Anthropic

Full briefing

Anthropic PBC was founded in 2021 by Dario Amodei and Daniela Amodei, along with several colleagues who departed OpenAI, and is headquartered in San Francisco. As a public benefit corporation, Anthropic is structured to prioritize AI safety alongside commercial viability, a distinction that has attracted significant investment: Amazon Web Services committed $4 billion (with an option to reach $4B total), Google committed $2 billion, and Spark Capital and other VCs have participated in rounds valuing the company at over $18 billion as of 2024. Anthropic's core product is the Claude family of large language models: Claude 3 Opus, Claude 3.5 Sonnet, and Claude 3 Haiku, with newer generations following a similar tiered naming convention. Claude is accessed through the Claude.ai consumer interface, the Anthropic API, and through AWS Bedrock (where it is a flagship model) and Google Cloud Vertex AI. AWS's investment comes with a commitment that Anthropic uses AWS Trainium and Inferentia custom AI chips alongside NVIDIA GPU clusters — making Anthropic one of the largest buyers of both NVIDIA H100/H200 GPUs and AWS custom silicon. Anthropic's research contributions to the AI safety field include Constitutional AI (CAI) — a method for training AI systems to be helpful, harmless, and honest by having the model self-critique against a constitution of principles — and the Responsible Scaling Policy (RSP), which commits Anthropic to pausing model development if evaluations indicate certain capability thresholds are crossed. These frameworks have influenced industry practice at other leading labs. From a supply chain perspective, Anthropic represents a new category of AI infrastructure buyer: frontier AI labs that are neither hyperscalers nor traditional enterprises, but require hyperscale GPU clusters for frontier model training. Training a frontier model like Claude 3 Opus is estimated to require tens of thousands of H100 GPU-hours, making these labs collectively a multi-billion dollar annual demand source for NVIDIA and the TSMC CoWoS packaging ecosystem. Anthropic also operates inference infrastructure at significant scale to serve the Claude API and Claude.ai consumer traffic. This inference fleet — deployed primarily on AWS and Google Cloud — represents ongoing GPU demand even between training runs, contributing to the sustained tightness in GPU supply that characterized 2023–2025.