← The AI Systems Map
L0 · LabsFrontier labUnited States

Anthropic

Safety-focused lab; maker of Claude

Website ↗GitHub ↗

Anthropic builds the Claude family with a research emphasis on interpretability, Constitutional AI, and safety, and is the leader in enterprise/coding usage via Claude Code and API. It runs multi-cloud on AWS (primary, using Trainium) and Google Cloud (using TPUs). In 2025-2026 it raised large rounds led by strategic backing from Amazon (~$8B total) and Google (~$3B+), and shipped Claude Opus and Sonnet generations targeting agentic and coding workloads.

Flagship model

Claude (Opus / Sonnet)

Primary cloud

AWS (Trainium) + Google Cloud (TPU)

Lead investors

Amazon, Google

Strength

Enterprise, coding, safety

How it fits the stack

Anthropic with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.

depends onhostsusesused bydepends onhostsused bydepends onsuppliesdepends ondepends onusesusesusespartners withcompetes withpartners withpartners withAnthropicLabsAmazon Web Services(AWS)AWS Trainium /InferentiaGoogle Cloud (GCP)Google TPUAI coding agents &IDEsAI Coding Assistants(Cursor, Copilot,Windsurf, Claude Code)Cognition (Devin)Cursor (Anysphere)GitHub CopilotGleanHarvey AIAI Safety &InterpretabilityResearch (ARC,Amazon AGI (Nova)Anduril / Defense AIDefense & governmentAI
AnthropicDepends on ↑Feeds ↓Related

Anthropic in the AI stack. Anthropic with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.

Graph data (text) — 16 entities, 18 relationships
  • Anthropic —depends on→ Amazon Web Services (AWS)
  • Anthropic —hosts→ Amazon Web Services (AWS)
  • Anthropic —uses→ AWS Trainium / Inferentia
  • Anthropic —used by→ AWS Trainium / Inferentia
  • Anthropic —depends on→ Google Cloud (GCP)
  • Anthropic —hosts→ Google Cloud (GCP)
  • Anthropic —used by→ Google TPU
  • AI coding agents & IDEs —depends on→ Anthropic
  • AI Coding Assistants (Cursor, Copilot, Windsurf, Claude Code) —supplies→ Anthropic
  • Cognition (Devin) —depends on→ Anthropic
  • Cursor (Anysphere) —depends on→ Anthropic
  • GitHub Copilot —uses→ Anthropic
  • Glean —uses→ Anthropic
  • Harvey AI —uses→ Anthropic
  • Anthropic —partners with→ AI Safety & Interpretability Research (ARC, Redwood, Apollo)
  • Anthropic —competes with→ Amazon AGI (Nova)
  • Anthropic —partners with→ Anduril / Defense AI
  • Anthropic —partners with→ Defense & government AI

Depends on ↑ · 22

depends onAmazon Web Services (AWS)AWS is Anthropic's primary cloud; trains on TrainiumhostsAmazon Web Services (AWS)Primary cloud + Project Rainier Trainium2 cluster for Anthropic trainingusesAWS Trainium / InferentiaProject Rainier Trainium2 cluster for Claude trainingused byAWS Trainium / InferentiaAnthropic trains on Trainium2 (Project Rainier)depends onGoogle Cloud (GCP)Anthropic also uses Google Cloud TPUs at large scalehostsGoogle Cloud (GCP)2025 deal for up to ~1M TPUs / ~1GW+ capacity for Anthropicused byGoogle TPUAnthropic committed to large TPU capacitysuppliesMercorexpert-domain labelers for post-training/evalssuppliesMLOps / experiment & observabilityExperiment tracking / observabilitydepends onNvidia Data-Center GPU (Blackwell/Rubin)⬦Also uses Nvidia GPUs alongside Trainium/TPUused byNvidia Data-Center GPU (Blackwell/Rubin)⬦Anthropic uses Nvidia GPUs alongside TPU/TrainiumusesPyTorchPyTorch/JAX are the substrate above the accelerator layerdepends onScale AI⬦RLHF & evaluation datasuppliesScale AI⬦Expert labeling and evalssourcesSurge AIRLHF/human-preference data from neutral vendor after Meta absorbed ScalesuppliesSurge AIPremium RLHF/human preference data to frontier labsdepends onSurge AI / Mercor (expert data)PhD-level domain annotationsuppliesSurge AI / Mercor (expert data)Expert human data / RLHF supplier to labsusesSurge AI / Mercor / Turing (frontier data labor)Expert human SFT/RLHF contractorsused bySWE-bench / Evaluation Benchmarkscoding benchmark used to evaluate ClaudeusesSynthetic data & distillationConstitutional AI and distillation use synthetic datasuppliesTraining Data & Web Corpus⬦data access/provenance is a contested non-hardware chokepoint

Context — capital, rivals, policy · · 30

partners withAI Safety & Interpretability Research (ARC, Redwood, Apollo)Interpretability/scheming research informs deploymentcompetes withAmazon AGI (Nova)Amazon's first-party Nova competes with its own investee on Bedrockpartners withAnduril / Defense AIClaude for defense via Palantir/AWS gov dealspartners withDefense & government AICDAO / DoD frontier contractspartners withDefense & National Security AI⬦Claude in classified/gov environmentspartners withEnterprise Data Cloud (Snowflake/Databricks)⬦Snowflake Cortex / Databricks model accesscompetes withGoogle DeepMindCompete for enterprise despite Google's investmentpartners withMETRAutonomy evals with pre-release accesscompetes withOpenAIDirect frontier and enterprise/coding rivalspartners withPalantir AIPAIP deploys Claude in enterprise/gov via ontologypartners withPalantir TechnologiesAIP + Claude for defense/enterprise deployment on AWSpartners withSalesforce AgentforceAgentforce integrates Anthropic modelspartners withUK AI Safety Institute (AISI)Pre-deployment model evaluation accessinvests inAmazon$8B invested; primary compute partnerinvests inAmazon AGI (Nova)Amazon invested ~$8B in Anthropic; primary cloud partnerinvests inAmazon Web Services (AWS)~$8B+ total strategic investment in Anthropicinvests inAndreessen Horowitz (a16z)holds Anthropic exposure among model-layer betsinvests inAnthropic capital stack (Amazon/Google)~$183B-round capital funds Claude training computeinvests inGoogle / Alphabet$3B+, ~14% stake, while competing via Geminiinvests inGoogle Cloud (GCP)Multiple equity investments in Anthropicinvests inGoogle DeepMindGoogle invested ~$3B+ in Anthropic and provides TPUsrestrictsAI Copyright Litigation (NYT et al.)⬦~$1.5B Anthropic books settlement (2025)restrictsAI Safety Institutes (US CAISI / UK AISI)Pre-deployment frontier evalsrestrictsCopyright & AI litigation⬦Bartz v. Anthropic settlement shapes data userestrictsEU AI Acttransparency/risk duties apply to frontier providers in EUrestrictsEU AI Act GPAI Code of PracticeSigned the GPAI Code of PracticerestrictsUS AI Safety Institute (CAISI/NIST)Pre-deployment evaluation of frontier modelsdesignsClaude (Opus / Sonnet)Anthropic designs the Claude Opus/Sonnet/Haiku familydesignsFrontier Safety Frameworks (RSP / Preparedness / FSF)Responsible Scaling Policy / ASL tiersdesignsModel Context Protocol (MCP)Anthropic authored MCP; adopted by OpenAI/Google/Microsoft
← Back to the AI Systems Map