3 startups solving the thing nobody warns you about: pricing AI you can't predict
When your cost-of-goods is a token meter that spikes with usage, the old SaaS pricing playbook breaks. Here's who's rewriting it.
If you ship an AI product, you've felt this: a single power user can cost you more than they pay. Flat monthly pricing was designed for software with near-zero marginal cost. Tokens are not that. Three startups are building for the world where your COGS is a live meter.
1. Meterline — usage billing built for token economics
What they built: metering and billing infrastructure that speaks "tokens, requests, and model tiers" natively instead of forcing you to bolt AI usage onto Stripe metered billing.
The builder's take: I've hacked this together with Stripe usage records and a cron job, and it's genuinely painful the moment you have more than one model or want to pass margin through cleanly. The wedge is real. The question is whether metering is a feature Stripe eventually absorbs — building infrastructure one layer below a giant is always a race.
2. Grayline — margin alerts before the invoice, not after
What they built: watches your per-customer unit economics in real time and tells you which accounts are underwater while you can still act.
The builder's take: this is the one I'd actually pay for today. The failure mode of AI products isn't slow growth — it's growing yourself into negative margin without noticing. Making unit economics observable per customer is the right altitude. What I'd want to know: does it plug into how I already tag usage, or does it demand I re-instrument everything? Adoption lives and dies on that.
3. Prepaid.dev — credits as a first-class primitive
What they built: drop-in prepaid credit balances, top-ups, and expiry for AI apps — the "buy 10,000 credits" flow, productized.
The builder's take: credits quietly solve two problems at once — they cap your downside and they get you paid before you incur the cost. That's a beautiful cash-flow property most founders underrate. The risk is UX: credits done badly feel like a arcade token scam. Done well (clear value, no dark-pattern expiry) they build trust.
The through-line: pricing is becoming a core engineering concern, not a spreadsheet you fill in once. If your product touches a model, the team that instruments margin early is the team that survives its own success.
Get the next issue first
The 3 startups worth knowing this week, with a builder's take on why each one matters. One email, every Tuesday.