Microsoft’s “old-school” seat pricing is a seat-of-the-pants strategy that ignores AI economics

By Alan Jacobson, AI Economics Strategist

Microsoft’s decision to double down on seat-based pricing for AI tools may feel familiar to enterprise buyers. But familiarity does not make it correct.

The company’s new E7 bundle—combining Office 365, Copilot, security software and new agent-management tools—extends the traditional enterprise software model: charge a fixed monthly price per user.

That model worked in the pre-AI world.

It does not work in the AI world.

AI systems do not behave like conventional software. Their costs are not tied to the number of employees using them. They are tied to compute—and compute varies dramatically depending on how the system is used.

Flat-rate pricing, whether sold as a subscription or a seat license, simply ignores that reality.

Flat-rate pricing does not scale in enterprise AI

Flat-rate pricing can work in limited ways for consumers. Streaming services charge everyone the same price even though some customers watch far more than others.

But even consumer services rely heavily on throttling and usage controls to keep infrastructure costs under control.

Enterprise AI is different.

Inside a company of 10,000 employees, usage patterns vary wildly:

• One employee barely touches AI tools
• Another runs Copilot constantly
• A third builds autonomous agents that operate continuously

Each user generates dramatically different compute loads.

Seat-based pricing treats them as identical.

In other words, seat-based pricing is simply another flat-rate mechanism—and flat-rate pricing collapses when underlying costs vary dramatically between users.

Token billing doesn’t solve the problem

Some vendors argue that token-based billing solves this problem.

It doesn’t.

Tokens measure words. They do not measure compute.

And words are a terrible proxy for reasoning cost.

This is the problem the industry keeps avoiding.

Flat-rate pricing does not scale at the enterprise level. Token-based billing does not measure cost. Tokens measure words. Words are a terrible proxy for compute.

Consider these two scenarios.

A user talks to AI for thirty minutes about his girlfriend:

How she seems distant.
How she is slow to respond to texts.
How she is mysteriously unavailable.

The system dutifully transcribes every word, responds empathetically and consumes a massive number of tokens — all while avoiding the four words a human would scream immediately: SHE’S CHEATING ON YOU!

Now consider a three-word query:

“Is God real?”

Few questions demand more reasoning, context, philosophy and depth. Yet under token-based billing, that interaction may never recover the cost of compute.

That alone should end the debate over billing.

Tokens are not compute. They are a proxy—and an inaccurate one.

If you want to bill for cost, you must meter compute.

What Microsoft is really betting on

Microsoft’s new E7 bundle—priced around $99 per user per month—bundles Copilot with Office apps, security tools and agent management software.

The strategy is familiar: increase average revenue per user by bundling new capabilities into a higher subscription tier.

But the bet goes deeper than pricing.

Microsoft is effectively betting that AI workloads will behave like traditional SaaS usage—predictable, evenly distributed and manageable through seat licensing.

That assumption may prove incorrect.

AI workloads are inherently uneven. A small percentage of heavy users can generate the majority of compute demand.

If AI adoption accelerates while seat-based pricing remains fixed, cost-of-revenue may rise faster than subscription revenue.

That dynamic would appear in financials as AI Cost-Revenue Divergence (ACRD).

The signal analysts should watch

The question is not whether Microsoft can sell Copilot seats.

The question is whether the cost of inference will scale faster than Copilot revenue.

If it does, the signal will show up clearly in financial statements:

Revenue growth slows or stabilizes while cost of revenue accelerates due to AI inference workloads.

When that happens, margins compress.

That is the core signal analysts should watch.

Because if AI pricing remains tied to seats instead of compute, the economics of enterprise AI will eventually force a correction.

Seat pricing measures employees.

AI costs measure compute.

Those two numbers are not the same