There is no Cowork SKU, so there is no single price. You buy a fixed license layer and a variable metered layer that bills when agents run. This is what each one costs, and which one your proposal quietly left out.
Copilot Cowork does not have its own SKU, so asking what it costs the way you would ask about a seat gives you a misleading answer. It bills as a license plus a meter.
That structure is the whole story. The license layer is fixed and easy to quote. The meter layer bills when agents actually run, has no rollover, and is the part a proposal tends to leave out.
The license layer is the entitlement that lets Cowork exist in your tenant. Microsoft 365 Copilot sits here at $30 per user per month, list, on an annual commitment. That figure is real, and it is the smallest part of the real cost.
If your negotiation is heading toward E7, the license layer is where that decision lands too. E7 is $99 against roughly $117 for the same components bought separately, which is a genuine bundle saving and also a governance floor rather than a finished price.
The meter layer is Copilot Credits. Credits are consumed when an agent executes work, which means your bill tracks adoption rather than headcount. Prepurchase runs at $200 for 25,000 credits.
Credits do not roll over. Whatever you buy for a period and do not consume is gone, and whatever you consume beyond it bills on. Security Copilot adds its own unit with overage at $6.
Across the Microsoft Copilot and agent proposals Morten Andersen reviewed between 2025 and 2026, roughly 30 enterprise deals, the seat count was almost always modelled carefully and the consumption layer almost never was.
The pattern was consistent enough to plan around: buyers negotiated hard on the per user rate, accepted the credit mechanism as an administrative detail, and met the real number two quarters later.
A seat is an entitlement you either use or waste quietly. A credit is a consumable that disappears at the end of the period whether you used it or not, so the same unused capacity that costs you nothing on a seat costs you cash here.
The practical effect is that credits punish both directions. Underestimate and you overrun. Overestimate and you have funded capacity that expired.
Prepurchase buys a block at a better unit rate. Pay as you go bills what you use with no commitment and no expiry risk. The right answer depends on whether you can forecast consumption yet, and in the first two quarters you usually cannot.
| Question | Prepurchase | Pay as you go |
|---|---|---|
| Unit rate | Lower | Higher |
| Unused capacity | Expires, you funded it | Never billed |
| Overrun behaviour | Bills on beyond the block | Bills continuously |
| Fits when | You have two quarters of real usage data | You are still discovering the curve |
| Negotiation value | Rate, and a rollover concession worth asking for | A ceiling and an alerting commitment |
E7 bundles the governance and security components that agent work tends to require anyway. At $99 against $117 in parts the arithmetic favours the bundle, provided you would genuinely have bought those parts.
Buy E7 to avoid deciding, and you have paid $99 to defer the decision. The bundle is only a saving against components you actually needed.
Security Copilot bills on its own unit with overage at $6, and agent activity generates security work. Model this alongside credits rather than treating it as a separate line, because the same rollout drives both.
The common advice is to pilot small, measure, then scale. That is right about sequence and wrong about what to measure. Most pilots measure whether people liked it, which is a satisfaction score, not a cost model.
Measure consumption per completed task instead, then multiply by the tasks you intend to move. A pilot that produces an enthusiasm number and no credits per task figure has told you nothing you can put in a budget, and it will price the wrong thing at scale.
Source: Redress Compliance advisory engagement file, 2025 to 2026. Microsoft Copilot and agent commercial reviews only.
For the full model, including the consumption worksheet, see what Microsoft Copilot Cowork really costs. Related tools: the Copilot credits cost calculator and cost per task model.
Primary sources: the Microsoft 365 Copilot product page, the Microsoft 365 Copilot documentation, Copilot Studio billing and licensing, and the Microsoft Product Terms.
There is no per user Cowork price, because Cowork is not a standalone SKU. You pay a license layer, where Microsoft 365 Copilot is $30 per user per month list on an annual commitment, plus a metered layer of Copilot Credits that bills when agents run. A quote giving one number has priced only the first layer.
No. Unused credits expire at the end of the period and consumption beyond the block bills on. That is why the meter penalises both over and under estimation, and why rollover is the most valuable concession to ask for in the consumption layer.
Pay as you go while you are still discovering the consumption curve, because nothing expires. Prepurchase once you have roughly two quarters of real usage data, because the unit rate is lower. Committing early to a block you cannot yet forecast funds capacity that will expire unused.
E7 is $99 against roughly $117 for the same components bought separately, so it is a real saving if you would genuinely have bought those components. It is a governance floor, not a finished price, and agent execution still bills separately through credits. Buying E7 to avoid making the component decision means paying $99 to defer it.
Credits consumed per completed task, then multiply by the tasks you intend to move. Most pilots measure user enthusiasm, which cannot be turned into a budget line. Without a cost per task figure you are scaling an unpriced meter.
Yes. Security Copilot bills on its own unit with overage at $6, and agent activity generates security workload. Model it in the same forecast as credits rather than as a separate line, because one rollout drives both.
Buyers negotiate the seat and sign the meter. The seat is the number in the proposal. The meter is the number in year two.
Independent buyer side advisory on Microsoft Copilot and agent spend: the license layer, the consumption meter, credit ceilings and rollover, E7 component testing, and taking Copilot into the wider agreement rather than a standalone order.
Independent. Buyer side. Written for CIOs, CFOs, and procurement leaders carrying Microsoft contracts. No vendor influence. No reseller margin.
Talk to the Microsoft buyer side practice. No obligation.
See Microsoft Advisory →
Independent buyer side advisory. No vendor influence. No reseller margin. We sit on your side of the table when you negotiate with Anthropic and the GenAI vendors.
Monthly. One email. Zero noise.
