Microsoft Said the Quiet Part: Per-Seat Licensing Just Died on an Earnings Call

The most consequential pricing announcement of 2026 did not come from a press release. It came from a Microsoft earnings call, and the procurement teams who needed to hear it were not on the line.

Satya Nadella, on the record: “the basic transformation of any per-user business of ours, whether it is productivity, coding, or security, will become a per-user and usage business.” Amy Hood, on the same call: “a licensed business plus a consumption business applied far more broadly than I think people have thought about.” That is not a hint. That is the dominant enterprise software vendor on the planet announcing the end of the model the entire procurement function was built to negotiate.

GitHub Copilot moved to usage-based pricing on June 1, 2026. Dynamics 365 Customer Service already has roughly 60% of customers buying usage credits. Copilot credit consumption is up nearly 2× quarter over quarter. Per-seat is not gone, but its job has changed. It is now the packaging on the front of the box. The contents are prepaid consumption, and the bill that matters is metered.

The fiction the invoice was hiding

For two decades, “seat” was a useful fiction. It billed against headcount because headcount was the one thing IT, Finance, and HR could agree on. Nobody actually used a seat the same way; they all paid the same. The fiction worked because the marginal cost of an active user was rounding error.

AI broke the rounding error.

A power user of GitHub Copilot can burn 100× the tokens of a casual one. A customer service agent equipped with Dynamics Copilot can drive ten times the inference of a colleague who barely opens it. The cost of a “seat” is no longer a flat number plus rounding, it is a long-tail distribution where 5% of users drive most of the spend.

A vendor cannot run that business at flat per-seat prices without subsidizing the heavy users from the light ones, which means the light users are subsidizing the experience the heavy users are getting. Nadella and Hood just told their shareholders they are done doing that.

The cost-truth buyers do not see

While Microsoft was rewriting the contract, sites.diy was running a quieter experiment that should be required reading for every CFO. They sat a proxy server between developer machines and the coding tools they pay for, then measured what those tools actually cost at the API level for the same work.

The numbers are not subtle.

A $20-per-month Codex subscription performs work that, billed at OpenAI’s API rates, would cost roughly $536 per month. That is a 26.8× subsidy. Claude Pro, by contrast, costs about $0.744 per million tokens for the same workloads, 9.3× more per token than Codex’s effective rate. Cursor, according to Replit’s Amjad Masad, runs at negative 23% margins.

These are not pricing differences. They are funding decisions. The subscription tier you are paying is not a price, it is the vendor’s current willingness to absorb your usage in exchange for being inside your developers’ editors. That willingness has an expiration date, and Microsoft just told the market when its expires.

What “per-user and usage” actually means in a contract

There are three things changing at once, and the procurement team needs to see all three before signing the next renewal.

The seat is becoming an entitlement wrapper. Buying a seat no longer buys outcomes. It buys eligibility to consume from a pool, plus a bundled allocation that resembles the old fiction. When the allocation runs out, the meter starts.

The credit pool is the real budget line. Microsoft’s Dynamics 365 model is the template. You buy seats, you buy credits, the credits get burned at variable rates depending on which feature is invoked. Two customers with identical seat counts will diverge in spend by an order of magnitude, and the person who notices first will be the CFO, not the CIO.

The vendor sets the conversion rate. A “Copilot credit” is not a unit of work. It is a unit of vendor accounting. The vendor decides how many credits an action costs. The vendor decides when to repackage. The vendor decides which features cost more this quarter than last. Nothing in the seat-licensing playbook prepared procurement for negotiating against a unit they do not control.

The 2026 procurement question

The right question at the next renewal is not “how many seats?” The right question is “what is our consumption envelope, and what does the team’s actual usage curve look like inside it?”

Most enterprises cannot answer that question. They know how many seats they bought. They do not know which 5% of users are driving 80% of the consumption, which workflows are cheap, which workflows quietly burn through credits, and what the spend looks like if adoption goes from 30% to 70%.

This is the same gap I wrote about in the $7 Doritos repricing and the three-axis pricing problem, but it is now operationally urgent. The seat-licensing era let you negotiate once a year. The consumption era requires you to model the envelope before you sign, monitor the burn while you operate, and intervene before the agent quietly approves its own overage, the failure mode I covered in the agent budget self-approval problem.

What “do this now” looks like for CFO and procurement

You have until the next renewal cycle, which for most enterprises is between 60 and 180 days. Three concrete moves.

One: instrument the burn. You cannot negotiate a consumption contract you cannot measure. Before the next renewal, every AI tool in the stack needs a per-user, per-workflow consumption baseline. Not a vendor dashboard, your own measurement, in your own data, against your own definition of useful work. This is the measurement layer that separates the 5% getting substantial ROI from the 95% who cannot answer the question. It is also what I argued was the only durable moat when the underlying capability becomes commodity.

Two: re-bid the contract as a consumption envelope, not a seat count. Walk into the renewal with a usage curve and ask the vendor to price against the curve. Their preferred packaging is a credit pool plus seats. Your preferred packaging is a unit price per outcome with caps. The negotiation lives in the gap between those two packagings, and the side with better data wins.

Three: build the kill switch before you need it. Per-seat licensing was forgiving, overspending meant buying too many seats. Consumption is unforgiving, overspending means a runaway credit burn nobody noticed until the invoice. Every team using a metered AI tool needs a circuit breaker: a budget that cannot be exceeded without a human approval, an alert that fires before the budget hits, and a fallback workflow that runs without the metered tool when the budget is gone.

The honest summary

Microsoft did not change the model on May 1. They confirmed, on the record, that the model already changed. Per-seat is now the box. Consumption is the contents. The earnings call moved the conversation from “is this happening?” to “how fast does the rest of the industry follow?”

The answer is “fast.” Salesforce already shipped a three-axis pricing model. Anthropic’s enterprise pricing is consumption-shaped. OpenAI’s enterprise contracts have been negotiating against credit pools for two quarters. The holdouts are the vendors whose customers have not noticed yet, and they will move the moment Microsoft’s renewal cohort starts arriving with usage curves in hand.

The companies that win the next twelve months of AI procurement are the ones whose CFOs walk in already knowing what their consumption envelope looks like. The companies that lose are the ones who think they are still buying seats.

The earnings call already happened. The contract template already changed. The only variable left is whether your procurement team finds out before or after the next invoice.

---

This analysis synthesizes Clouded Judgement 5.1.26, The Death of Per-Seat Pricing? (Jamin Ball, May 2026), Coding Plan Comparisons Based on Actual Usage (sites.diy, May 2026), and Replit’s Amjad Masad on the Cursor deal (TechCrunch, May 2026).

Victorino Group helps enterprises model the consumption envelope before contract renewal. Let’s talk.