Amazon (AWS AI)

Amazon launched AWS with EC2 (Elastic Compute Cloud) and S3 (Simple Storage Service), creating the on-demand cloud computing model that would become the infrastructure backbone of the entire AI industry.

Amazon launched Amazon Echo and the Alexa voice assistant, bringing AI-powered natural language interaction into millions of homes. Alexa would become the first widely deployed consumer AI assistant and spawned the Alexa Skills Kit developer ecosystem, establishing Amazon's leadership in consumer AI.

AWS launched Amazon SageMaker at re:Invent 2017, a fully managed service for building, training, and deploying machine learning models, making ML accessible to developers without deep ML expertise.

AWS launched Inferentia, its first custom-designed machine learning inference chip, alongside the Inf1 instance family. This marked Amazon's entry into purpose-built AI silicon, reducing inference costs by up to 45 percent compared to GPU-based alternatives and reducing dependence on NVIDIA hardware.

AWS launched Amazon Bedrock, a fully managed service offering foundation models from AI21 Labs, Anthropic, Cohere, Meta, Stability AI, and Amazon via a single API, positioning AWS as a neutral AI model marketplace.

Amazon announced a partnership with Anthropic, investing $1.25 billion initially (totaling $8 billion by 2024). Anthropic became an AWS primary customer and made Claude available to AWS customers via Amazon Bedrock.

Amazon announced a $4 billion strategic investment in Anthropic, with Anthropic selecting AWS as its primary cloud provider for training and deploying future models. The partnership gave Anthropic access to AWS's Trainium and Inferentia chips while locking in AWS as the infrastructure provider for frontier AI development.

AWS launched Amazon Q, a generative AI-powered assistant for workplace tasks including answering questions, generating content, and connecting to enterprise data sources. Amazon Q represented AWS's direct entry into the enterprise AI assistant market, competing with Microsoft Copilot and ChatGPT Enterprise.

AWS announced Trainium2 chips and expanded AI infrastructure at re:Invent 2023, including Project Ceiba (a massive AI supercomputer with 20,000+ Trainium2 chips for Anthropic) and AWS Generative AI Innovation Centers.

AWS launched Trainium2, its second-generation custom AI training chip designed to reduce training costs by up to 50% compared to GPU-based instances. Trainium2 was deployed in EC2 Trn2 instances and large-scale UltraClusters, enabling AWS customers to train foundation models at significantly lower cost.

AWS announced Amazon Nova, its own family of foundation models including Micro, Lite, Pro, and Premier tiers for text, multimodal, and image generation. The entry positioned AWS as both a model provider and model marketplace, competing directly with Google, Microsoft, and standalone AI labs.

At AWS re:Invent 2025, Amazon launches Nova 2 model family — including Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic (speech-to-speech), and Nova 2 Omni — alongside Nova Forge (custom model building service) and Nova Act (browser agent service). Nova 2 Pro delivers 1M-token context and matches frontier models on key benchmarks.

Amazon entered advanced discussions to invest up to $50 billion in OpenAI as part of a potential AI partnership, alongside SoftBank ($30 billion) and Nvidia ($30 billion), in a $120 billion round valuing OpenAI at $852 billion.

Amazon overhauls its AI strategy, deprecating most flagship Nova models (Premier, Omni, Reel, Canvas) and redirecting resources to a new frontier-model initiative called Frontier Model Research (FMR), led by UC Berkeley researcher Pieter Abbeel. The restructuring includes layoffs in the AGI organization and shutdown of AGI Lab, the Adept-acquisition research group.