~/opinions/read-small-print-microsoft-copilot-cost-chart
// opinion · 18 Nov 2026

Read the small print on Microsoft's Copilot cost chart

Ask four questions before Microsoft's $30, $81 and $205 chart goes into a business case.

the stance

Microsoft's chart is a vendor-funded estimate built on modelled workloads at list prices, and Microsoft's own write-up says the results are directional. It's a fair reason to work out what AI costs your own organisation. It shouldn't be the number in a business case until someone has rerun the sums on the organisation's own tasks and prices. This is Cliff's judgement, from two decades of IT delivery experience.

what's uncertain

Figures and method come from Microsoft's published summary of the GigaOm report, GigaOm's disclosure and GeekWire's report of the keynote, checked 18 November 2026. This piece works from Microsoft's summary and not from the full report text. The summary doesn't name the ChatGPT and Claude plans beyond Enterprise, give the Copilot Credit price it used, or show how the usage charges for the other two products were worked out.

what would change the view

A rerun by someone Microsoft didn't pay, or the same method applied to an organisation's real tasks at its negotiated prices, would make the comparison worth quoting. Published Copilot Credit rates for Cowork and Autopilot jobs would make the advanced-user figure checkable.

Microsoft's Ignite 2026 keynote on 17 November showed a chart pricing an everyday business user at $30 a month on Copilot, $81 on ChatGPT Enterprise and $205 on Claude Enterprise. Before that chart goes into a budget paper, ask who ran the study, what it measured, which prices it used and what your own usage would add. Microsoft's own write-up answers the first three.

The figures and method come from Microsoft's post on the study, published 16 November 2026, GeekWire's report from the keynote and GigaOm's report page, all checked 18 November 2026. What to do with the chart is my judgement, from two decades of IT delivery experience.

Who ran the study?

Microsoft did. GigaOm's report page says Microsoft commissioned and paid for the report, the two companies co-authored it, and Microsoft conducted the testing and the cost analysis. GigaOm helped design the research and validated the method, the results and the costs.

The numbers may well hold. They're still a supplier's estimate of its own product against two rivals, and that label should travel with them into any paper.

What did it measure?

The testing ran in October 2026 on 37 prompts across Copilot Chat, Cowork and Copilot in Word, Excel and PowerPoint, with each product given the same fictional company and data. From those runs Microsoft modelled two people. The everyday user does 100 tasks a month, such as finding information, preparing for meetings, analysing data and drafting content. The advanced user does 140, which adds 20 agentic workflows where a multi-step job is handed to an agent. For that advanced user the chart reads $96 for Copilot, $149 for ChatGPT and $299 for Claude.

A run passed only if it met every item on a checklist written before testing. Microsoft's summary says quality came out broadly similar across the three products over the whole workload, with results varying by task.

Which prices did it use?

The study used October 2026 list prices, or assumed prices where no list price is published, and left out negotiated discounts. Copilot's everyday chat and Office use sits inside the $30 licence, while the everyday ChatGPT and Claude figures have usage charges added to their subscriptions. Microsoft's summary says the other two subscriptions can cost about $10 less than Copilot's, and that usage charges wiped out that gap within the first week of modelled use.

Microsoft calls the results directional, meaning a rough guide, and says real workloads and negotiated prices will differ. Both Microsoft and GigaOm recommend running the evaluation on your own workloads and pricing, and that's the sentence I'd quote.

What would your own usage add?

The $30 covers what Microsoft calls everyday AI. Microsoft Learn lists Cowork, Copilot Code, Autopilot and the newest frontier models under usage-based billing, paid in Copilot Credits on top of the licence, which is why the advanced Copilot user costs $96 in Microsoft's own chart. The summary doesn't give the credit price behind that figure.

Microsoft announced a way to watch that spend at the same event. Agent 365 Cost Management gives admins one view of usage-based AI services, with spending policies, limits and alerts. The Ignite 2026 Book of News lists it as a preview, with forecasting limited to Frontier customers in November 2026.

Before anyone signs, I'd ask for two numbers of your own: what a month of agent work costs in credits on your real tasks, and what you'd pay for each of the three products after negotiation. A small pilot group with a spending cap would produce the first.

If the chart turns up in a deck you're reviewing, put Microsoft's footnote back under it: commissioned by Microsoft, modelled workloads, list prices, October 2026.

Sources

Microsoft Copilot Blog, What AI at work really costs, 16 November 2026. Figures, method, user definitions, pricing basis and Microsoft's caveats. Checked 18 November 2026.

GigaOm, Demystifying the TCO of Enterprise AI. The disclosure of who commissioned, paid for and ran the study. Checked 18 November 2026.

GeekWire, Microsoft jabs at OpenAI and Anthropic, touting cheaper, customizable AI models of its own, 17 November 2026. The chart shown in the keynote. Checked 18 November 2026.

Microsoft Learn, Understanding the user subscription license (USL) and usage-based billing (UBB). What the licence covers and what runs on Copilot Credits. Checked 18 November 2026.

Microsoft, Ignite 2026 Book of News. Agent 365 Cost Management and its preview status. Checked 18 November 2026.