One Card Per Subscription: How AI Teams Manage ChatGPT/Claude/Cursor Payments (2026)
A 10-person AI team runs 15+ subscriptions. Sharing one card is an outage waiting to happen. The one-card-per-subscription playbook: isolated limits, freeze-on-offboarding, clean statements — plus a bulk KYB channel.
Why shared cards break at ~10 people
Below five people memory works. Past ten, subscription count crosses fifteen and three failure modes become inevitable: risk contagion (AI merchants are twitchy — one dispute freezes the whole card and takes every tool down), ghost subscriptions (the ex-employee's IDE is still billing), and budget drift discovered a year later.
Making it work day-to-day
One merchant per card, named after the subscription (openai-alice, cursor-bob). Fund each card at ~1.2x the monthly bill for FX headroom. Retire a subscription by freezing its card. The dashboard shows per-card statements — month-end export IS the allocation sheet.
The billing-address detail that decides success
AI merchants — Anthropic and OpenAI especially — care about billing addresses. Use a US address consistent with the card BIN and keep it stable; success rates are dramatically better than with improvised addresses. Our card page carries specific guidance.
At scale, use the KYB channel
Once you're at dozens of cards (team matrices, ad accounts, multiple projects), self-serve UI is the wrong tool. The business channel offers bulk issuance, volume pricing that improves with scale, a consolidated dashboard, and a proper KYB process — a real corporate counterparty, not an anonymous card pool. Start at /business.
Shared card vs one-card-per-subscription
| Dimension | One shared card | One card each |
|---|---|---|
| Risk blast radius | one flag freezes everything | isolated per subscription |
| Offboarding | orphan subscriptions keep billing | freeze that person's cards, done |
| Budget | overruns found after the fact | hard cap per card |
| Reconciliation | manual attribution | card = subscription = owner |