AI Pricing Just Got Harder to Predict
For twenty years, business software pricing was gloriously boring: a fixed amount per user per month, times your headcount, done. AI is quietly ending that. Microsoft 365 price increases have now taken effect across business and enterprise plans, with frontline plans rising more steeply, and more importantly some AI features are shifting to consumption pricing, where you buy credits and each task draws them down. Reported examples of a single Cowork use range from about a dollar to seven dollars or more depending on the task and model. Your AI bill is becoming a variable, and variables need watching.
From a known number to a moving one
The shift matters more than the increase. A price rise is annoying but plannable, you adjust the budget line and move on. Consumption pricing changes the shape of the cost. Instead of knowing your bill before the month starts, you find out at the end, and the total depends on how much your team used the tool, what kinds of tasks they ran, and which model those tasks touched, since heavier models cost more per use. Vendors will fairly point out you only pay for what you use. True. But it also means nobody is capping your exposure except you.
Who wins and who gets surprised
Consumption pricing is not a scam, it is a different distribution of risk. Whether it favours you depends almost entirely on a usage pattern most businesses have never actually measured.
| Consumption pricing works well | Consumption pricing bites |
|---|---|
| Occasional, bursty use across a team | A few power users running heavy tasks daily |
| You stop paying for idle seats | An automated workflow quietly loops |
| Someone watches the meter | Nobody looks until the invoice arrives |
That last row is where most bad surprises come from, and it is the cheapest one to fix. It also connects to something we wrote about yesterday: if you cannot list what AI you pay for, you certainly cannot spot a bill drifting upward. Start with an inventory of your AI stack.
Four habits that keep it sane
None of this requires a finance department. Turn on spending limits and alerts wherever the vendor offers them, and do it before you need them rather than after a shock. Give one person responsibility for reviewing AI spend monthly, the way someone already watches your other bills. Steer routine work to cheaper, lighter models and save premium models for tasks that genuinely benefit, because model choice is frequently the largest single driver of cost. And watch your first two or three months closely to learn your real pattern, then decide honestly whether consumption or a flat plan suits you better. Sometimes the boring flat rate wins.
The bottom line
AI is turning software from a predictable subscription into something closer to a utility bill, and utilities need meters, alerts, and someone paying attention. That is manageable, but only if you treat it as an operating cost rather than a fixed one. Check which of your tools have moved to consumption pricing, find out what a typical task really costs, switch on the alerts today, and build the rising base prices into your annual budget instead of discovering them at renewal. The businesses that get burned here will not be the heavy users. They will be the ones who never looked at the meter.
Frequently Asked Questions
What is changing about AI pricing?
Two things at once. First, base software prices are rising to help pay for AI, Microsoft 365 increases have now taken effect across business and enterprise plans, with frontline plans rising more steeply. Second, and more significant, some AI features are moving away from flat per-user pricing toward consumption models where you buy credits and each task draws them down. Microsoft’s Cowork, for example, went generally available on a consumption basis, with individual uses reported to range from around a dollar to seven dollars or more depending on the work and the model used.
Why does consumption pricing matter so much?
Because it changes your AI cost from a known number into a variable one. With per-user pricing you know your bill before the month starts. With consumption pricing, your bill depends on how much your team uses the tool, which tasks they run, and which model those tasks touch, and heavier models cost more. That is not inherently bad, you only pay for what you use, but it removes the predictability that makes budgeting easy and creates real potential for a surprise at month end if usage climbs faster than anyone expected.
Is consumption pricing a bad deal?
Not necessarily, it is just a different deal. It can be excellent if your usage is genuinely occasional, since you stop paying for seats that sit idle. It can be expensive if a few enthusiastic users run heavy tasks repeatedly, or if an automated workflow quietly loops. The honest answer is that it shifts risk from the vendor to you: they no longer eat the cost of heavy users, and you no longer enjoy the certainty of a flat fee. Whether that is good depends entirely on your actual usage pattern, which most businesses have never measured.
How do I keep a variable AI bill under control?
Four practical habits. Set spending limits or alerts wherever the vendor offers them, before you need them. Make someone responsible for watching AI spend monthly, the same way someone watches your other bills. Use cheaper, lighter models for routine work and reserve premium models for tasks that genuinely need them, since model choice is often the biggest driver of cost. And review your first two or three months closely to learn your real usage pattern, then decide whether consumption or a flat plan actually suits you. Most surprises come from nobody watching.
What should a Canadian business do right now?
Check whether any AI feature you rely on has moved, or is moving, to consumption pricing, and find out what a typical task actually costs. Turn on budget alerts today rather than after a shock. Estimate your realistic monthly usage and compare it honestly against flat-rate alternatives, sometimes the flat plan is cheaper, sometimes it is not. And factor rising base software prices into your annual budget rather than treating them as a surprise. AI costs are becoming an operating expense that needs the same attention as any other variable cost in your business.
Keep your AI bill predictable
We help Canadian businesses match AI plans to real usage, set sensible guardrails on spend, and use the right model for each job so costs stay under control.
Related Articles
Voice AI Just Got Smart and Cheap: What Reasoning Voice Agents Mean for Customer Calls
Local AI Just Went Mainstream: Should You Run It Yourself?
NotebookLM Pricing Canada 2026: Free vs Plus ($27 CAD/mo) and Whether You Need It
AI consultants with 100+ custom GPT builds and automation projects for 50+ Canadian businesses across 20+ industries. Based in Markham, Ontario. PIPEDA-compliant solutions.