One of the quiet victories of the Subscription Economy was killing off cost-plus pricing. You stopped pricing to your cost structure and started pricing to the value the customer actually got (outcomes, usage, tiers), anything but “what did this cost me to make, plus a markup.” AI is dragging that logic backward, and lots of vendors don’t seem to have noticed.
Here’s how it happens. You ship an AI feature. It runs on tokens, and tokens cost real money. So you do the obvious thing: price the feature a little above what the tokens cost. AI gross margins are already sitting around 50-60%, well below the 80-90% software companies are used to, and the fastest way to protect that margin looks like tracking cost per request and pricing just above it. That’s cost-plus. It just arrived wearing a token counter instead of a factory ledger.
Here are three reasons cost plus pricing is a trap.
You’ve outsourced your pricing to your supplier.
When your price is a thin markup on tokens, every time a model provider cuts inference prices, your customer expects a discount. Every time they raise prices, or a new model needs more compute per query, that hike becomes your problem. You are no longer setting your price. Anthropic and OpenAI are setting it for you, and you’re just adding a margin sticker on top of whatever number they hand you this quarter.
It prices the wrong thing.
Cost-plus charges for effort: tokens burned, compute consumed. But nobody wakes up wanting to buy tokens. They want a resolved ticket, a drafted contract, a forecast that was right. The vendors actually winning this earnings season priced the outcome, not the exhaust. Zendesk charges $1.50 per automated resolution, not per token the resolution consumed. Fin charges $0.99 per resolved conversation. If your AI gets more efficient and starts answering the same question in half the tokens, cost-plus pricing means your revenue on that transaction falls. You get punished for exactly the efficiency gain you were supposed to be selling.
It caps your upside at your COGS.
Value-based pricing lets you capture a share of what the customer actually gained: hours saved, deals closed, tickets resolved. Cost-plus pricing puts a hard ceiling on your ambition. You can never charge more than a modest multiple of your input cost, no matter how much value showed up on the other side. Salesforce doesn’t bill based on the marginal costs of its “agentic work units” (notice the choice of words!). They’re billing based on what those units were worth to their customer.
The compute-cost conversation is real, and companies do need to protect margin. But margin protection is a floor, not a pricing strategy. Use cost as the guardrail that tells you when a deal stops making sense, but never as the number you build the price around.
SaaS spent two decades teaching software to stop thinking like manufacturing. Don’t let AI teach it to start again.
