Shipping in companies

Claude on Bedrock, Vertex or the Anthropic API: pick for control, not fashion

The right Claude runtime is about data residency, IAM and procurement, not a leaderboard. Match the platform to the review you have to pass.

Claude runs on the Anthropic API, Amazon Bedrock and Google Vertex. Quality differences on a given model generation are usually smaller than the differences in IAM, networking, logging and legal paper. Pick the runtime that your security review can sign.

A startup with no residency constraint can start on the Anthropic API and move later. A bank that already lives in AWS will often want Bedrock. A public-sector buyer may dictate Vertex or a private region. None of those choices replace evals or tool design.

What actually changes

ConcernWhy it matters
Data processing termsTraining, retention, subprocessors
IdentityHow the app authenticates, not a shared key in Slack
NetworkingPrivateLink, VPC-SC, no public egress
Quotas and batchHow you absorb a launch spike
Feature lagCaching, tools, computer use land on platforms on different clocks

Do not fork the product per cloud

Keep the agent loop, schemas and evals in your repository. Treat the model client as an adapter. If you fork prompts per vendor, you will never know which change broke quality.

Isolation is a requirement, not a slogan

Healthcare, finance and government buyers will ask where bytes sit. Answer with a diagram, not a brand preference. That is the same conversation as HIPAA and SOC 2.

We meet your stack. The production build does not import a mystery platform you have to keep renting.

Related notes