Before deciding how to buy Claude, it helps to see the whole menu. Anthropic sells Claude in two fundamentally different shapes, and most of the confusion inside buying committees comes from mixing them up. For an organisation of any real size the menu narrows fast, though, and this blog explains why it usually lands on Enterprise.
Purchase options
Seats are for people who use Claude directly, priced per person per month. Tokens are for software calling Claude through an API, priced by what it consumes. Everything on claude.com/pricing is one or the other.
Individual plans, for one person with a card:
- Free, at $0, covering chat, web search, memory and connectors
- Pro, $17 a month on annual billing or $20 monthly, which adds Claude Code, Cowork, Design, Science, projects and Research
- Max 5x, $100 a month, giving 5x Pro's usage with higher output limits and priority access at busy times
- Max 20x, $200 a month, at 20x Pro's usage. This is the only tier that reaches 20x; no Team or Enterprise seat matches it
Organisation plans, bought for a group:
- Team Standard, $20 per seat per month annually, $25 monthly, at 1.25x Pro's per-session usage
- Team Premium, $100 per seat per month annually, $125 monthly, at 6.25x Pro's per-session usage; Enterprise, $20 per seat per month plus usage at API rates, annual only
Claude API plans
The Claude API is priced per million tokens: Fable 5 at $10 in and $50 out, Opus 5 at $5 and $25, Sonnet 5 at $2 and $10, Haiku 4.5 at $1 and $5. Refer Anthropic pricing page for latest rates. Batch processing halves those rates and prompt caching cuts cached input by up to 80-90%, which together move the bill more than most negotiated discounts do.
One detail on Team that's easy to miss: you can mix and match seat types, giving Standard seats to occasional users and Premium seats to heavy ones inside the same organisation.
Where you buy it, and where it actually runs
Two separate questions, and people collapse them constantly. Where you buy affects procurement, contracting and which budget the spend lands in. Where inference runs affects your security review. They don't move together.
One clarification worth stating plainly, because it turns up in a lot of proposals: you cannot buy Claude Enterprise through Amazon Bedrock. Bedrock is a token-metered API service. It sells inference, not seats, and there is no mechanism in it for provisioning licences. When someone offers you “Enterprise via Bedrock,” they mean either Enterprise via AWS Marketplace or Bedrock API consumption, and those two things solve completely different problems for completely different people.
Team Standard vs Team Premium vs Enterprise
Read down the Enterprise column and almost nothing in it is about more Claude. What you're buying is control, evidence, contracting and cost control, which is the set of things that stops being optional the moment Claude graduates from pilot to payroll.
The economics usually favours Enterprise too
This one gets missed, and it's the argument I'd lead with in a finance conversation.
A Team Premium seat costs $100 a month, flat, whether the person hammers it daily or barely opens the tab. An Enterprise seat costs $20 plus whatever they actually consume. Break-even sits around $80 of tokens per user per month, and that is a lot of tokens. An engineer running Claude Code across a large monorepo typically lands well under it. A finance analyst asking a handful of questions a day spends a few dollars.
So for most organisations Enterprise isn't the premium option at all. It's cheaper for heavy users and much cheaper for light ones, because you stop paying for headroom nobody touched. There's also no allowance to divide up: every person's consumption is metered and billed at API rates, so one heavy user never eats into what anyone else can do. Add admin spend caps at user and org level and you get better economics and tighter control than a flat per-seat plan can offer.
The one gate is the 20-seat minimum. Below that the maths is academic, because you cannot buy Enterprise at all.
When Enterprise is the right call
Any one of these is enough on its own:
- Don’t worry about the 5 hour and weekly usage limit. Seat-based plans meter usage in rolling five-hour sessions with weekly caps layered on top, so a heavy day can lock someone out until a reset. Usage-based Enterprise has no plan or seat-level usage limits at all. Consumption is simply billed at API rates as it happens, so your people work at their own pace instead of rationing prompts on a Thursday afternoon.
- You've outgrown 150 seats. Team stops there, and there is no negotiating past it. In the other direction, you need at least 20 seats before Enterprise is available to you at all.
- Someone has to prove who did what. Audit logs and the Compliance API let a security team monitor usage programmatically, and Team cannot produce that evidence for a regulator or an internal auditor.
- Joiners and leavers are a live risk. SCIM matters far less at onboarding than at offboarding: it's what guarantees the person who resigned on Friday can't sign in on Monday.
- Not everyone should see everything. Role-based access with fine-grained permissioning, plus network-level controls and IP allowlisting.
- You're regulated. Custom data retention controls, plus a HIPAA-ready offering with a BAA. Note the BAA is available on the sales-assisted route only, not self-serve.
- Usage is uneven. Enterprise pools consumption across the organisation instead of capping per person, and admins set spend limits at both levels. Configure those in week one, because they don't arrive switched on.
- You need long context. 500K on the default model against Team's 200K.
- Procurement wants a real contract. MSAs, non-standard terms, PO-based invoicing, and tiered incentives against committed spend, all on the sales-assisted route.
In practice almost every organisation past a few hundred people crosses one of those lines, usually the audit or identity one, and usually the first time somebody asks a question nobody can answer. The honest exception is small teams: fewer than 20 seats, steady usage, nobody asking for SCIM. Below the 20-seat floor the decision is made for you anyway. Stay on Team, and move up when it stops being true.
Why buy Enterprise through AWS Marketplace
The product is identical either way. What changes is the commercial mechanics, and for most enterprises that's where the value sits.
- It draws down your AWS commitment. Marketplace purchases retire your EDP or PPA, typically capped at 25% of annual commit, and that cap is a negotiated term worth pushing on at renewal. On a $5M commit that's up to $1.25M of software bought with money you've already promised AWS.
- One condition governs it. Since May 2025, AWS only counts products fully deployed on AWS infrastructure, evidenced by the "Deployed on AWS" badge on the listing. Anthropic's carries it. Get eligibility confirmed by your account team in writing anyway, because this single clause decides whether the whole exercise works.
- Procurement gets shorter, since Anthropic is already onboarded with AWS, you take standardised Marketplace terms rather than negotiating a fresh MSA, and Claude lands on the invoice you already reconcile.
- Retirement is not the same as discount. Your EDP percentage doesn't apply to Marketplace spend. The discount on Claude comes from an Anthropic private offer instead, negotiated on seats, term and committed spend.
- A partner can usually do better. An authorised partner like cloud.in issues a Channel Partner Private Offer with custom pricing, one contract, and licences plus services in a single transaction. Note that AWS excludes professional services from retirement, so bundling implementation is procurement convenience rather than extra drawdown.
The short version
If Claude is going to more than a handful of people in a company that has auditors, an identity provider and a security review, Enterprise is the plan. Buying it through AWS Marketplace then converts committed spend you've already promised AWS into working software, on terms a partner can improve.
How Cloud.in helps
Cloud.in is authorised reseller for Claude on AWS, AWS Premier Tier Services Partner, the highest tier in the AWS Partner Network, and a validated Managed Service Provider. Reseller authorisation is what lets us issue a Channel Partner Private Offer (CPPO) directly, rather than handing you back to Anthropic to transact, and it puts you on the sales-assisted route by default. Premier isn't a badge you buy either: AWS audits delivery practice, certification depth and customer references before granting it, and only a handful of partners in India hold it.
Behind that sits 27+ years of AWS work, competencies in GenAI, Migration and Modernization, SMB, Education and Microsoft Workloads, Well-Architected Partner status, and 24×7 NOC and SOC operations from Mumbai and Pune.
On a Claude Enterprise decision, here's what we do that a licence reseller doesn't.
- Tell you which column of that table you belong in, over two weeks of discovery with engineering, security and finance in one room. If Team Premium suits you for another year, we'll say so.
- Model the line that ruins budgets. Seat cost is arithmetic; token consumption needs workload-level modelling, and you leave with a forecast built on your usage rather than a vendor's best case.
- Structure the commercials. Private offer or CPPO, retirement treatment confirmed with your AWS account team in writing, and existing discounts checked so they carry across, since they can't be applied retroactively.
- Land it properly. SSO and SCIM into your identity provider, role-based access on your real org chart, audit logging your security team will actually read, spend caps set before the first large invoice.
- Drive adoption. Four hundred licences and forty active users is the standard failure mode, and it has never once been a technical problem.
Most customers start with a twenty-five-seat pilot and a real cost model, then expand from what they learn.
Talk to Cloud.in and we'll get the first session in the diary.
The blog is written by Swapnil Vaidya (Lead Cloud Solutions Architect @ Cloud.in)
No comments:
Post a Comment