Getting Started

How to Choose Between Bedrock, Vertex AI, Foundry, and Claude Platform on AWS

Four doors to the same models. The good news: for most enterprises, three of them eliminate themselves in the first five minutes.

Claude 3P 101 · Updated July 2026 · Unofficial guide

Choosing a Claude platform sounds like a technology evaluation. In practice it is mostly an organizational question: which cloud does your company already trust with identity, billing, and audit? The models are identical everywhere, list prices match the first-party API across marketplaces, and the same Python SDK works on all four platforms. What differs is operational fit and feature timing. Here is a framework that resolves the decision in order of importance.

Step 1: Start from the cloud you already run

If 90% of your infrastructure is on one cloud, the Claude offering on that cloud is your default: Amazon Bedrock on AWS, Vertex AI on Google Cloud, Microsoft Foundry on Azure. You inherit the identity model your engineers already know, your existing monitoring sees the traffic, spend lands on the bill your finance team already reconciles, and committed-use discounts are negotiated with the provider you already negotiate with. Deviating from this default requires a concrete reason, not a preference.

AWS shops get a second option the other clouds do not have: Claude Platform on AWS, operated by Anthropic itself but running inside AWS, authenticated with AWS SigV4. It is the only 3P route with same-day API parity — new Claude features land the day Anthropic ships them.

Step 2: Check the feature gaps against your roadmap

Core capabilities — Messages API, streaming, tool use, vision, extended/adaptive thinking, prompt caching — are on all four platforms, so a chat assistant or document summarizer runs anywhere. The differences appear in newer capabilities:

CapabilityBedrockVertex AIFoundryClaude Platform on AWS
Core API, streaming, tools, vision, thinking, cachingYesYesYesYes
Batch API, Files API, code execution, web fetchNoNoMostly betaYes
Web search toolNoBasic variantBetaYes
Managed AgentsNoNoNoYes
New-feature timingProvider scheduleProvider scheduleBroad, some features betaSame day as 1P

If your twelve-month roadmap includes batch pipelines, file-centric workflows, or agentic features, Bedrock and Vertex AI drop out for those workloads regardless of which cloud you run — which is exactly why some AWS-centric organizations land on Claude Platform on AWS.

Step 3: Weigh operator model and risk appetite

Bedrock, Vertex AI, and Foundry are managed services operated by the cloud providers; the provider is your support channel and controls the release calendar. Claude Platform on AWS is Anthropic-operated, which buys feature velocity but adds a second operational relationship. Foundry deserves a specific caveat: its feature support is broad, though some newer features are still in beta — fine for pilots, worth a conversation with your engineering leads before production commitments. Compliance-wise, each option largely inherits the hosting cloud's posture; confirm the specific certifications and data-handling terms that matter to you with your provider.

Rule of thumb: single-cloud company, mainstream workload → use your cloud's native offering. AWS company that needs day-one features or Managed Agents → Claude Platform on AWS. Genuinely multi-cloud → pick one primary platform first; the shared SDK keeps a later second platform cheap to add.

Procurement is the quieter tiebreaker. Marketplace purchases typically flow through the cloud agreement your legal team has already reviewed, and committed-use discounts are negotiated with the cloud provider rather than Anthropic. If one of your clouds has an under-consumed spend commitment, routing Claude usage toward it can make the finance conversation noticeably easier — a consideration that has settled more platform decisions than any benchmark.

Step 4: Decide with a two-week test, not a document

Because the SDK is shared, standing up the same prompt on two candidate platforms is an afternoon of work, not a project. Run your actual workload — your documents, your prompts, your traffic pattern — on your shortlist for two weeks. Compare developer experience, quota-raising responsiveness, and how cleanly each fits your logging and access-control standards. That evidence beats any comparison matrix, including this one. And remember the choice is reversible: swapping platforms later mostly means changing a client class, credentials, and model ID strings.

Where to go next

Once you have a candidate, jump to its first-call guide — Bedrock or Claude Platform on AWS — or scan the platform overview and feature matrix on the main guide.