← All posts
Cost · Economics

The per-seat AI bill that never stops growing

Altern8 AI · 16 July 2026 · 5 min read

Per-seat pricing feels reasonable when you buy it. You’re paying for the people who use the tool — what could be fairer? The problem is what happens next, as AI stops being a tool a few people use and becomes something everyone uses all day.

Cloud AI products — Microsoft Copilot, ChatGPT Enterprise, Claude for Business — are almost all priced the same way: a fee per user, per month, indefinitely. That model has a structural feature the sales demo never dwells on: your cost scales with your headcount, not with the value you get.

The line goes up and to the right — and it’s your bill

Consider how AI adoption actually unfolds inside an organisation. It starts with a pilot team. It works. Word spreads. Six months later, three departments want it. A year later, leadership wants everyone to have it, because being the employee without AI is starting to look like being the employee without email.

Every one of those new users is another seat, another monthly fee, forever. The better the tool works, the more people want it, the more it costs — and there is no ceiling. You are, in effect, taxed on your own successful adoption.

You are taxed on your own successful adoption. The more AI helps, the more it costs — with no upper bound.

For a 200-person firm that’s a meaningful annual line item. For a 2,000-person firm it’s a budget conversation at board level. And it recurs every year, rising with every hire.

The hidden multipliers

The headline seat price also isn’t the whole cost. A few things tend to compound it:

A different cost structure

There’s another way to price AI that changes the shape of the curve: pay for the infrastructure, not the seats.

When AI runs on infrastructure you own or control, the dominant cost is the hardware and hosting — which is a function of usage volume, not headcount. Adding a user doesn’t add a licence fee. Once the infrastructure is provisioned for your load, more people using it more often doesn’t send the bill up in the same linear way.

The two models diverge over time. Per-seat cost rises with every hire and every year. Fixed-infrastructure cost is dominated by capacity you provision once and grow deliberately. The more heavily an enterprise uses AI, the more the fixed model’s relative advantage compounds — the opposite of the per-seat curve.

We’ll be honest about the caveats, because the comparison is often oversold. A fixed-infrastructure model has real costs the seat price hides: you (or your vendor) provision and maintain the hardware, and heavy usage needs enough capacity to serve it well. The economics favour ownership most clearly at scale and with sustained usage — a five-person team of light users may find cloud seats perfectly economical. The point isn’t that fixed cost is always cheaper; it’s that the curves point in opposite directions, and for an enterprise expecting deep, broad AI adoption, that direction matters enormously.

The question to actually ask

The useful question isn’t “what’s the seat price?” It’s: where do we expect our AI usage to be in three years, and which cost curve do we want to be sitting on when we get there?

If the honest answer is “everyone, all the time,” then a model whose cost rises with every one of those people, forever, is worth examining before it’s locked in across the whole organisation.

Cost comparisons here are directional and illustrative. Actual costs depend on vendor pricing, your usage, your infrastructure choices, and your scale — model your own numbers before making a decision.

Model the fixed-cost alternative

Altern8 AI runs on your own infrastructure at a fixed cost — grounded, cited answers across every department, without a per-seat fee that grows with headcount.

Talk to us →