At Google Cloud Next 2026, Anthropic took the stage to announce its partnership with Google Cloud. The presentation included the usual talking points about scale, speed, and integration. But one slide had fine print that told a different story:
“Amazon Web Services is Anthropic’s primary cloud provider and primary training partner.”
On Google’s stage. Disclosing that AWS is Anthropic’s real home base.
🔍 THE BOTTOM LINE: Anthropic uses Google Cloud for inference and distribution, but AWS for training. The fine print reveals the multi-cloud hierarchy — and who actually pays the bills.
🎭 The Irony
Anthropic has partnerships with all three major cloud providers:
- AWS — Primary cloud provider and training partner. Amazon invested $4 billion in Anthropic.
- Google Cloud — Distribution and inference partner. Google invested $2 billion.
- Microsoft Azure — Available via Azure OpenAI Service… for OpenAI models, not Claude.
So when Anthropic presents at Google Cloud Next, they’re appearing on the platform they use for serving Claude to customers — while the models those customers use were trained on AWS infrastructure that Amazon paid for.
The disclosure wasn’t accidental. It’s required by both contractual obligations and regulatory transparency. When you accept investment from multiple cloud providers, each one wants their relationship acknowledged.
But the optics are striking: the AI safety company, on Google’s stage, disclosing that their primary infrastructure partner is Google’s biggest cloud competitor.
🏗️ What This Means Practically
For developers and enterprises choosing where to deploy Claude:
- On AWS: Full feature parity, native Bedrock integration, Anthropic’s primary training environment
- On Google Cloud: Vertex AI integration, A5X inference optimisation, Google’s investment relationship
- On Azure: Not available for Claude directly — Azure’s Anthropic partnership doesn’t exist in the same way
Anthropic isn’t cloud-agnostic. It’s multi-cloud with a clear hierarchy: AWS for training, Google for inference scale, and Azure isn’t really in the picture.
🇳🇿 NZ Angle
For NZ businesses evaluating AI platforms:
- Vendor lock-in is real — Your AI provider’s cloud hierarchy affects your deployment options
- Data residency matters — If you need NZ or APAC data residency, check which region and which cloud provider your AI service actually uses
- Multi-cloud doesn’t mean equal — Anthropic on Google Cloud ≠ Anthropic on AWS. The underlying infrastructure, update cadence, and feature availability differ.
