AWS has long sold computing power by the hour. Now it is selling the model itself.
On December 3, at the company’s annual re:Invent conference in Las Vegas, Amazon announced the Nova family of foundation models. The move is a direct play for the corporate AI market — the same market where OpenAI, Anthropic, and Google have been jockeying for months. But Nova is different. It is not a standalone product sold to consumers. It is a tool built inside Amazon Web Services, designed to be used by the same companies that already rent servers from AWS.
Foundation models are large AI systems trained on vast datasets. A single model can handle text generation, image recognition, translation, and more. Until now, most companies that wanted to use such models had to go through an API from a third party — OpenAI’s GPT-4, say, or Anthropic’s Claude. AWS is offering its own alternative, baked directly into its cloud infrastructure.
The timing matters. AWS already dominates cloud computing. According to the company’s own materials, its services are used by individuals, companies, and governments around the world. Clients pay for computing power on demand, scaling up during busy periods and scaling down to cut costs. Nova slots into that existing ecosystem. A company that already stores its data on AWS and runs its applications on AWS can now also run its AI models on AWS — without ever leaving the platform.
The announcement came with little warning. AWS did not preview Nova at earlier events. The company simply unveiled the family of models during a keynote at re:Invent, a conference that has historically been the venue for major AWS product launches. The message was clear: Amazon sees foundation models as a core part of its cloud business, not a side experiment.
Amazon is not the first tech giant to offer foundation models through its cloud platform. Google Cloud has its own family of models, called Gemini. Microsoft Azure offers access to OpenAI’s models. But AWS is the largest cloud provider by market share. Its entry into the foundation model space shifts the competitive landscape. Companies that were considering which AI model to adopt now have a reason to stay inside the AWS ecosystem.
The Nova family is not a single model. It is a set of models, each presumably tuned for different tasks — text, image, possibly code. That structure mirrors the approach taken by Google and Meta, both of which have released families of models rather than a single one-size-fits-all system. AWS has not yet released detailed performance benchmarks, but the company has framed Nova as a tool for driving innovation and efficiency.
For AWS clients, the arrival of Nova means one fewer reason to look elsewhere. The company’s pitch has always been about simplicity: you do not manage your own servers, you do not manage your own operating systems, and now you do not need to manage your own AI model provider. Everything lives inside AWS. That is the logic behind the announcement, and it is the logic that has made AWS the dominant force in cloud computing for a decade.
The real test will be adoption. Will companies that already use AWS switch to Nova, or will they stick with the models they know? AWS is betting that convenience will win. The company has spent years making it easy for clients to do everything inside its walls. Nova is the latest brick in that wall.
