Amazon

Amazon.com, Inc. develops the Nova foundation models, distributes models through Amazon Bedrock, and supplies AI infrastructure through AWS and its Trainium chips. Its consumer assistant Alexa+ uses models from both Nova and Anthropic. In 2026, Amazon announced major investments in OpenAI and Anthropic alongside agreements covering cloud services and AI compute capacity.[3][7][2][6]

Amazon introduced Nova at AWS re:Invent on December 3, 2024. Micro was a text-only model; Lite, Pro, and Premier were designed to process text, images, and video and generate text. Micro, Lite, and Pro were generally available at introduction, while Premier was scheduled for the first quarter of 2025. The family also included Canvas for image generation and Reel for video generation.[3]

At introduction, Micro supported 128K input tokens, while Lite and Pro supported 300K tokens or 30 minutes of video processing. Bedrock made Amazon’s models and those from other AI companies accessible through a single API. Nova supported fine-tuning on labeled proprietary data and distillation, transferring specific knowledge from a larger teacher model to a smaller model. Integration with Bedrock Knowledge Bases supported retrieval-augmented generation, grounding responses in an organization’s own data. Amazon also designed Nova for agentic applications that interact with proprietary systems through multiple APIs to execute multistep tasks.[3]

Rohit Prasad, then identified as SVP of Amazon Artificial General Intelligence, described the internal scale behind the launch: “Inside Amazon, we have about 1,000 generative AI applications in motion”. The release also described SAP integrating Nova into its SAP AI Core generative AI hub to support new skills for Joule, SAP’s AI copilot, and applications using SAP business data. Dentsu Digital was integrating Reel into campaign creation, while Caylent was using Nova for video understanding across media, sports, and retail.[3]

On February 27, 2026, OpenAI announced that Amazon would invest $50 billion. Latent.Space’s February 28 account described an initial $15 billion followed by another $35 billion in the coming months when certain conditions were met. That account also described an OpenAI-powered “Stateful Runtime Environment” on Bedrock, AWS as the exclusive third-party cloud provider for OpenAI Frontier, and an agreement for two gigawatts of Trainium capacity spanning Trainium3 and Trainium4, worth $100 billion over eight years. These were announced investment and infrastructure terms.[2][1]

Amazon’s April 20, 2026 expansion of its Anthropic collaboration distinguished $8 billion previously invested from a newly announced $5 billion investment that day and up to another $20 billion tied to future commercial milestones. Anthropic committed to spend more than $100 billion on AWS technologies over ten years and secure up to five gigawatts of capacity to train and power its models. The spending and capacity figures were commitments, rather than realized AWS revenue or delivered capacity.[6]

The collaboration included direct engineering work: Anthropic supplied feedback from Claude training workloads to Annapurna Labs to help develop and optimize future Trainium chips. Amazon reported that AWS and Anthropic had launched Project Rainier with nearly half a million Trainium2 chips and that Anthropic was actively using the cluster to train and deploy Claude. As of the April announcement, Amazon reported more than 100,000 customers running Claude models on AWS.[6]

Alexa+ uses a new architecture powered by large language models from Amazon Nova and Anthropic. Its capabilities include remembering context across ongoing conversations, making restaurant reservations, booking rides, and coordinating family calendars. Amazon offers the assistant across Alexa-enabled devices, the Alexa app, and the web.[7]

Amazon’s Alexa+ access announcement, originally published February 4, 2026 and updated September 1, described availability to all U.S. customers. Prime members received unlimited household access to the full experience at no additional cost; customers without Prime could subscribe for $19.99 per month. A separate free text-chat tier through Alexa.com and the Alexa app was limited based on use. The September update clarified tier details for the Prime benefit that began in February.[7]

These AI products and agreements followed earlier expansions in online retail and developer infrastructure. Amazon opened its online bookstore in July 1995; AWS launched the S3 storage service in March 2006.[5][4]

Jeff Bezos launched Amazon.com as a small online book retailer from his garage. Amazon’s July 2000 retrospective dates its opening for business to July 1995. It recalls moving the website to a new office in the back seat of the chief programmer’s Honda that November, with 20 minutes of downtime.[5]

Amazon went public on NASDAQ as AMZN in May 1997 and introduced 1-Click shopping that September. It acquired IMDb in April 1998 and opened Amazon.co.uk and Amazon.de in October 1998. In September 1999, zShops enabled outside sellers—from small businesses to large manufacturers—to offer merchandise on Amazon.com. By its fifth anniversary in July 2000, Amazon reported twenty million customers served.[5]

AWS launched Amazon S3 on March 14, 2006, giving developers storage through SOAP and REST interfaces. At launch, each data block could be up to five gigabytes and had a user-defined key, key/value metadata, and an access-control list governing reading and writing. Data was transparently stored in multiple locations. The pay-as-you-go price was 15 cents per gigabyte stored for a month, plus 20 cents per gigabyte transferred in or out—launch-era specifications and prices for Amazon’s developer infrastructure.[4]