Microsoft AI Costs: What’s Licensing, What’s Azure

Microsoft doesn’t bill all its AI products the same way. Some are traditional per-user licences that appear on your Enterprise Agreement, while others are consumption meters billed through Azure. Several products offer both options, which is what makes budgeting complicated.

The split creates a practical problem: licensing and Azure consumption typically sit in different budgets, require different approval processes, and land on different desks. If you’re forecasting Microsoft AI costs for next year, you need to know which products appear where.

The Two Billing Worlds

Microsoft has two distinct billing mechanisms for AI products:

The budgeting problem

When your CFO asks “what will Microsoft AI cost next year?”, the honest answer is: it depends on which products, and you’ll need two different numbers.

The licensing portion you can forecast with reasonable confidence. The Azure portion depends on how much your organisation actually uses the tools, and that’s harder to predict, especially in year one.

Which Products, Which Bill

The pattern:

The general rule is that per-user products are billed through licensing, while per-action or per-compute products are billed through Azure. Some products offer both options, and those hybrid products are the tricky ones. Security Copilot can be standalone (billed through Azure) or included with E5 (no separate charge). Copilot Studio can be capacity packs (licensing) or PAYG (Azure). You might have both models in the same organisation, which means you’re tracking costs in two places.

When Costs Stack

A single AI interaction can trigger costs across multiple billing systems, which makes forecasting difficult.

Example 1: M365 Copilot user builds an autonomous agent

A knowledge worker has an M365 Copilot licence. They use Copilot Studio to build an agent that monitors a shared mailbox and processes incoming requests automatically.

The licence is fixed at $30/month. But if the mailbox receives 500 emails per month, the agent consumes 18,500 credits ($185/month) on top of that. The $30 appears on your EA; the $185 appears on your Azure bill.

Example 2: Copilot Studio capacity packs with overage

You’ve bought 3 capacity packs ($600/month) expecting 75,000 credits of usage. A busy month pushes consumption to 90,000 credits.

The capacity packs are on your licensing agreement. The overage is PAYG through Azure. Two different budget lines for the same product.

Example 3: M365 Copilot user’s agent serves unlicensed colleagues

An M365 Copilot user builds an HR policy agent and shares it with the whole company. The builder has a licence, but most employees don’t.

The builder’s usage is covered by their licence. Everyone else generates PAYG consumption.

Example 4: Security analyst with both Copilots

A security analyst on M365 E5 uses both M365 Copilot and Security Copilot.

If they stay within the E5 allocation, Security Copilot costs nothing extra. If they exceed it, the overage hits Azure. The M365 Copilot licence is always on the EA.

Example 5: Power Automate flow triggers Copilot Studio agent

A Power Automate Premium user builds a flow that calls a Copilot Studio agent for document processing.

The flow licence is fixed. The agent consumption scales with how many documents get processed.

The lesson from these examples is that you can’t just add up licence counts and call it a forecast. You need to understand what happens when people actually use these tools, and how the costs flow to different billing systems.

The Visibility Gap

Here’s the practical problem: there’s no single dashboard showing all your Microsoft AI consumption.

• M365 Copilot usage appears in the Microsoft 365 admin centre

• Copilot Credits appear in the Power Platform admin centre

• Security Copilot SCUs have their own usage dashboard

• Azure OpenAI consumption appears in Azure Cost Management

• Dataverse capacity appears in Power Platform admin centre

As Jukka Niiranen, a Power Platform licensing specialist, observed: partners working on customer analytics “need to combine multiple data sources just to see what their organisation has consumed today.”

What the visibility gap means for budgeting:

The practical consequence is that you can’t pull a single report showing total AI spend. Different consumption types require different monitoring tools, and autonomous agents compound the problem because they generate consumption without any user activity to track. Month-end reconciliation requires manual effort across multiple portals, and until Microsoft provides better tooling, there’s no way around it.

What you can do:

In the meantime, set up spending alerts in both Power Platform admin centre and Azure Cost Management. Designate someone in your organisation to own cross-product visibility, and have them review consumption weekly rather than waiting until month-end. You’ll need to build a manual consolidation process, at least until Microsoft provides something better.

How to Budget

Given the split billing model, here’s how to approach forecasting:

For licensing-based products

Licensing-based products are straightforward to forecast. You count users, multiply by the price, and apply any negotiated discounts.

Build in a buffer for mid-year expansion, but the base cost is predictable.

For consumption-based products

Consumption-based products require usage estimation, and usage estimation is inherently uncertain until you have real data to work with.

The best approach is to start with pilots. Run a small deployment for two to three months, measure actual consumption, and then extrapolate with a safety margin built in.

As you plan, identify the cost drivers that will affect your consumption. Consider how many agents will run, how often autonomous triggers will fire, what interaction volume you expect, and which AI tools your agents will invoke.

You should also set spending limits from the start. Both Power Platform admin centre and Azure allow you to configure budgets and alerts, and you should use them before consumption surprises you.

Capacity packs vs PAYG

If your Copilot Studio usage is predictable, capacity packs ($200/month for 25,000 credits) save 20% versus PAYG ($0.01/credit). But unused credits expire monthly.

Recommendation: Start with PAYG to establish baseline usage, then shift to capacity packs once you understand consumption patterns.

Pre-Purchase Plan (P3)

For high-volume or variable usage, the Copilot Credit Pre-Purchase Plan offers a one-year upfront commitment with tiered discounts ranging from 5% to 20% depending on volume. Unlike capacity packs, P3 credits expire annually rather than monthly, which gives you more flexibility. P3 is also MACC-eligible, meaning the credits count toward your Microsoft Azure Consumption Commitment.

P3 makes sense if you’re spending $3,000+/year on Copilot Credits and can forecast with reasonable accuracy. Below that threshold, PAYG is simpler.

MACC interactions

If you have a Microsoft Azure Consumption Commitment, P3 credits count toward it. The MACC eligibility can help you hit targets while getting discounted AI capacity.

But be careful: MACC commitments create pressure to spend. Don’t buy P3 credits you won’t use just to satisfy a commitment.

Questions for Your Next Renewal

When you’re negotiating a Microsoft agreement that includes AI products, these questions will surface the information you need:

Need Help?

The split billing model for Microsoft AI creates complexity that traditional licensing expertise doesn’t fully address. If you’re trying to forecast costs, negotiate a Microsoft agreement, or make sense of how licensing and Azure consumption interact, get in touch.

We don’t sell Microsoft licences, so our advice is unbiased. We help you understand what you’re actually paying for.