Why your AI project can go viral and still lose money

For about twenty years, software obeyed one rule. You build it once, and serving the next user costs almost nothing. So more users meant more profit, every time. Growth was the whole game.

AI broke that rule. Every action in your product now calls a model, and that call costs real money. Here's what nobody warns you about: an AI project can go viral and lose money faster the more it grows.

The one number that matters

Contribution margin per user. It's what's left from a customer after the cost of serving them, and for AI products that cost now includes tokens. Negative margin means every new user digs the hole deeper. Everything else is downstream of this one figure.

The three tests

The calculator runs three, and your idea has to pass all three.

1. Does each user pay for themselves?

   Positive contribution margin, and a payback period you can actually survive. This is where token cost quietly eats you.

2. Does the math clear a real number?

   Your market size, times your price, times the slice you can realistically win. A $10 tool needs a million buyers. A $10,000 tool needs a thousand. Those are two completely different companies, and almost nobody does the multiplication before they build.

3. Can you afford to get the customer?

   Lifetime value against what it costs to acquire them. The one that gets people is the danger zone: priced too low for anyone to sell it for you, too high to spread on its own.

The move that fixes most of it

If you change one thing, change your pricing. Flat monthly plans are usually wrong for AI, because your cost to serve moves with usage while your revenue sits still. The power user you love is the one draining you. A base fee plus a usage or outcome component keeps revenue tracking cost, and that's where most of the pricing that survives has ended up.