Amazon said it will invest $50 billion to build and deploy AI and high performance computing AWS infrastructure built for the US government.
Construction on the infrastructure will break ground in 2026. The new investment will add almost 1.3 gigawatt of compute capacity for AWS Top Secret, AWS Secret and AWS GovCloud Regions for multiple classification levels.
Hyperscale cloud providers have been investment multiple billions of dollars to build out capacity for AI. These cloud providers also see a big total addressable market for public sector and government customers.
- OpenAI inks $38 billion deal with AWS, starts renting GPUs immediately
- AWS re:Invent 2025 to feature Trainium3, scaling secure AI agents
- AWS fires up Project Rainier, Trainium2 cluster for Anthropic
According to AWS, which kicks off its re:Invent 2025 conference next week, the US government focused infrastructure will be powered by its AWS Trainium AI chips as well as Nvidia GPUs. Federal agencies can use the infrastructure to access Amazon SageMaker AI, Amazon Bedrock and multiple models including Anthropic's Claude and Amazon Nova.
Use cases for the new infrastructure will range from national security to scientific research and innovation. AWS has been building US government focused clouds and infrastructure since 2011.
AWS CEO Matt Garman said the company's first-ever supercomputing infrastructure for government customers will give agencies "expanded access to advanced AI capabilities."
Constellation Research analyst Holger Mueller said:
"AWS has the largest government share of the cloud providers, and as such has practically the most to lose. Implicitly AWS confirms that HPC is a potential disruptor for commercial relations of cloud vendors - and therefore has to invest in it. The key question is going to be if AWS can pony up the capex for other key hardware innovations, AI super computers first, quantum platforms next."
