An enterprise buyer’s briefing on Microsoft Copilot Cowork — generally available since June 16, 2026 — and the usage-based pricing model that most Microsoft 365 customers have not yet budgeted for.
Microsoft Copilot Cowork reached general availability on June 16, 2026, and it changes the economics of Copilot in a way the per-seat conversation does not capture. Cowork is an agentic AI worker: you hand it a task and it carries out multi-step work across Outlook, Teams, Word, Excel and your organisation’s data, pausing at checkpoints for your approval. The capability is genuinely useful. The pricing is genuinely different — it is metered, usage-based, and far harder to forecast than the flat seat license most finance teams are used to.
This is the same shift that has repeatedly caught enterprises out on consumption-based cloud services: a per-unit rate that looks trivial, attached to a usage pattern nobody has modelled, billed against a budget nobody set a cap on. The headline rate for Cowork is a single cent. The risk is what a few hundred power users do with it across a quarter. This briefing explains what Cowork actually costs, what drives consumption, and how to put governance and negotiation around it before the first invoice lands.
In this briefing
1. How Copilot Cowork Is Priced
Cowork uses usage-based billing at $0.01 per Copilot Credit. There is no separate per-seat Cowork fee. Instead, every task a user hands to Cowork consumes credits in proportion to how much work it does. A quick, single-step task on a light model with little context costs a handful of credits; a long, multi-tool task grounded across an entire mailbox and run autonomously in the background sits at the expensive end. The model is consumptive, like metered cloud, not subscription, like a seat.
Critically, Cowork is not standalone. It requires an active Microsoft 365 Copilot user license — roughly $30 per user per month on enterprise plans — and the metered usage accrues on top of that seat cost, not instead of it. So the true cost of Cowork is the Copilot seat plus consumption. See our Microsoft 365 Copilot license cost breakdown for the base number that sits underneath this.
One scheduling detail matters for 2026 budgets: tenants that had at least one user in the Frontier preview programme (March 30 to June 16, 2026) receive a grace period and will not be billed for Cowork usage until July 1, 2026. If that is you, the meter is effectively already running on usage you may not yet be tracking.
2. What Drives Credit Consumption
Credit usage is calculated from four inputs. Understanding them is the entire game when it comes to forecasting and controlling spend:
| Consumption driver | What it means | Why it matters for cost |
|---|---|---|
| Model use | Heavier, more capable models cost more per call. Cowork runs on advanced models, including Anthropic’s Claude. | Routing every task to the top model is the fastest way to overspend. |
| Context retrieval | How much of your mailbox, files and Teams history the agent must ground against. | Broad, whole-estate grounding is far more expensive than scoped context. |
| Tool calls | Each action — send an email, build a document, post in Teams — adds consumption. | Multi-action workflows multiply credits per task. |
| Runtime | Longer, autonomous tasks running in the background accumulate more credits. | “Set and forget” automations are the least predictable cost. |
The implication is that two users on identical seat licenses can generate wildly different Cowork bills depending entirely on how they use it — which is exactly why a flat per-seat mental model fails here.
3. The Budget Risk Nobody Warns You About
Flat per-user pricing is predictable; metered agent billing is not. A single power user automating dozens of multi-step workflows a day can consume credits at a rate that dwarfs their seat license, and consumption tends to concentrate in exactly the roles most eager to adopt the tool. Without spend caps and visibility, finance teams will see consumption spikes they did not forecast — the same dynamic that has burned enterprises on Copilot Studio message packs and Azure OpenAI consumption.
There are three specific failure modes to watch. Power-user concentration, where a small group drives most of the bill. Runaway autonomy, where long background tasks accrue runtime cost with no human watching the meter. And context bloat, where agents are pointed at the entire estate by default when a scoped data source would do the same job for a fraction of the credits.
Negotiating your Microsoft agreement?
We negotiate Microsoft 365, Copilot, and Azure commitments on a 25% gainshare basis — you keep 75% of every dollar saved, and if we don’t save you money you pay nothing.
4. Microsoft’s Levers — and Your Counters
Consumption pricing comes with its own sales psychology. The levers are softer than a seat negotiation but cost just as much if unmanaged:
| Microsoft’s lever | How it shows up | Your counter |
|---|---|---|
| “It’s only pennies” | A per-credit rate framed as trivial, with no aggregate forecast. | Model credits-per-task by workflow and multiply by realistic volume. Pennies times millions is a budget line. |
| No caps by default | Usage enabled broadly with governance treated as a later step. | Set tenant- and user-level spend caps and alerts before rollout, not after the first invoice. |
| Bundle into the Copilot commit | Cowork consumption folded into a larger Copilot deal with no unit visibility. | Demand separate, transparent unit economics for agent consumption so you can benchmark and forecast it. |
| Adoption-led expansion | Encouragement to enable Cowork org-wide to maximise “value”. | Pilot with a measured group first. Expand on evidence of ROI per credit, not enthusiasm. |
5. How to Control and Negotiate Cowork Costs
Treat Cowork like any consumption-based cloud service, because that is what it is. Four disciplines keep it in budget.
Govern before you scale. Set tenant- and user-level spend caps and alerts before broad rollout. The default posture should be a ceiling you raise deliberately, not an open meter you discover.
Pilot and meter. Run a controlled pilot and measure credits-per-task by workflow type, so you can forecast cost at full scale instead of guessing. This is the consumption equivalent of building a usage baseline before you negotiate.
Route models deliberately. Decide which workflows justify a heavy model and which can run on a lighter one. Model choice is the single biggest swing in credit consumption, and most of it can be set by policy.
Commit volume for a rate. Once you can forecast consumption, bring projected volume to your Microsoft EA renewal and negotiate committed-use economics rather than letting Cowork accrue at uncommitted list. Metered services are negotiable when you arrive with credible volume — and far cheaper to govern than to claw back.
6. The Executive Checklist
1. Know the meter is running. If you were in Frontier, billing starts July 1, 2026 — instrument usage now.
2. Cap before you scale. Tenant- and user-level spend limits and alerts go in before broad enablement.
3. Forecast from a pilot. Measure credits-per-task by workflow; do not budget from the headline cent.
4. Set model-routing policy. Reserve heavy models for tasks that justify them.
5. Demand unit transparency. Keep agent consumption priced separately from your seat commit so you can benchmark it.
6. Commit volume into the EA. Negotiate consumption economics with forecast volume rather than accruing at list.
A Worked Consumption Example
Because Cowork is metered, the only way to budget it is to think in tasks, not seats. The mechanics below are illustrative — Microsoft does not publish a fixed credit price per task type — but they show how consumption scales. A light task, such as drafting and sending a single email from a short prompt, uses one model call, minimal context and one or two tool calls: a handful of credits, effectively a few cents. A heavy task — research a topic across your mailbox and recent Teams history, draft a multi-section document, build a summary table and post it to a channel — stacks a capable model, broad context retrieval, several tool calls and meaningful runtime: easily tens to low hundreds of credits, i.e. dollars rather than cents.
Now scale it. Take 200 active Cowork users each running, say, five light tasks and one heavy task per working day. The light tasks are nearly free individually but add up across the population; the heavy tasks are where the bill concentrates. Across a month, that pattern can run from the low thousands to well into five figures depending on model routing and context scope — entirely on top of the ~$30/user seat cost. The figure is illustrative of the mechanics, not a quote, but the lesson is fixed: a few power users on heavy, broadly-grounded tasks can dominate your invoice. Forecast from your own pilot before you enable Cowork widely.
How Cowork Differs from Copilot Chat and Copilot Studio
Microsoft now has three distinct Copilot cost models, and conflating them leads to budget surprises. Copilot Chat is the conversational assistant included in your per-seat Microsoft 365 Copilot license — no incremental usage charge. Copilot Studio is the low-code platform for building custom agents, priced on a message-capacity model with tenant-level packs. Cowork is the autonomous task executor, billed per Copilot Credit on actual consumption. The seat license covers Chat; Studio and Cowork both add consumption costs on top, but on different meters. Any Copilot business case should account for all three rather than assuming the seat fee is the whole story.
Cowork and Your 2026 Budget Cycle
Because Cowork lands mid-year and bills on consumption, it does not fit neatly into a seat-based budget line set months ago. Treat it as a new, separate consumption line rather than folding it invisibly into Copilot seats. Set a deliberate starting monthly ceiling based on your pilot forecast, then review actuals monthly through the first two quarters of usage — consumption patterns take time to stabilise, and early adopters skew heavy. Reconcile each month against the per-task credit forecast from your pilot so variance surfaces while it is still small.
Two governance practices pay for themselves quickly. First, attribute Cowork consumption to departments or cost centres so accountability sits with the teams generating the spend; visibility alone tends to moderate usage. Second, set alert thresholds well below your ceiling so a runaway automation or a newly enthusiastic power user triggers a review before it triggers an overage. The organisations that stay in control of Cowork are the ones that instrumented it on day one, not the ones that waited for the first surprising invoice.
Frequently Asked Questions
How much does Microsoft Copilot Cowork cost?
Cowork uses usage-based billing at $0.01 per Copilot Credit. There is no separate per-seat Cowork fee, but it requires an active Microsoft 365 Copilot license (around $30 per user per month on enterprise plans) underneath it.
Do I need a Microsoft 365 Copilot license to use Cowork?
Yes. Cowork requires an active Microsoft 365 Copilot user subscription license; the metered Cowork usage accrues on top of that seat cost.
When does Copilot Cowork billing start?
Cowork reached general availability on June 16, 2026. Tenants that had a user in the Frontier preview programme (March 30 to June 16, 2026) receive a grace period and are not billed for Cowork usage until July 1, 2026.
What determines how many credits a Cowork task uses?
Four inputs: the model used, how much context the agent retrieves, the number of tool calls (actions) it makes, and how long the task runs. Heavier models, broader context and longer autonomous runtime all increase credit consumption.
Can I cap or control Copilot Cowork spending?
Yes. You can set tenant- and user-level spend caps and alerts, and you should do so before broad rollout. Piloting to measure credits-per-task and setting model-routing policy are the most effective controls.
Bottom Line
Copilot Cowork costs $0.01 per Copilot Credit on top of a required ~$30/user/month Microsoft 365 Copilot license, with Frontier tenants billed from July 1, 2026. The headline rate looks trivial; the real exposure is unbudgeted consumption at scale, concentrated among power users and long autonomous tasks. Put spend caps and governance in place now, forecast from a metered pilot, and negotiate committed volume into your Microsoft agreement. We manage this as part of broader Microsoft contract negotiations on a 25% gainshare basis. No savings, no fee.