Same Product, A Different Bill
AI is getting cheaper by the unit and more expensive by the invoice.
Start with Microsoft changing how Copilot is sold through its Cloud Solution Provider program, the channel MSPs use to resell Microsoft. As Channel Insider reports, beginning November second, new Microsoft 365 Copilot Business licenses bought through CSP include usage-based billing by default. That applies to standalone licenses and to bundles. Each new license comes with a preset pay-as-you-go limit of ten dollars per user per month, and the customer can adjust it or add prepaid credits. For a twenty-five-person client, that’s up to two hundred fifty dollars a month that didn’t exist on the old invoice. The usage that counts against it includes Copilot Cowork, the Work IQ APIs and GitHub Copilot Harness.
Microsoft is direct about the purpose. It says the default is meant to reduce setup friction and open a path toward more consumption and more upsell conversations. And since September first, it has run an activation incentive that rewards CSP partners whose customers activate and consume more.
So the seat is still sold at a fixed price. What a customer actually pays now depends on what that customer does with it.
Now look at the same thing from the buyer’s side. McKinsey studied American government entities, a large and fast-growing buyer of enterprise AI, as Semafor reports. Its finding: the per-token price of AI is collapsing, but the total bill is rising, because agents need far more compute. Tim Ward, who leads McKinsey’s global public sector practice, told Semafor that for many governments, using AI may have been close to free until now. Very soon, he said, “it won’t be.”
Put the two side by side. The vendor publishes a falling price per unit. The buyer receives a rising bill. And the gap between those two numbers is decided by consumption, one customer at a time.
Two customers buying the identical product can now pay very different amounts. The price list no longer tells either of them which one they’ll be.
So who decides which one you’ll be?
The Seller Holds The Record
Every price is a guess about what the buyer will pay. AI makes that guess cheap, and it puts the guess in the seller’s hands.
Look at how it works when it’s pointed at consumers. McDonald’s uses an AI pricing engine to recommend what each of its nearly fourteen thousand restaurants should charge for every item, as Reuters found and TNW reports. The engine analyzes millions of daily transactions. One factor it weighs is an estimate of how much each store’s customers are willing to pay. It also pulls prices from the online menus of nearby Wendy’s and Burger King restaurants. McDonald’s calls it a tool, not a mandate. But Reuters found a June franchisee document showing the company records in detail when owners deviate from what it recommends. McDonald’s disputes the reporting.
The reaction to that kind of pricing in consumer markets has been sharp. Walmart’s chief executive, John Furner, wrote to customers ruling it out: “We price the product, not the person.” The Federal Trade Commission has published a draft policy statement on it.
In business, there is no letter like that. Different customers paying different prices is called negotiation, and it’s normal. What AI changes is how much each side knows when it sits down.
The fight between hospitals and health insurers shows the business version. The Blue Cross Blue Shield Association says hospitals’ use of AI tools to code insurance claims added nine hundred forty-two million dollars in spending over two years. That’s the insurers’ own analysis, so weigh it that way. They argue the coding changed while the care didn’t. The New York Times reports that AI is now on both sides of that fight. One side got better at reading the record, the number moved, and the other side had to build its own analysis to argue back.
Now put a buyer on the usage-priced side of that table. McKinsey says the cost of an AI agent completing the same task can vary by as much as thirty times, as Business Insider reports. Picture a task that costs a dollar one time and thirty dollars the next, with nothing on the invoice to tell you why. The seller meters every run. The buyer usually can’t predict its own bill.
In plain terms, in a usage-priced market, the advantage goes to whoever holds the better record of what the buyer actually consumes. Right now, that’s the vendor.
And for most of your clients, it isn’t close.
Which raises the question of who’s in a position to close the gap.
Tracking What Your Clients Use
For an MSP, that gap shows up in two places at once. You buy from these vendors, and your clients buy through you. Either way, somebody sits down to talk price, and the question is who brought the numbers.
Start with how few buyers have them. KPMG’s Global AI Pulse surveyed more than two thousand senior leaders across twenty countries, all at organizations with more than fifty million dollars in annual revenue. KPMG sells advisory work on exactly this problem, so keep that in mind. Thirty-five percent said their AI operating costs are fully visible and actively monitored. Thirteen percent said they only see the cost once the bill arrives. And the ones with full visibility were five times more likely to report an established return on their AI investment: fifteen percent against three.
Those are companies big enough to have finance departments. Your clients sit below that line.
The large buyers have noticed, and they’re buying the record. AI procurement firm Vertice acquired software-pricing company Vendr, as TNW reports. Vendr’s chief executive, Ryan Neu, framed the deal around the company’s founding premise: buyers signing big contracts had far less information than the vendors across the table. According to Vertice, the combined data covers more than seventy-five billion dollars in spend across thirty-two thousand vendors, drawn from two hundred fifty thousand negotiated contracts. It feeds an AI agent that negotiates renewals on the buyer’s behalf.
So the enterprise is closing the gap by buying data. A twenty-five-person accounting firm isn’t buying Vertice. The only party in its world positioned to keep track of what it actually uses is the one that already administers its tenants. That’s you. And notice who’s paying you to drive that consumption up.
So here’s the choice. Track each client’s consumption yourself, on a schedule, and keep the history, so you walk into every renewal and every pricing conversation already knowing what that client uses. Or find out at renewal, when the vendor shows you its numbers, and pay what its data says your clients will bear.
The first test of that choice won’t be a renewal. It’ll be your client’s first metered bill.
Because your clients are about to get an AI bill that changes month to month, and the first thing they’ll ask is whether it’s right. Before November second, sit down with every client on Copilot, explain the ten-dollar-per-user default that comes with new licenses, and ask who they expect to use it most and for what work. That conversation starts the consumption history. At renewal, the same history becomes the basis for pricing your own AI line per client, with the record behind the number on the table.
What to Consider
- Export before the window closes. In the Microsoft 365 admin center, the Copilot usage report can be viewed over the last 7, 30, 90 or 180 days. That’s the longest it looks back. A year of a client’s consumption at renewal exists only if you exported it and kept it, monthly and per tenant.
- Map where each meter lives. Even inside Microsoft, AI spend reports in more than one place:
- The Copilot Cost Management dashboard covers Copilot Credit spending.
- A Copilot Credits usage report under Reports > Usage shows metered Copilot Chat consumption.
- Agents built in Microsoft Foundry are billed as ordinary Azure consumption, not through the Copilot billing policy at all.
For every AI tool you sell, write down where its usage shows up per client. Otherwise your record will have gaps exactly where the bill grows.
- Set the ceiling with the client, not for them. In Cost Management, a spending policy can be set to limit monthly spending, which caps the credits it can spend each month. Agree that number with the client in the same meeting where you explain the November change. A ceiling the client chose is one they can check the bill against, and your consumption history is what tells both of you whether it was right.
If this trend continues: Within a year, as more Copilot features move onto credits, the renewal conversations that matter will be about the metered line rather than the seat count. The MSPs negotiating from their own numbers will be the ones who started keeping them before November second