Did Microsoft Cancel Anthropic Licences? What Changed and How Claude Billing Works in 2026

Updated 25 July 2026: We found no primary evidence supporting the old claim that Microsoft cancelled internal Anthropic licences. Microsoft may or may not have made a private internal procurement change, but that remains unverified. The public evidence instead shows Microsoft expanding customer access to Claude across Copilot and Microsoft Foundry.

Verdict: the cancellation claim was unverified

Claims about a large technology company changing its internal AI subscriptions can sound convincing, particularly when token costs are part of the story. But without a procurement document, official statement or attributable reporting, they should not be treated as fact.

What can be verified is that Microsoft deepened its commercial support for Anthropic models. Claude arrived in Microsoft 365 Copilot products during 2025, reached Copilot Chat in June 2026 and became generally available through Microsoft Foundry later that month.

The official Microsoft and Claude timeline

September 2025: Claude joined Copilot products

On 24 September 2025, Microsoft announced Anthropic models for Researcher and Copilot Studio. Access initially required an organisation to opt in, giving administrators control over whether Anthropic models could be used.

This was an expansion of model choice rather than a replacement for Microsoft's existing models. It allowed organisations to choose a model according to the task, governance requirements and available product features.

June 2026: Claude expanded into Copilot Chat and Foundry

Microsoft's Microsoft 365 Copilot release notes record Claude becoming available as a model option in Copilot Chat on 16 June 2026.

On 29 June, Claude in Microsoft Foundry became generally available. In this arrangement, the service is hosted on Azure while Anthropic operates inference as a data processor and service-level agreement provider. Inference is the process of running a trained model to generate a response.

Microsoft also states that Global and US data zones, plus zero data retention for relevant workloads, are available. Those options need to be checked against each organisation's location, workload and compliance requirements.

Claude billing in Microsoft has three separate routes

The most important budgeting point is that there is no single universal Microsoft Claude price. Costs depend on which Microsoft product is being used and how the user or application reaches the model.

1. Microsoft 365 Copilot user licensing

Microsoft 365 Copilot is primarily a user-licensing route. Whether a particular Claude experience is available or included depends on the product, licence and user scenario.

A Copilot seat should not automatically be treated as an unlimited Claude API allowance. Administrators should use Microsoft's current licensing guidance to confirm which users and features are covered.

2. Copilot Studio and pooled Copilot Credits

Copilot Studio measures agent activity in Copilot Credits. Prepaid capacity is pooled across the tenant, while pay-as-you-go billing is also available. Bring-your-own Foundry models are billed separately.

Credit rates vary by feature. Some scenarios involving authenticated Microsoft 365 Copilot users are included, while other activity consumes credits from the organisation's pool. Microsoft's Copilot Studio usage documentation explains the applicable meters.

3. Foundry and Claude Consumption Units

Foundry is the token-based route. A token is a small unit of text processed or generated by a language model. Microsoft converts Claude token usage into Claude Consumption Units, or CCUs, using Anthropic's current rates for each model.

CCUs are metered hourly and invoiced monthly through Azure Marketplace. CCU billing applies automatically to new Claude deployments. Deployments created before CCU became generally available remain on their existing per-model plan unless replaced with a new deployment.

A practical monthly Claude budget framework

Measure Microsoft 365 Copilot seats

Monthly seat budget = eligible users x current Copilot licence cost, plus any prerequisite Microsoft 365 licences not already held. Add separately billed Studio or Foundry workloads.

Measure Copilot Studio activity

Estimated monthly credits = sum of feature events x the applicable credit rate

Measure real agent activity, including tests, failed runs and repeated actions. Compare the resulting total with the organisation's pooled capacity, then identify any connected Foundry models billed outside that pool.

Measure Foundry token consumption

Monthly model cost = ordinary and cached input, cache writes and output, each at the current rate for that Claude model. Apply Microsoft's CCU conversion and Azure invoicing rather than blending models into one average.

This workload-based approach also makes comparisons with other AI coding tools and human development costs more meaningful. The relevant question is not simply what a model costs, but what useful work it completes and how much review it creates.

UK and EU administrators need to check data controls

For Microsoft Online Services, Microsoft identifies Anthropic as a subprocessor. However, Anthropic models are currently outside the EU Data Boundary and applicable in-country processing commitments. They are disabled by default for EU, EFTA and UK tenants, so an AI Administrator or Global Administrator must opt in and select the users or groups allowed to use them.

Preview models labelled "with Data Retention" are separately controlled, use Anthropic's commercial terms and treat Anthropic as an independent data processor rather than a Microsoft subprocessor.

• Confirm the tenant setting, allowed users and applicable data zone.
• Check whether zero data retention applies to the intended workload.
• Review permissions before connecting internal files or business systems.
• Document the billing and processing route for each team, agent and application.

These checks are operational rather than theoretical. Model choice can affect data processing, invoices and supplier oversight even when everything appears through one Microsoft interface.

Practical ways to control Claude costs

• Start with measured pilots: capture runs, credits, input tokens and output tokens before forecasting wider adoption.
• Set Azure budgets and alerts: monitor consumption early instead of waiting for the monthly invoice.
• Reduce unnecessary context: sending entire documents or conversation histories increases input usage.
• Limit output length: ask for concise structured responses where lengthy prose adds no value.
• Separate testing from production: tag deployments so experimentation does not distort operational costs.
• Use the appropriate model: reserve more expensive or capable options for workloads that genuinely need them.
• Track human review: low token spend is not a saving if unreliable output creates additional checking and rework.

Infrastructure costs also have a physical dimension. My guide to AI, data centres and water use explains why environmental claims need the same evidence-led treatment as licensing rumours.

Frequently asked questions

Did Microsoft cancel Anthropic licences?

No primary evidence supporting that claim has been found. A private internal procurement change cannot be confirmed or ruled out from the public material, so it remains unverified.

Is Claude included with every Microsoft 365 Copilot licence?

Not necessarily. Availability and billing depend on the Microsoft product, licence and user scenario. Check the current licensing page and tenant controls.

Are Copilot Credits the same as Claude Consumption Units?

No. Copilot Credits meter Copilot Studio activity through prepaid pooled capacity or pay-as-you-go billing. Claude Consumption Units meter Foundry usage billed through Azure Marketplace.

What should businesses monitor first?

Track licensed users, Studio credits, Foundry tokens, connected data sources and human review time separately. That produces a far more useful budget than assuming all Claude access sits inside one Copilot subscription.

The verified record shows Microsoft expanding Claude access, not publicly abandoning Anthropic. For wider context, see my analysis of the OpenAI and Anthropic competitive landscape in 2026.