AI token economics: what the OpenRouter numbers mean for your budget
OpenRouter’s usage data tells a story most AI cost breakdowns skip. Open-weight models handle roughly 60% of weekly tokens on the platform but account for only 26% of spending. The gap between volume and dollars is wide, and it matters when you’re sizing a budget.
The number on the pricing page is not your real cost
Per-token list prices don’t reflect what a completed task actually costs you. Two factors change the bill significantly: cache discounts reduce what you pay on repeated context, and failed requests still cost money even when they produce nothing useful.
The article from OnlyCFO puts it plainly. If each request costs $1 and your success rate is 80%, every successful result costs $1.25 in practice. That 25-cent gap compounds fast at volume.
How to budget this correctly
The right unit is cost per successful task, not cost per token. To get there, apply your actual cache discount rate and your real retry rate to the list price before you compare providers.
- Pull your cache hit rate from your provider’s dashboard
- Track failed and retried requests separately from successful ones
- Divide total spend by successful completions to get your true unit cost
- Run that calculation for each provider you’re evaluating before switching
The compute access problem for startups
A second piece from Evan Conrad adds a harder constraint to the picture. The article argues that most startups will not be able to secure compute capacity within the next six months. Getting GPU capacity online is increasingly difficult for companies that are not investment grade.
Startups are being asked for larger upfront payments. Investment-grade GPU buyers are reportedly beginning to require that token purchasers either carry investment-grade credit themselves or put up larger deposits. The few investment-grade tenants taking on more debt are themselves being downgraded as a result.
For solo operators and small teams, the practical read is this: open-weight models running on third-party infrastructure are currently the most accessible path to predictable AI costs. The economics favor them on price, and the access constraints favor them on availability.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
