← All articles

Craft

Stripe didn't buy a billing company. It bought the line between product and business.

Credits look like a finance detail. Each one carries a dozen product decisions — and in 2025, the companies that got them wrong found out in public.

On January 14, 2026, Stripe completed its acquisition of Metronome, a usage-billing company that, in Stripe CEO Patrick Collison's words, "already powers some of the most ambitious AI companies, including OpenAI, Anthropic, and NVIDIA." Stripe didn't disclose a price. Upstarts Media, citing eight sources, reported about $1 billion.

Collison's explanation of the deal is the line worth keeping: "Metering and billing are in a very direct sense the interface between 'product' and 'business'."

Most PMs treat billing as someone else's system. Then look at what happened in 2025.

Every credit crisis last year was a product decision

  • Cursor — on June 16, 2025, Cursor changed its Pro plan from 500 requests a month to $20 of usage at API prices, plus unlimited use in Auto mode. On July 4, CEO Michael Truell wrote: "We recognize that we didn't handle this pricing rollout well, and we're sorry." Cursor offered refunds, on request, for unexpected usage between June 16 and July 4.
  • Anthropic — on July 28, 2025, Anthropic announced new weekly rate limits for its Claude Pro and Max plans, starting in late August and aimed at heavy Claude Code use. The reason: "Some of the biggest Claude Code fans are running it continuously in the background, 24/7." One user "consumed tens of thousands in model usage on a $200 plan." The limits were expected to affect "less than 5% of subscribers."
  • GitHub Copilot — GitHub said premium requests would go live on May 5, 2025. On May 7, GitHub delayed enforcement. On June 18, the limits went live — with a default spending limit of $0 for extra requests.
  • Replit — effort-based pricing rolled out to new users on June 18, 2025, and to everyone else on July 2. Replit itself acknowledged it "can end up being more expensive over the lifetime of a project." A billing error on July 11 hit about 6% of paying users, who were reimbursed.

Only one of these is a bug. The rest are decisions: what a unit is, what happens at the limit, who sees the meter, and when the gate closes. That's product work — done in public.

A credit has settings

Open the documentation of any modern billing platform and you'll see it. A credit isn't a number in a database. It's an object with properties, and every property is a decision.

  • Stripe — credit grants carry a priority, where a "lower number indicates a higher priority" and applies first. "Credits won't expire unless you set this field." A grant can be limited to specific metered prices.
  • Metronome — you can "specify what percent of the initial balance of the credit should roll over to the new contract."
  • Orb — credits "expiring soonest are used first," and free credit blocks are spent before paid ones.

Here's the product question behind each setting:

SettingThe product question behind it
ExpiryHow long before an unused credit disappears — and how will that feel to the customer?
Burn orderPromotional credits first, or paid? Oldest first?
ScopeCan a credit be spent anywhere, or only on one feature or model?
RolloverDoes an unused balance carry into the next period or contract?
PoolingOne company-wide balance, or a limit per user?
DenominationAbstract units, or dollars?
Behavior at zeroBlock, downgrade to a cheaper model, or refill automatically?
VisibilityDoes the user see a raw number, or the work they have left?

Ship without deciding, and the platform's defaults decide for you. Stripe's default, for example, is that credits never expire.

A credit isn't a number. It's a dozen product decisions — and if you don't make them, your billing vendor's defaults will.

Three jobs, and one of them can't wait

Under the settings, every credit system does three jobs.

  • Record — a ledger of every grant, spend, and expiration. Never overwrite a balance; add entries and calculate the balance from history. When a customer disputes a charge, that history is your only defense.
  • Count — metering: what was used, by whom, on what.
  • Gate — enforcement: does this user have the balance to proceed, right now?

Recording and counting can run a little late. An invoice doesn't care whether usage arrived a few minutes behind. Gating can't. Picture five agents on one account, each checking a balance of 500 credits at the same instant. All five see 500. All five proceed. If that check isn't a single all-or-nothing operation, the account is deep below zero before anyone notices.

Anthropic's Claude Code limits show the gate as a product decision too. A $200 plan with no weekly ceiling met users running it around the clock — one of them consumed tens of thousands in model usage. The fix wasn't a faster database. It was a new rule about what the plan allows.

Who owns this in your company?

Collison's line is the reason to care. If metering is the interface between product and business, it can't belong only to finance, or only to engineering. Someone has to decide what a credit means to the customer — and that's a product job.

Your next pricing incident probably won't be the price. It'll be a setting nobody chose.

Could you name, today, who decided your credits' expiry, their burn order, and what happens at zero?

Free companion

The Credit Engine — Infrastructure Playbook

The technical playbook behind credit-based pricing in a single PDF — the three pillars (ledger, metering, enforcement), why balances must never be overwritten, credit leasing for high-frequency agents, the full anatomy of a credit, lifecycle rules, and a buy-vs-build vendor comparison.

Get the playbook →

This is one piece of a longer framework I teach in Chapter 5 of Product Strategy in the AI Era — including the twelve decisions behind a credit system: units vs. currency, expiry, burn order, bundle sizes, renewal defaults, what happens on churn, and revenue recognition.

Sources

  • Stripe Newsroom, "Stripe completes Metronome acquisition" (January 14, 2026): stripe.com
  • Upstarts Media, Stripe pays $1 billion for Metronome (December 3, 2025): upstartsmedia.com
  • Cursor, "Clarifying our pricing" (July 4, 2025): cursor.com
  • Simon Willison, quoting Anthropic's Claude Code weekly limits (July 28, 2025): simonwillison.net
  • GitHub Changelog, Copilot Pro+ (April 4, 2025): github.blog
  • GitHub Changelog, premium request limits delayed (May 7, 2025): github.blog
  • GitHub Changelog, premium request billing in effect (June 18, 2025): github.blog
  • Replit, "Effort-based pricing recap": replit.com
  • Stripe docs, billing credits: docs.stripe.com
  • Metronome docs, applying credits and commits: metronome.com
  • Orb docs, prepurchase and credits: withorb.com

Newsletter

The Strategic AI Corner

One idea on product strategy in the AI era — weekly. No spam, unsubscribe anytime.

Free. Unsubscribe anytime.

More articles

📊
Strategy

Figma gave customers three months to learn its AI credits. One still ended up at 209% of the limit.

Measuring usage before you charge for it is the right move. Figma did it by the book — and its forum still shows exactly where warm-up periods break.

Read →
📉
Strategy

Your engineers just made your biggest pricing decision

Two AI products can charge the same price and keep very different shares of it. The difference gets decided in the architecture — and nobody calls it a pricing decision.

Read →
📋
Strategy

Your customer's procurement team now has a checklist for your AI pricing

Buyers have spent two years absorbing AI price increases. Now they're organizing. Here's what they'll ask — and how to design pricing that passes before they do.

Read →
🤝
Strategy

Your usage pricing won't fail with customers. It'll fail in the comp plan.

With seats, the deal ends at signature. With usage pricing, it starts there — and most sales teams are still paid for the old ending.

Read →
🧭
Strategy

Anthropic pricing and model — a lesson for all respected SaaS

Anthropic's top plan unlocks four things, and not one of them is a feature. A masterclass in AI-era segmentation — and what it means for your own pricing.

Read →
📝
Craft

Your PRD now has three readers — and one of them is an AI

PRDs didn't die. They got a new audience — and old habits don't survive it.

Read →
💰
Strategy

Pricing is the feature you forgot to ship

Why pricing decisions belong in the product roadmap, not in finance.

Read →
🎯
Positioning

Every AI feature you add is a positioning decision you didn't make

The hidden cost of adding AI features without a positioning frame.

Read →