Stripe just fixed the problem killing AI startups: token billing that lets you charge a 30% markup on LLM costs instead of eating them. The feature launched March 2, 2026, in preview. If you’re building agentic AI systems that burn through Claude or GPT tokens, you can now pass those costs to users automatically—and add a profit margin on top.
This isn’t theoretical. It’s the difference between scaling profitably and bleeding cash every time a user clicks “generate.”
Stripe’s 30% markup turns AI costs into profit—if you can get access
Here’s how it works: Stripe tracks your OpenAI, Anthropic, or other LLM API usage in real time, calculates the token cost per user, and bills them accordingly. You set the markup—say, 30% above raw compute costs—and Stripe handles the rest. No manual invoicing, no spreadsheet hell, no guessing whether your AI feature will bankrupt you next month.
The model works at scale. Gamma, the AI presentation tool, is the proof point everyone’s citing—though Stripe’s case study doesn’t include the $100M ARR figure floating around industry coverage. What we know for certain: Gamma uses Stripe’s infrastructure to handle payments for its 70 million users, and CEO Grant Lee says the platform “gives us the financial infrastructure to support Gamma’s expansion.” That’s the pitch: turn your biggest cost center into a line item customers accept because the AI feature delivers value.
The catch: it’s waitlist-only. Most startups can’t access this yet, which means competitors with early access are scaling profitably while everyone else keeps subsidizing their users’ AI habits.
The data says AI features make money, not lose it
Stripe’s betting on a counterintuitive truth: AI features aren’t the problem. Billing infrastructure is. According to Stripe’s monetization research, properly calibrated AI features boost customer satisfaction 15%-20% and revenue 5%-8%. That contradicts the narrative that AI drains margins—but only if you can charge for it.
And Stripe’s been solving payment problems at scale for years. The company recovered $6.5 billion in revenue through automated retries and payment optimizations. That credibility matters when you’re asking startups to trust you with their entire monetization strategy. Gamma’s reliance on Link wallet—Stripe’s one-click checkout product—for over 40% of transactions shows the ecosystem advantage: you’re not just getting token billing, you’re getting the full payment stack.
But here’s the thing. If companies underestimate AI’s impact on revenue, they’ll keep treating these features as cost sinks instead of growth drivers. Stripe’s data suggests the opposite: AI increases willingness to pay when you bill transparently.
The uncapped bill problem nobody’s solved yet
Token billing doesn’t eliminate risk. It shifts it. One misconfigured agent can still generate thousands in compute costs overnight, and if your metering fails, you’re the one eating the bill—or passing an unexplained charge to a customer who churns immediately.
Stripe isn’t alone here. OpenRouter charges a 5.5% markup on token fees for its first-tier plan—lower than Stripe’s example, but without the distribution. Stripe’s advantage is reach: if you’re already processing payments through them, token billing is one API call away. If you’re not, you’re waiting on a list while competitors scale.
A Stripe product manager acknowledged third-party gateways like Vercel and OpenRouter on social media, signaling awareness that this isn’t a winner-take-all market. But awareness doesn’t solve the access problem. The waitlist means most startups are still locked out, watching a two-tier economy form: those with Stripe token billing can scale profitably, those without keep bleeding.
The feature works. Gamma proved it. But if you can’t get in, the proof doesn’t matter.
