Anthropic and Amazon have signed a major expansion of their strategic partnership, securing up to 5 gigawatts (GW) of compute capacity to support the training and deployment of Claude AI models. The agreement includes a long-term commitment exceeding $100 billion over 10 years for infrastructure built primarily on Amazon Web Services, reinforcing AWS as Anthropic’s primary cloud and training platform for mission-critical AI workloads.
The expanded collaboration centers on scaling custom silicon and cloud infrastructure. Anthropic will leverage AWS-designed chips including Trainium2, Trainium3, and future generations through Trainium4, alongside AWS Graviton CPUs. Amazon expects significant Trainium2 capacity to come online in Q2 2026, with nearly 1 GW of combined Trainium2 and Trainium3 capacity deployed by year-end. The agreement also extends inference infrastructure globally, with expanded deployment of Claude via Amazon Bedrock across Asia and Europe to support growing enterprise demand.
Anthropic also plans deeper integration of its Claude platform within AWS, allowing customers to access models directly through existing AWS accounts with unified billing, security, and governance controls. The company reports that more than 100,000 customers already use Claude on Bedrock. In parallel, Amazon will invest an additional $5 billion in Anthropic, with the option to increase total investment by up to $20 billion, building on its prior $8 billion commitment. Anthropic said its annualized revenue has surpassed $30 billion, driven by accelerating enterprise adoption and surging consumer usage, which has placed increasing pressure on compute infrastructure.
- Up to 5 GW of AI compute capacity secured for Claude training and inference
- $100+ billion infrastructure commitment over 10 years on AWS
- Deployment roadmap includes Trainium2 (2026), Trainium3, and future Trainium4 chips
- Nearly 1 GW of Trainium capacity online by end of 2026
- Global inference expansion across Asia and Europe via Bedrock
- Claude platform fully integrated into AWS environment (billing, security, governance)
- Amazon to invest $5B immediately, with potential to reach $20B additional
- Anthropic run-rate revenue exceeds $30B, up from ~$9B at end of 2025
“Our users tell us Claude is increasingly essential to how they work, and we need to build the infrastructure to keep pace with rapidly growing demand,” said Dario Amodei, CEO and co-founder of Anthropic. “Our collaboration with Amazon will allow us to continue advancing AI research while delivering Claude to our customers, including the more than 100,000 building on AWS.”
🌐 Analysis: This deal underscores the intensifying competition in AI infrastructure, where hyperscalers are vertically integrating custom silicon, cloud platforms, and AI model ecosystems. Amazon’s Trainium roadmap positions it directly against GPU-centric strategies from competitors, while locking in Anthropic as a long-term anchor tenant for its AI infrastructure stack.
At 5 GW scale, the agreement places Anthropic among the largest consumers of AI compute globally, comparable to hyperscaler internal deployments. It also highlights a broader shift toward multi-platform model availability—Claude remains accessible across AWS, Google Cloud, and Microsoft Azure—even as infrastructure commitments increasingly concentrate with a primary partner.
| Anthropic AI Infrastructure Stack: Capacity Sources & Technology Suppliers (April 2026) | |||
|---|---|---|---|
| Cloud / Partner | Primary Silicon | Scale & Commitments | Role in Anthropic Strategy |
| Amazon Web Services (AWS) | Trainium2 / Trainium3 / Trainium4 Graviton CPUs | Up to 5 GW capacity (2026+) ~1 GW Trainium2/3 by end of 2026 $100B+ 10-year infrastructure commitment Project Rainier cluster (≈1M Trainium2 chips, reported) | Primary training and inference platform Core foundation for Claude deployment Deepest strategic and financial partnership |
| Google Cloud | Tensor Processing Units (TPUs) (v5e and next-generation) | ~1 GW-class capacity ramping in 2026 Multi-GW expansion starting 2027 (via Google/Broadcom supply chain) Up to ~1M TPU-scale deployments (planned) | Second major training pillar High-efficiency large-scale model training Rapidly expanding capacity footprint |
| Microsoft Azure | NVIDIA GPUs (Hopper, Blackwell, future Vera Rubin) | Up to ~1 GW NVIDIA-based capacity (announced 2025) ~$30B compute commitment (reported partnership scope) | GPU-based training and inference Complements custom silicon strategy Ensures access to industry-standard AI hardware |
| Multi-Cloud Distribution | Combined: Trainium + TPU + NVIDIA GPU | Claude available across AWS (Bedrock), Google Cloud (Vertex AI), and Azure Global inference expansion (U.S., Europe, Asia) | Workload optimization across chip types Supply chain diversification Enterprise flexibility and resilience |
