In this post
You've probably seen the headlines. As of 16 June 2026, Microsoft Copilot Cowork moved out of the Frontier early-access programme and into general availability, and it's now a billable, metered service. If your business was in the Frontier preview, there's a grace period: billing doesn't start until 1 July 2026, giving you a short window to set budgets and controls before the meter runs.
Running advanced AI models at scale isn't free, and somebody has to pay for all those tokens eventually. But the real story isn't that Cowork costs money; it's that businesses are entering a new phase of AI adoption. Until now, the focus has been on possibility: what can AI do, where can it create value, and how fast can we scale it? Now there's a sharper question on the table: how do we scale AI responsibly?
AI has never really been free. The difference is that we're now seeing the cost of consumption rather than it being hidden within licensing - Jay Fitzhenry, vCTO for AI at Node4:
As Jay points out, paying for what you consume isn't unusual - it's how cloud has worked for years. Storage, compute, APIs, and data processing have all followed consumption-based pricing because they scale with usage. AI is no different: every prompt and every response requires computational resources, so charging based on what's consumed is a logical evolution, not an anomaly.
The Capability Conversation is Becoming a Cost Conversation
Microsoft 365 Copilot hasn't changed. Copilot in Word, Outlook, Teams and Chat stays licensed on a predictable per-user basis. What's changed is Copilot Cowork.
The meter now runs when AI agents perform multi-step work on a user's behalf, building reports, pulling research from across systems, and producing stakeholder-ready content with a cost based on consumption. That means usage can vary significantly between employees, teams and workloads. And that's where it gets interesting, because for many businesses, AI is moving from experimentation into day-to-day operations.
How Copilot Cowork Pricing Actually Works
Copilot Cowork pricing is built on two layers:
- The Microsoft 365 Copilot licence (predictable, per user). This is the entry point, the same seat that powers Chat and Copilot across your apps. Microsoft publishes it in pounds: currently around £13.80–£16.10 per user/month (ex VAT), depending on your plan. It's the flat, forecastable part of the bill.
- Copilot Credits (variable, usage-based). The multi-step work Cowork does is billed separately in Copilot Credits, around a penny per credit (Microsoft lists pay-as-you-go at $0.01 per credit). Commit to a prepaid annual volume, and the cost per credit is lower, for more predictable spend as usage grows.
Each task's cost comes down to four things: the model it runs, the context it pulls in, the tools it calls, and the runtime it needs. A quick summary is cheap; a deep, multi-output report grounded across your whole mailbox sits at the expensive end. As a rough guide, a light task (a Monday status update) runs to a pound or two, a medium task (a meeting-prep pack with a doc and a deck) to a few pounds, and a heavy task (six months of usage data turned into a leadership report) more again.
You're not locked into one model, either, and model choice is one of the biggest cost levers. Right now, Cowork runs on Anthropic's Opus 4.8 and Sonnet 4.6, with Microsoft's own cost-optimised Cowork 1 model arriving soon for everyday tasks at lower credit consumption.
Why Visibility and Value Matter More Than Ever
The challenge with consumption-based AI is that cost isn't always obvious. Tasks run in the background, agents execute multiple steps, context windows grow and usage can climb without anyone watching the meter.
The controls are there: type /cost to see what a task costs in credits, and use the Cost Management dashboard in the Microsoft 365 admin centre to set spending limits at tenant, group and user level, with alerts and per-team reporting. Cowork is even off by default until an admin switches it on.
But Jay's view is that visibility is only half the story and that the model still has room to mature. He's candid that the pricing is still a little too broad; a simple cost per token feels generic, and he'd like to see more granular options as the platform matures. He's equally clear that businesses don't yet have the cost governance they'd expect from an enterprise platform; better visibility, forecasting and spending controls will be essential as adoption grows.
The bigger gap, though, is ROI:
The biggest surprise isn't the pricing model. It's how few businesses have measured ROI before the bill arrived." - Jay Fitzhenry, vCTO for AI
Because Cowork was effectively included for many customers, there wasn't always pressure to quantify the value it delivered. Now that there's a visible cost attached, those conversations are happening after adoption rather than before it. In Jay's view, there was an opportunity for customers and Microsoft alike to build stronger ROI narratives from day one: organisations that had measured productivity gains, time savings and business outcomes from the start would find today's pricing conversations far easier.
Not Everyone Should Be Using AI in the Same Way
One of the biggest mistakes businesses make is treating AI adoption as a blanket rollout. Different people work differently. A technical specialist analysing data, a salesperson preparing for customer meetings, and a senior leader reviewing performance will all consume AI very differently.
Before thinking about licences and credits, understand your workforce across four key personas:
- Corporate knowledge workers
- Customer-facing knowledge workers
- Technical workers
- Managers and senior leaders
Mapping how each group is likely to use AI creates a far stronger foundation for forecasting cost and spotting the high-value opportunities. (Microsoft's free Customer Cowork Estimator is a useful starting point for modelling this.)
Stop Thinking about Cost, Start Looking for Value
The question to ask isn't "how much will AI cost?" - it's "what value will AI create?".
The businesses seeing the strongest returns aren't the ones spending the least. They're the ones that:
- Apply AI intentionally.
- Know where it creates value.
- Understand which use cases deserve agentic AI and which don't.
- Have the visibility and controls to scale confidently.
The Node4 View
The arrival of consumption-based AI pricing shouldn't slow businesses down. If anything, it's a sign AI is maturing. The conversation is shifting from experimentation to accountability, from capability to outcomes. And that's exactly where successful transformation happens.
Our vCTO for AI, Jay Fitzhenry, puts it simply: despite the pricing change, his overall view hasn't shifted. Copilot Cowork still delivers significant value for the right use cases; it can save hours of effort, sharpen decision-making and lift productivity across the organisation. What changes now is that the value has to be measurable.
I still believe Copilot Cowork is worth the investment. The conversation has simply shifted from adoption to accountability." - Jay Fitzhenry, vCTO for AI
The next phase of AI won't be defined by who adopts it first. It'll be defined by who governs it best because achieving more with AI is about using it intelligently.
Get in touch with one of our AI experts today. We'll walk you through token usage and show you how to scale AI while staying cost-conscious.
