Copilot in SharePoint Went GA — And Quietly Started Metering You

The pricing model changed, and the announcement buried it

Copilot in SharePoint reached general availability and most of the coverage led with the features. The more consequential change is underneath: as of September 30, 2026, Copilot in SharePoint has usage-based billing.

Your per-seat licence still covers everyday work. Above it sits a new metered tier, priced in Copilot Credits at $0.01 per credit, with a default allowance of 4,000 credits per user per month — up to $40 per user per month on top of the licence you already bought.

For a 2,000-seat tenant, that default allowance describes a theoretical ceiling of $80,000 per month. Nobody will hit that. But most organizations currently have no idea what number they *will* hit, because the model changed from "count the seats" to "predict the behavior," and nobody budgeted for the second one.

---

What is actually metered

Three SharePoint capabilities consume credits at GA:

| Feature | What triggers it |

|---|---|

| Image generation and editing | Directly in document libraries and sites |

| AI-generated site analytics reports | Site owner asks for performance numbers |

| Advanced metadata autofill | Near-real-time column population |

Everyday Copilot work — asking questions of your content, summarizing a document, standard chat — stays on the per-seat licence.

The dividing line is roughly: reading and summarizing is included; generating and running at scale is metered.

---

The part that should worry you: who holds the spending pen

Look again at that table and ask who performs each action.

All three are site-owner surfaces. Not admin consoles. Not anything behind a Purview policy or an approval workflow. The people who can trigger metered spend are the people who run SharePoint sites — a population that in most enterprises is measured in hundreds, was never selected for financial judgment, and has never seen a purchase order.

There is no approval step between an enthusiastic site owner and a metered API call.

This is a genuine inversion. For the entire history of Microsoft 365 licensing, cost was an administrative decision — someone bought seats, assigned them, and spend was bounded by that assignment. Metered features move the spending decision to the edge while leaving the invoice in the middle. Nothing in your existing governance model was designed for that.

---

Why a flat per-action cost model will mislead you

Microsoft calculates each task's price from four inputs:

• Model use
• Context retrieval
• Tool calls
• Runtime

Two of those — context retrieval and runtime — scale with how much content the task has to read.

So "an analytics report" is not a fixed price. A report over a tidy 300-item team site and the same report over a 40,000-item document library with a decade of accumulated versions are not the same purchase, and the gap between them may be large.

This produces a counterintuitive result worth sitting with:

Your messiest sites are your most expensive AI sites.

The libraries with sprawling item counts, unpruned version history and no information architecture are exactly the ones where retrieval and runtime are highest. The same content hygiene problems that have been a performance annoyance for years now carry a direct, recurring, per-invocation cost.

That reframes a cleanup backlog most organizations have been deferring indefinitely. It stopped being a tidiness project and became a unit-economics project.

---

The exposure math, honestly

The $40/user/month figure is a default allowance, not a forecast. Treating it as a budget line would be as wrong as ignoring it.

What you actually need is three numbers:

1. Your ceiling. Seats × the per-user allowance. This is the number that should never surprise you, because it is arithmetic, not prediction. Know it before enablement.

2. Your enabled population. Not every user needs metered features available. The honest question is which roles genuinely benefit from image generation in libraries and on-demand analytics — and the answer is almost never "everyone."

3. Your actual consumption curve. Only measurable after the fact, which is precisely why the first two matter. A pilot on a known population for a known period gives you a real per-user rate to extrapolate from. Skipping the pilot means your first data point is an invoice.

---

What good governance looks like here

Decide the allowance deliberately. The 4,000-credit default is a number Microsoft picked, not one you chose. Review the pay-as-you-go configuration and per-user allowance in your billing settings and make it a decision rather than an inheritance.

Scope enablement by role, not by tenant. Metered features should reach the populations with a defensible use case first. Broad enablement is the expensive default.

Clean before you enable, on your largest sites. Given that retrieval and runtime drive cost, the version-history and item-count cleanup you have been deferring now pays twice — once in performance, once per invocation.

Monitor from day one, not from the first surprise. Establish the reporting cadence before broad rollout. Retroactive discovery of a consumption pattern is the failure mode to avoid.

Tell site owners the truth. The single cheapest control available is telling the people who trigger these features that they cost money. Most will moderate their own behavior immediately. Almost nobody does this, because the features do not say so at the point of use.

---

Our read

The capabilities are good and the metered model is defensible — generating images and running analytics over large corpora genuinely costs Microsoft money, and seat-based pricing cannot absorb that.

The problem is not the pricing. It is that the spend decision moved to the edge of the organization and the invoice did not. Every existing Microsoft 365 cost control assumes spend is bounded by assignment. Metered, self-serve, site-owner-triggered consumption breaks that assumption quietly, and the feedback loop is an invoice arriving weeks later with no attribution a site owner would recognize.

Organizations that treat this as a licensing update will find out what it costs retroactively. Organizations that treat it as a governance change with a billing consequence will set the allowance, scope the population, clean the expensive sites, and know their ceiling before the first report runs.

---

Want the exposure modelled against your actual tenant before you enable metered features? See SharePoint & Copilot and SharePoint Consulting, or get in touch.

Sources

• What's new in Copilot in SharePoint: October 2026 — Microsoft Community Hub
• SharePoint AI innovations hit GA, powering new Copilot app and agents — Microsoft Community Hub
• Copilot Credits explained: October 2026 rate card — Fiveforward