Microsoft has refreshed the Copilot Studio Licensing Guide for April 2026. The change log is short, but the bigger story is what has happened around it over the last six months: the unit got renamed, two pre-purchase plans appeared, the default Dataverse allowance tripled, and Microsoft quietly unified its agent commercial model across Copilot Studio and Microsoft Foundry. If you are budgeting for agents in 2026, this is the moment to make sure you understand which lever you are actually pulling.

We will start with the mechanics — the four ways you can pay, with the math visible — then name the playbook Microsoft is running across them, and finish with what to push for if a Copilot Studio commitment lands inside your next EA or MCA-E renewal.

Copilot Studio (MCS) is Microsoft’s conversational-AI platform for building, customizing, and publishing agents — both the kind your employees use inside Microsoft 365 and the kind your customers hit on a website or in Teams. It sits on top of Power Platform, uses Dataverse as its data layer, and connects to the wider 1,200+ connector library.

The thing to anchor on for licensing: you do not pay per user, per conversation, or per agent. You pay for consumption, measured in Copilot Credits.

A Copilot Credit is Microsoft’s measure of the work an agent does to handle a turn — retrieving information, calling a tool, running a flow, invoking generative AI. Different actions cost different numbers of credits, so a simple FAQ-style turn is cheap and a multi-tool, RAG-heavy turn is not.

Two facts to keep in mind:

• The pay-as-you-go list price is $0.01 per Copilot Credit.
• Until September 1, 2025 the unit was called a “message”. Same plumbing, new name. If your internal models or vendor quotes still say “messages”, they refer to the same thing.

Everything else in the guide is just a different way of paying for that same currency.

There are now four mechanisms. They are not mutually exclusive — most enterprises will end up using two of them.

1. Pay-as-you-go (PAYG) meter — $0.01 per credit, billed in arrears.
No commitment, no caps, no surprises beyond the bill. You attach an Azure subscription to your Copilot Studio environment and Microsoft meters consumption monthly. This is the right starting position when you do not yet know what your agents will cost.

2. Copilot Credit pack (the monthly subscription) — $200 per pack, 25,000 credits, billed annually.
Effectively $0.008 per credit — a 20% discount on PAYG, but with a hard catch: unused credits do not roll over. Buy six packs (150,000 credits/month), use 90,000 in a quiet month, and you have just paid for 60,000 credits you never used. This SKU only makes sense once your monthly usage is reasonably predictable and consistently above the pack count.

3. Copilot Credit Pre-Purchase Plan (P3) — annual upfront, tiered discount up to 20%.
Launched October 2025. You buy a one-year pool of Copilot Credit Commit Units (CCCUs); each CCCU equals $1 of credit consumption, i.e. 100 credits. Discount tiers run from 5% at 3,000 CCCUs ($2,850) up to 20% at 3,000,000 CCCUs ($2.4m). Two structural advantages over the monthly pack: the discount scales, and the pool runs for a full year rather than expiring monthly. Once you exhaust the pool, billing falls back to PAYG automatically — no service interruption.

4. Microsoft Agent Pre-Purchase Plan (P3) — annual upfront, broader scope.
Added in February 2026 and the most strategically interesting move in the guide. You buy Agent Commit Units (ACUs), which work like CCCUs ($1 of usage = 100 credits) but draw down against eligible consumption across Copilot Studio and Microsoft Foundry — and, per Microsoft’s own commentary, ultimately Fabric and GitHub agentic services as well. Three tiers: 20,000 ACUs at $19,000 (5%), 100,000 at $90,000 (10%), 500,000 at $425,000 (15%). Smaller maximum discount than the Copilot Credit P3, but a single budget line for everything agentic.

A quick worked example. Suppose you forecast 3,000,000 Copilot Credits a year — say 250,000 a month across a handful of internal agents.

• PAYG: 3,000,000 × $0.01 = $30,000/year.
• Credit pack: 10 packs × $200 × 12 = $24,000/year — if you actually use the 250,000 every month. Underuse one month and the saving evaporates.
• Copilot Credit P3, Tier 3 (30,000 CCCUs): $27,900/year — 7% off PAYG, no monthly waste risk.
• Agent P3, Tier 2 (100,000 ACUs ≈ 10,000,000 credits): $90,000/year — only relevant if you also intend to consume Foundry/Fabric/GitHub agents against the same pool.

Show the math to your CFO before signing anything. The “right” answer depends entirely on the shape of your usage curve.

If your users hold the Microsoft 365 Copilot User SL ($30/user/month), they already have Copilot Studio use rights for employee-facing agents — Teams, SharePoint, M365 Copilot — at no extra charge. Agents built for those surfaces, operating under the licensed user’s identity, are covered.

The catch is “fair usage”. Microsoft does not publish the exact threshold, and reserves the right to update it as patterns emerge. In practice that means: do not plan a high-volume customer-facing agent on the assumption that M365 Copilot covers it. External channels (web, WhatsApp, Facebook) and any heavy programmatic consumption will route through Copilot Credits regardless.

Dataverse capacity. Every Copilot Studio tenant gets default capacity: 15 GB Database, 20 GB File, 2 GB Log. The Database default tripled from 5 GB to 15 GB in December 2025, which is genuinely useful — but agents that lean on RAG or store conversation history can blow through it. Add-ons run $40/GB/month for Database (or $30/GB/month at the 1,000 GB tier), $2/GB/month for File, $10/GB/month for Log. Track this in the Power Platform Admin Center; technical enforcement is real once you exceed purchased capacity.

Managed Environments. If you turn it on — and Copilot Studio expects you to for any non-trivial deployment — every active user in that environment must be covered by a standalone license or a PAYG meter. The limited Power Platform rights bundled in your Dynamics 365 or M365 licenses do not satisfy this. It is the most common compliance trap we see in MCS environments.

The April update itself is small: it adds the Microsoft Agent P3 plan to the document (the SKU was announced in February) and tidies a couple of feature references. The substantive changes happened earlier — the Credit P3 in October 2025, the messages-to-credits rename in September 2025, the Dataverse default increase in December 2025, the Agent P3 in February 2026. April just consolidates them into one document.

Read the four mechanisms together and the design intent is hard to miss. Each piece looks reasonable on its own. The composition is what to watch.

The rename was not just a rename. When “messages” became “Copilot Credits” on September 1, 2025, Microsoft also redefined what a single billable unit represents — credits scale with the work done by an agent, not with a single conversational turn. Any benchmark, ROI model, or vendor quote built before September is no longer directly comparable. The same agent doing the same job can consume materially more or fewer credits than it consumed messages, and most customers have not yet rerun their numbers.

Pricing complexity is a feature, not a bug. Four mechanisms — PAYG, monthly pack, Credit P3, Agent P3 — with three different commit terms (none, monthly, annual), two units of consumption (credits, $-denominated commit units), and overlapping discount curves. Every additional dimension of choice shifts the modeling burden onto the customer. Customers who cannot model accurately tend to over-buy out of caution, or under-buy and get caught on PAYG overage. Microsoft wins both.

The forecasting risk sits with you. Pack credits expire monthly, with no roll-over. CCCUs and ACUs both expire at the end of the annual term. Agents are by definition a new workload — usage is genuinely unpredictable in year one — and Microsoft is asking you to commit before you have a clean run-rate. The asymmetry is deliberate: under-use, you waste; over-use, you fall onto PAYG at the highest rate per credit.

The Agent P3 is a strategic steer, not a generosity. It looks like a discount, but the maximum is 15% versus 20% on the Copilot Credit P3. Microsoft is offering a worse rate in exchange for tying your agent budget to Copilot Studio and Microsoft Foundry and (per Microsoft’s own commentary) Fabric and GitHub agentic services. That is portfolio lock-in dressed as flexibility. If you are not actively planning to consume those broader services, the Agent P3 is the wrong instrument and the Credit P3 will save you more.

“Fair use” is unenforceable as a customer. Employee-facing Copilot Studio usage covered by the Microsoft 365 Copilot User SL is gated by a fair-use limit Microsoft does not publish and explicitly reserves the right to update. Practically, that means: build a high-volume internal agent on the assumption that M365 Copilot covers it, and Microsoft can reset the threshold underneath you. The bundle is excellent for low-volume employee scenarios. It is not a serious commercial foundation for anything heavy.

Managed Environments quietly converts a deployment decision into a licensing event. Switch it on — and the guide expects you to for any non-trivial deployment — and every active user in that environment must hold a standalone Power Platform / Copilot Studio license or be covered by a PAYG meter. The limited Power Platform rights bundled with M365 or Dynamics 365 do not satisfy this. We see this as the single most common compliance trap in real Copilot Studio environments, and it is the easiest one for an audit to surface.

Dataverse overage is engineered to be expensive. Subscription add-on Database capacity is $40/GB/month. The pay-as-you-go Database meter is $48/GB/month — 20% more for the same gigabyte. Customers who exceed entitled capacity unexpectedly pay the premium rate; customers who plan ahead pay the steady-state rate. The 5-to-15 GB default increase in December 2025 was a real concession, and worth taking — but it does not fix the price gap above the entitlement.

None of this is unusual. It is the standard Microsoft commercial pattern: a public unit price as the anchor, a complex set of mechanisms that reward forecasting accuracy you do not have, and softer levers (fair-use, Managed Environments, overage rates) that quietly shape behavior. The customers who do well here are the ones who recognize the pattern early and refuse to commit before they have data.

If a Copilot Studio commitment is going to land inside your next EA or MCA-E, treat it as a separate negotiation thread. Bundling it into the wider deal — without its own forecast, its own term language, and its own out-clauses — is how organizations end up paying for credits they never consume.

Six things we would push for, in order:

1. Get pay-as-you-go enabled now. Before the renewal conversation, you want three to six months of real consumption data. Microsoft’s account team will arrive with a recommended P3 size; your job is to be able to disagree from a position of evidence rather than feeling. PAYG is unglamorous and slightly more expensive per credit — but the data it generates is worth far more than the marginal cost in year one.

2. Use the Copilot Credit P3 against your MACC. The Copilot Credit P3 counts toward your Microsoft Azure Consumption Commitment. If you have a meaningful MACC in place — most enterprises with material Azure spend do — routing predictable agent consumption through the Credit P3 effectively pays down two obligations with one budget. Get Microsoft to confirm MACC eligibility in writing for your specific scenario, including how Foundry agent consumption interacts if you are also evaluating the Agent P3.

3. Push back on the Agent P3 unless your roadmap genuinely spans Foundry, Fabric, and GitHub. If you are buying Copilot Studio and only Copilot Studio, the Credit P3 gives you a higher maximum discount (20% vs 15%) and a cleaner commercial scope. The Agent P3 only earns its place if your agent strategy crosses Microsoft’s broader stack — and even then, model it carefully: a 15% discount on a much bigger commit is not automatically better than a 20% discount on a tighter one.

4. Lock pricing for the EA term. Microsoft has already flagged that the Microsoft Agent P3 pricing is “subject to change”. Get an explicit price hold for Copilot Credits and pre-purchase plan tiers across your EA term, and define what happens if Microsoft restructures the SKUs mid-term. We would not sign a multi-year EA without language that protects you against unilateral repricing of the agent commercial model.

5. Negotiate written scope for Microsoft 365 Copilot fair-use. Where you intend to rely on the M365 Copilot User SL to cover employee-facing agent usage — Teams, SharePoint, M365 Copilot surfaces — get Microsoft to confirm in writing what “fair use” means for your top three or four scenarios. This will not turn it into a hard contractual cap, but it materially shifts the conversation if Microsoft later asserts you have crossed the line.

6. Refuse the ramp. Microsoft account teams often propose a “ramp” structure — small Year 1 commit growing automatically across the term — to lock in volume before you know what you need. The cleaner posture is a flat baseline commit at the level your data supports, with PAYG as the spillover, and a true-up mechanism rather than a pre-baked ramp. You can always buy more. You cannot easily buy less.

Two timing notes worth keeping in your back pocket. Microsoft’s fiscal year ends June 30; the April-to-June quarter is the period when account teams have the strongest incentive to close. If your renewal cycle has any flexibility, that window is where additional discount tends to appear. And the broader pattern — Copilot Credit P3 in October 2025, Microsoft Agent P3 in February 2026 — suggests Microsoft will keep iterating on the agent commercial model. Anything you sign should assume more change is coming, not less.

For most enterprises, the right sequence is unchanged by the harder reading: start on pay-as-you-go to learn what your agents actually cost, move predictable baseline volume onto the Copilot Credit Pre-Purchase Plan once you have three to six months of data, keep PAYG attached as the spillover, and only consider the Agent P3 if your roadmap genuinely spans Microsoft’s wider agent stack.

What changes when you read the model critically is the negotiating posture. This is not a SKU you commit to before you have the data — and it is not a line item that should be quietly bundled into a renewal without its own term language. The mechanism Microsoft is selling rewards accurate forecasting and punishes the absence of it. The way to neutralize that asymmetry is to refuse to commit until your forecasts are real, and to bring your own numbers to the table when they are.

If you want a second pair of eyes on the math, the MACC interplay, or the renewal language, that is what we are here for.

Floris Klaver is CTO and co-founder of LicenseQ. We are an independent Microsoft licensing and contracting consultancy based in Vaassen, the Netherlands. No fees from Microsoft, ever.

Talk to us: floris.klaver@licenseq.com · Schedule a call