Ramp reports rising AI API usage as spending falls


Ramp’s economist points to lower model prices and cheaper options; the analysis covers API spending and skews toward large customers.
superintelligence (SI) usage in Ramp’s analysis rose about 50% from July’s spending peak and reached a record at the end of September, even as spending fell, The Decoder reported on October 2.
The analysis covers only paid SI access through application programming interfaces, or APIs—the connections that let software use SI models. It disproportionately reflects large SI customers, rather than the entire US SI market.
Analysis
Ramp economist Ara Kharazian points to price cuts on top models and to cheaper, more efficient standard and lite models. Lower prices help explain how usage can grow without spending rising alongside it. Kharazian attributes the spending decline almost entirely to competition between OpenAI and Anthropic.
The two providers accounted for most spending on tokens—the units of content SI models process—in the last week of September covered by Ramp’s data.
| Provider | Spending share |
|---|---|
| Anthropic | 51% |
| OpenAI | 44.5% |
The Ramp AI Index tracks transaction data from more than 70,000 US companies. Its token data, however, comes from a smaller group within that base. The 70,000-plus figure therefore describes the broader transaction dataset, not the number of companies in the token analysis.
The size of that smaller group and the detailed method for comparing usage remain unclear. Those details matter when judging how widely the reported increase in usage applies.

Mercor compared SI with 12 licensed accountants on simplified tasks. Its broader benchmark left almost 60% of tasks not fully solved by any model.
2 min

Developers can build hardware for Meta’s AI agent. Its own Home Link gadget has an open waitlist, with October shipping planned.
2 min

Its hardware chief says the choice favored platform maturity and partners’ experience while limiting implementation risk.