I clicked because a 54x price-to-sales multiple on a $140M revenue business is wild. Three months ago OpenRouter was valued at $1.3B; Stripe just paid 5.7x that. The top Hacker News comment nailed the skepticism: why is an API middleman worth more than Lyft or Alaska Airlines?
Don't read this as Stripe losing its mind. OpenRouter sits on 10M+ developers and a unified billing entry point across 400+ models. That distribution position is genuinely scarce. If AWS or Google bought it, neutrality evaporates overnight—model providers would flee a channel controlled by a competitor. Stripe doesn't build foundation models, so the stance at least holds.
The sharpest part is data convergence. Before the deal, Stripe only saw the revenue side of AI customers—when subscriptions hit, refund rates, retention curves. OpenRouter only saw the cost side—which agent is hammering which model, token consumption, latency. Once merchant IDs link the two systems, Stripe can see both sides of an AI company's P&L simultaneously.
Real scenario: two code-generation startups show identical monthly revenue on Stripe. One has flat token costs; the other's token spend climbs 30% month-over-month. Old Stripe couldn't tell them apart. Now it can spot the second company's margin erosion immediately. Token costs hit during inference, but subscription revenue lags by 30–60 days—the second company hits a cash crunch sooner. Stripe Capital can extend credit before traditional lenders even see the problem.
Where I'd discount: the heavy users on OpenRouter are mostly small teams and long-tail apps. Large AI companies negotiate private volume deals directly with model labs and bypass public routing. So this credit model serves smaller, higher-risk borrowers—not the whole AI lending market.
What the piece doesn't finish: the feedback loop. Stripe can currently do cost-based smart routing—pick the cheapest model that meets requirements. What companies actually want is routing by business return: which model drives higher conversion and customer lifetime value. That requires feeding payment outcomes back into routing decisions, and the engineering isn't there yet.