nomura Research Institute, OpenBabylon, Reddit and Sony are building models with Nova Forge.

“Working with Nova Forge is allowing us to improve content moderation on Reddit with a more unified system that’s already delivering impressive results,” says Chris Slowe, CTO at Reddit. “We’re replacing a number of different models with a single, more accurate solution that makes moderation more efficient. The ability to replace multiple specialised ML workflows with one cohesive approach marks a shift in how we implement and scale AI across Reddit.”

Nova Act provides infrastructure for building and deploying AI agents that execute actions in web browsers. Powered by a custom Nova 2 Lite model trained through reinforcement learning on thousands of tasks across hundreds of simulated web environments, the service delivers 90% reliability on customer workflows.

Startup Sola Systems has integrated Nova Act to automate hundreds of thousands of workflows per month for clients across tasks including reconciling payments, coordinating shipments and updating medical records. Hertz accelerated its software delivery by five times and reduced quality assurance testing from weeks to hours using Nova Act.