What a free tier costs when every user burns inference

For an AI product, the free tier is an acquisition budget spent automatically — every free request carries real inference cost. The question is never whether to spend it, but what the spend buys.

Traditional software free tiers cost almost nothing to scale. AI models burn real money with every request.

Run your own number

Take your cost per request, multiply by the requests an active free user makes in a month, multiply by your free-user count. That is your monthly acquisition spend: committed, recurring, and rising with every signup. Most teams have never written it down as a marketing line item. It is one.

Why AI broke the old freemium math

Freemium was rational when marginal cost per user rounded to zero. A million silent free users cost roughly nothing and occasionally converted. Inference removed the rounding: the free tier now has a metered bill, and "they might convert someday" has to cover it.

What the spend buys today

Under a classic free tier: silence. Free users arrive, consume, and tell no one, and the funnel only widens if you pay again elsewhere to drive traffic into it. The full accounting is worse than the average, because the heaviest free usage skews toward the users who will never convert.

Making the same spend buy exposure

An earned tier does not cut the budget. It names the purchase. The serving cost is the cost you were already paying; what changes is that every grant now has a public, disclosed post attached, seen by the poster's own network. The spend stops buying silence.

It also re-prices the abuse. Earning requires one social account, publicly attached to every grant, which is a cost the farmer pays and a real user does not.