98% of FinOps practitioners now manage AI spend (1). What does that mean for a Microsoft licensing team?
The FinOps Foundation’s State of FinOps 2026 survey puts 98% of 1,192 respondents managing AI spend, up from 63% in 2025 and 31% in 2024. On 25 August 2026, the Foundation published its own account of why managing that spend is not the same job as a licensing team tracking Copilot seats – a distinction that matters as Microsoft builds metering infrastructure deeper into the Microsoft 365 estate.
Tokenomics is not token-counting
The Foundation’s framing, from an insight piece by Mike Fuller, is worth considering: FinOps and Tokenomics aren’t the same discipline wearing different names, and neither is a subset of the other.
- FinOps answers the budget question; what you were charged, who owns the gap against forecast, whether the number moved
- Tokenomics answers the design question; which model, what caching strategy, whether a workload is worth running at its actual cost
Run with only one of these, and you get a confident answer to a question that was never the one that mattered.
For a licensing team, this split has a direct translation. Tracking what Microsoft 365 Copilot costs per seat is FinOps work. The seat cost is a fixed line on the invoice and doesn’t say anything about why that number is what it is, or what would make next quarter’s number different.
What this means for a Microsoft estate
Microsoft has been quietly adding the plumbing that makes this distinction concrete. Agent 365 (generally available from May 2026) carries its own licensing prerequisite, confirmed in Partner Center announcements in June and again in July 2026. That’s a second metering axis sitting alongside the Microsoft 365 Copilot seat license: one tracks who has access, the other starts to track the consumption of what the agents built on that access.
A licensing team watching only the seat count will see Copilot spend as flat or predictable. The same estate (read through a Tokenomics lens – what agents are doing, how often, at what model tier) can be moving in a completely different direction underneath that flat number. Neither reading is wrong; they’re answering different questions, and only one of them is currently inside most licensing teams’ remit.
What this means in practice
Cost ownership
Keep the FinOps view where it belongs: seat cost, run-rate, budget variance, owned by whoever already owns the Microsoft 365 spend line.
Tokenomics gap
Flag the Tokenomics gap to whomever owns AI deployment decisions. Model choice, agent sprawl and caching strategy sit outside a licensing team’s usual scope, but those are the decisions that will drive next year’s AI cost.
Cost separation
Treat Agent 365’s per-agent metering as the first concrete signal that ‘Copilot cost’ and ‘AI cost’ are no longer the same line item on a Microsoft estate.
Why this is worth watching now
In my view, the FinOps Foundation’s frameworks are provider-neutral, which leaves a gap for Microsoft-specific measures. No standardised, vendor-neutral set of Tokenomics KPIs for Microsoft estates has yet been published. Some consultancies have released competing frameworks, but none has become an accepted reference.
The useful work now is to define measures that link Copilot license cost, agent consumption and business outcome on a Microsoft estate. This will give licensing, finance and AI teams a clearer shared view of where costs arise and what value they deliver.
If you want to build that view across your Microsoft estate, talk to us about bringing FinOps and Tokenomics together.
