AI companies must watch out for freeloaders burning tokens for free—and wrecking margins one sign-up at a time

Software used to love scale. Build it once, sell it again and again, and the cost of serving one more customer was close to nothing. AI has broken that tidy little assumption. Every prompt, every answer, every agent task has to be generated fresh, and that means paid for fresh too.

That shift changes the whole business model. The internet economy was built around people making choices: search, click, subscribe, check out. AI agents don’t behave like that. They can research, write code, call APIs, execute tasks and even pay on a user’s behalf. Useful, yes. Also expensive in ways that can pile up fast.

The real danger is not just high compute bills. It’s abuse that slips in under the radar. Free trials and multi-account sign-ups can drain tokens without any intention to pay, and the damage shows up in margins long after the first request. Stripe research found that more than one in six sign-ups at AI companies are tied to multi-account abuse. By the time a payment fails, the user may already be gone.

That is pushing AI companies toward usage-based or hybrid pricing, where tokens become the unit that matters. Lovable, the AI software creation platform that uses Stripe, started with subscriptions and then moved to usage-based billing. Customers go through their free allowance, then get charged for token consumption. It is cleaner on paper. It is also a nicer target for freeloaders.

The message from Stripe’s Patrick Collison is blunt enough: software in the AI era is more like pizza, made to order at the moment of use. Fine. But pizza has to be paid for before the box gets thrown away. Companies now need to spot abuse at sign-up, not after a payment request bounces, and they need to distinguish a thief gaming free tokens from a legitimate customer whose usage suddenly spikes.