Choose where to buy Claude on who processes your data and which features and quotas you get. The Claude API, Claude Platform on AWS, and Microsoft Foundry all charge Anthropic's per-token rates. Amazon Bedrock and Google Cloud (Google managed) set their own prices and process your data themselves.
When you need this
- Your organization has a cloud spending commitment it wants Claude usage to count toward.
- Procurement wants one invoice, or security wants traffic inside an existing cloud boundary.
- A feature you plan to use, such as the Batch API, is not on every platform.
Evaluation criteria
Operator
Is the platform operated by Anthropic or by the cloud provider?
Look forAnthropic operates the Claude API, Claude Platform on AWS, Microsoft Foundry, and Claude Models on Google Cloud (Anthropic managed), launched October 22, 2026. AWS and Google operate and price Claude on Amazon Bedrock and Claude Models on Google Cloud (Google managed), called Google Cloud below.
Feature coverage
Does the platform offer the features you need?
Look forThe Message Batches API runs on every Anthropic-operated platform except Microsoft Foundry. The Files API, code execution, and the MCP connector run on all four. Fast mode for Opus 5.5 runs on the Claude API, Azure-hosted Microsoft Foundry deployments, and Google Cloud (Anthropic managed). Fast mode for Opus 5 and Opus 4.8 runs on the Claude API only, in pre-release.
Billing
How does usage appear on your invoice?
Look forClaude Platform on AWS and Microsoft Foundry bill through the cloud marketplace in Claude Consumption Units at US$0.01 each, with token usage rated at standard Claude API prices. Google Cloud (Anthropic managed) bills the same units through Google Cloud Marketplace, metered by Anthropic.
Quotas
Who sets your rate limits, and how do you raise them?
Look forOn partner platforms the provider sets quotas. A team on Google Cloud found the Opus quota defaulted to zero and its increase requests were auto-rejected.
Regions and pricing premiums
Do you need traffic pinned to a region?
Look forRegional and multi-region endpoints on Bedrock and Google Cloud cost 10% more than global endpoints for Claude 4.5 and later models. US-only inference on Anthropic-operated platforms costs 1.1x.
Model lifecycle
Whose retirement schedule applies to the models you use?
Look forAnthropic's dates cover its own platforms. Amazon Bedrock and Google Cloud set their own retirement schedules, so check both before you commit a workload.
Questions to ask vendors
- Does our existing Bedrock or Google Cloud private offer carry over if we move to Claude Platform on AWS?
- Which features on our roadmap are available on the platform we plan to buy through?
- What are the default quotas for the models we need, and how long do increases take?
Red flags
- Planning a batch-heavy workload on Bedrock, Google Cloud, or Microsoft Foundry, where the Message Batches API is not offered.
- Assuming Anthropic's retention terms apply on Bedrock or Google Cloud, where the provider is the data processor.
- Choosing a platform before checking its default quota for the model you need.
Products in this piece
Sources
- Official documentation
- 11
- Public source
- 4
- Claude API docs: Features overview
- Claude API docs: Pricing
- Claude API docs: Claude Platform on AWS
- Claude API docs: Claude in Microsoft Foundry
- Claude API docs: Claude in Amazon Bedrock
- Claude API docs: Claude on Google Cloud
- Claude API docs: Claude Models on Google Cloud (Anthropic managed)
- Claude API docs: Fast mode
- Claude API docs: Claude Platform release notes
- Claude API docs: API and data retention
- Claude API docs: Model deprecations
- Hacker News comment on default Opus quotas on Google Cloud (Mar 27, 2026)
- Hacker News comment on running Claude through Vertex AI (Mar 27, 2026)
- Hacker News comment on clients buying through Bedrock or Azure (Oct 9, 2026)
- Hacker News comment on Opus 5 requiring provider data sharing on Bedrock (Jul 24, 2026)