Where it runs
Anthropic’s service
An organization on claude.ai, run from its Organization settings.3 Sales-assisted plans are also sold through AWS Marketplace.11
Enterprise AI · Claude Enterprise
We set up Claude Enterprise for your company, join it to your systems, set its spend limits, and coach each team until it is part of the work.
Updated

Claude Enterprise is Anthropic’s own service, run from an organization you administer. Its seat is $20 a month billed annually and buys access only: every token is billed on top at API rates, with no cap unless an owner sets one.1, 2 So the setup is identity and data, and also spend: who can use what, and how much.
In Anthropic’s own order, with where each step trips companies up and what we do about it.
Give at least two people the Owner role before touching single sign‑on.10
Choose SCIM, which Anthropic recommends, just‑in‑time or invite only, and map the identity provider’s groups.10
Turn on Require SSO for Claude.10
Where it trips Once it is required, anyone not assigned to the app in the identity provider can’t sign in, and an expired certificate or client secret locks everyone out.3
What we do We assign the app by group before enforcing it, and put the certificate’s expiry where your IT team will see it.
Turn on Restrict organization creation, and optionally claim the domain’s existing accounts, in a 30‑day window that can’t be undone.10
Where it trips Claiming needs enforced single sign‑on and SCIM or just‑in‑time provisioning, and anyone not in the identity provider is deactivated at the 30‑day deadline.12
What we do We list who already uses Claude on your domain before claiming, so nobody loses their work on the deadline.
Set limits for the organization, its groups and each person; usage-based plans are uncapped by default.10
Where it trips With no limit, spend has no cap, and hitting the organization’s limit cuts off everyone at once.2, 10
What we do We set limits per group from the first weeks of real use, with room under the organization’s cap.
Set retention, audit logs and Compliance API access, and turn on connectors from an approved list.10
Create custom roles and groups, move a pilot group to them, then turn features on for the organization.10
Where it trips Moving people to custom roles before their groups are covered removes their access, and role changes take up to 15 minutes.4
What we do We build the group coverage first and move people in small batches.
Confirm the seats, and have engineers sign in with their Claude account through single sign‑on.10
A pilot, then department by department, then everyone, with training and champions.10
What we do This is the coaching we do: each team on its own work, with a champion inside it.
What a buyer asks first, from Anthropic’s own pages as read in October 2026.
Where it runs
An organization on claude.ai, run from its Organization settings.3 Sales-assisted plans are also sold through AWS Marketplace.11
Price
$20 a seat a month billed annually, from 20 seats self‑serve or 50 through sales. The seat includes no tokens; all use is billed at API rates.1, 11
Data residency
An owner can keep inference in the US, billed at 1.1 times on newer models. It controls where requests run, not where data is stored, and no storage region is stated for Claude Enterprise.13
Your own keys
Keys in AWS, Google Cloud or Azure. Once attached they can’t be removed or replaced, and they turn off some features, among them Claude in Slack and Claude Code on the web.6
HIPAA and FedRAMP
A BAA on Enterprise once the primary owner turns on the HIPAA‑ready setup; the mobile app and some newer features are not covered.14 FedRAMP Moderate is in process, with no date given.15
Audit
Audit logs export the past 180 days and carry IDs, not chat content.16 The Compliance API reads activity, chats, files and projects.11
Claude Enterprise is strongest where writing, analysis and code matter most: Claude Code is in every seat.9 It sits outside both suites and reaches Google Workspace and Microsoft 365 through connectors.7 Its billing suits a company ready to watch use, since the seat buys access and every token costs extra. If the whole company lives in Microsoft 365 or Google Workspace and wants one assistant for everyone, Microsoft 365 Copilot or Gemini Enterprise start closer to your data. If you need FedRAMP today, Copilot17 and the FedRAMP version of ChatGPT Enterprise18 are further along. Our comparison sets the four side by side, and where the data can’t leave the building, private AI on your own hardware fits better than any of them.
The way every enterprise AI engagement runs: four stages, one team throughout, starting with a call.
A $20 seat billed annually, plus every token at API rates. On self‑serve plans usage is prepaid as credits and stops when they run out, and owners set spend limits for the organization and each person.1, 2
Team stops at 150 seats and has no SCIM, audit logs, Compliance API, custom retention or HIPAA.1 Moving from Team to Enterprise yourself can’t be reversed, and some features, such as skills and code execution, are off after it until turned on again.19
Yes, in the single Enterprise seat: in the terminal, the IDE, the desktop app, the web and Slack. Its usage is billed at API rates like everything else.9
Yes, and it is the most common place we start. We review the setup as it stands, fix what keeps people from trusting it, join it to your systems, and coach each team on its own work.
We set up all four, and compare them side by side.
Each fact on this page, read from the vendor’s own pages in October 2026.