The New Economics of Microsoft Copilot: Managing Seats and Consumption

Microsoft’s workplace AI pricing has officially split in two. Microsoft 365 Copilot remains a fixed per-user subscription, while agentic capabilities like Copilot Cowork are billed separately through consumption-based Copilot Credits. For IT procurement leaders, this isn’t just another licensing change. It’s the introduction of variable AI costs alongside fixed subscriptions, requiring a new approach to budgeting, governance, and contract strategy.

What Changed in Copilot Pricing

Copilot Cowork, Microsoft’s agentic AI system, reached general availability in June 2026. Two commercial facts define it. First, Cowork requires a Microsoft 365 Copilot license as a prerequisite (list price of $360 per user per year). Second, and more consequentially, the subscription includes no Cowork entitlements at all. Every task Cowork performs is billed on top of the license, in proportion to the work performed.

The billing unit is the Copilot Credit. This is the same currency Microsoft already uses for Copilot Studio, and now the common meter across Copilot Cowork, Work IQ APIs, Dynamics 365 agents, and Power Platform AI workloads. Credits pool at the tenant level, and an organization’s total cost is the sum of credits consumed across all these experiences.

Meanwhile, Microsoft has stated its continued commitment to the per-user subscription. Copilot Chat, in-app Copilot experiences, and native agents such as Researcher and Analyst remain covered by the seat license with no incremental charge. But the direction of travel is clear: the highest-value agentic work sits on the consumption meter.

Copilot Credits usher in a new chapter in Microsoft pricing history.

While Microsoft remains “committed” to the per-user subscription for some Copilot experiences, the direction of travel is clear. The highest value agentic work sits on the consumption meter.

How Copilot Credits Are Priced

Microsoft offers two ways to buy Copilot Credits. Pay-as-you-go bills monthly at $0.01 per credit with no upfront commitment. Or organizations can prepay for a one-year pool of credits through the Copilot Credit Pre-Purchase Plan (P3). The catch? Any unused credits expire at the end of the term.

The P3 discount schedule is worth reading closely, because the headline discount is far steeper than what most organizations will actually qualify for:

The 20% discount tier requires a $3M annual credit commitment. But the reality is a typical enterprise pilot lands in single-digit discount territory. The practical implication: P3 is a commitment decision far more than it is a savings decision, and the expiration clause converts any over-forecast directly into waste.

The Copilot Credit Pre-Purchase Plan (P3) comes with serious caveats.

First, any unused credits expire at end of term. Second, most customers won’t qualify for a meaningful discount anytime soon.​

What Drives Copilot Consumption

Credit consumption per task is variable, driven by four cost components:

Microsoft does not publish deterministic per-task pricing. Instead, its Copilot Credits Guide (June 2026) frames consumption through illustrative scenarios: a light task such as a recurring weekly status draft runs roughly 70–200 credits; a medium task such as assembling a customer-meeting briefing from emails, calendar, CRM, and file context runs roughly 400–600 credits; and a heavy task such as analyzing six months of product usage data into a leadership-ready analysis exceeds 1,500 credits. In dollar terms at pay-as-you-go rates, that is roughly $1-$2 for light work, $4-$6 for medium work, and $15 or more for heavy work – per task.

Two properties of this model deserve emphasis. Consumption is userinitiated and effectively unbounded. Nothing in the license structure caps what an enthusiastic user can spend. And because credits also meter Work IQ API calls, Copilot Studio agents, and Dynamics 365 agents, the tenant level pool becomes a shared budget across workloads that are typically owned by different teams.

What actually drives Copilot consumption?

Microsoft frames consumption in overlysimplistic terms – light, medium and heavy tasks. But the reality is consumption is userinitiated and effectively unbounded.

There are multiple factors that must be analyzed to really understand (and thus forecast) consumption costs. It’s not as simple as Microsoft would have customers believe.

Sizing the Cost Exposure of Microsoft Copilot Pricing Model Changes

Consider an enterprise that has licensed Microsoft 365 Copilot broadly and enables Cowork for a 150-user population across sales, marketing, and management. Assume moderate adoption: each active user runs four light tasks, two medium tasks, and one heavy task every other month.

Using the midpoints of Microsoft’s illustrative ranges and pay-as-you-go pricing, the monthly picture looks like this:

Under these moderate assumptions, consumption adds roughly 75% on top of the license cost. That’s about $41,000 per year in usage against $54,000 in seats. Annualized consumption of roughly 4.1 million credits would qualify for only the 7% P3 tier, worth under $3,000 per year.

And these are planning-level assumptions built on Microsoft’s own illustrative figures. Actual consumption will vary with workflow complexity, and adoption curves for tools of this kind rarely stay flat. If usage doubles (an entirely plausible scenario), the consumption line exceeds the license line.

Get ready for bill shock.

Under the most moderate assumptions, Microsoft’s new Copilot pricing approach could easily add 75% on top of the license cost.

Don’t Forget Azure: The MACC Intersection

Copilot Credit spending also counts toward your Microsoft Azure Consumption Commitment (MACC), which can work for or against you depending on your position. If you have unused MACC capacity, Copilot Credits are a viable lever to reduce shortfall risk before renewal. But if you’re negotiating a new commitment, Microsoft now has another consumption stream to justify a larger MACC. The challenge is that Copilot Credit usage is still highly unpredictable.

Our advice is the same as with any Azure commitment. Sizing must be based on proven consumption data, not adoption forecasts. Twelve months of actual Cowork usage is evidence. Early-stage credit projections are assumptions, and they shouldn’t be treated as committed spend.

Governance Before Enablement

Microsoft has, to its credit, shipped meaningful FinOps controls alongside the meter. Copilot Credit usage is managed centrally through the Microsoft 365 admin center, where administrators can monitor spend, configure spend policies, set usage thresholds, and allocate credits across the organization. The existence of controls, however, is not the same as a governance posture.

Before enabling Cowork, organizations should have answers to a short list of questions:

• Which personas get access, and against which sanctioned scenarios?
• Who owns the approval path for expanding either? What spend thresholds and alerting are configured before the first credit is consumed?
• Who owns the tenant-level credit pool when Cowork, Copilot Studio, and Dynamics 365 agents all draw from it across different budget lines?
• How will task-level value be measured? A $5 medium task that saves an hour is a good trade, but someone must own that justification.

The organizing principle we recommend is the one we apply across public cloud sourcing generally: govern variable spend separately from predictable spend.

Strong Copilot cost management borrows from cloud governance best practices. Govern variable spend separately from predictable spend.

Negotiation Implications

For organizations approaching an EA renewal or a Copilot expansion decision, the credit model changes the negotiation surface in several ways:

• Resist bundling a credit commitment into the deal before usage data exists.
  Pay-as-you-go carries a modest price premium over P3, but at pilot volumes that premium is small, and it buys the option value of committing later against real telemetry. Optimize first, commit last.
• Treat the P3 expiration clause as a risk to be negotiated or sized around.
  An over-forecast pool that expires is a 100% loss on the unused portion, which dwarfs a 5-10% tier discount.
• If Copilot Credits are being positioned as MACC-eligible, insist that the forecast be labeled as an assumption and stress-tested.
  Recognize that Microsoft’s fiscal-year timing, ECIF-style incentives, and milestone structures remain available levers here just as they are in any Azure commitment negotiation.
• Benchmark the credit economics themselves.
  Bundle pricing already discounts to $0.008 per credit in market, and as adoption data matures, fair market value for committed credit pools will become as benchmarkable as any other consumption construct.

The Bottom Line

The seat-plus-consumption model should not be treated as a pricing footnote. It is the template for how Microsoft intends to monetize agentic AI across its entire stack. Organizations that treat Copilot Credits as a minor add-on will discover the variable meter the hard way.

Organizations that put governance in place before deployment, model usage based on their own workforce instead of vendor assumptions, and delay commitment decisions until real consumption data exists will be best positioned to capture the productivity upside without taking unnecessary financial risk.

Where NPI Can Help

NPI works on the buy side only. We advise enterprise IT and procurement teams on Microsoft commercial strategy, grounded in fair market value benchmarking across thousands of enterprise transactions. On Copilot Credits specifically, clients engage us in four ways:

• Exposure modeling and business-case validation. We help clients pressure-test Microsoft’s illustrative estimates against their own user populations and workflows, so the cost of enablement is understood before the first credit is consumed rather than discovered on the first invoice.
• Fair market value benchmarking. We assess whether proposed Copilot license pricing, P3 credit commitments, and bundled constructs are competitive against comparable enterprise deals. Based on this analysis, we set defensible targets before you respond to a quote.​
• Commitment sizing and MACC strategy. We help separate confirmed consumption from working assumptions, right-size credit and Azure commitments accordingly, and structure milestones, timing, and incentive levers so the commitment serves your adoption curve rather than Microsoft’s fiscal calendar.
• Consumption governance. Formal governance frameworks for Copilot Credits are still emerging across the industry. NPI brings the perspective we’ve developed governing variable Azure spend to help clients think through ownership, spend policies, and thresholds for the tenant-level credit pool before enablement outpaces accountability.