Ramp makes AI spend visible

A company can tell you exactly who expensed a $38 airport meal, but until now it often could not tell you which team burned through 40 million artificial intelligence tokens on a model over a weekend. Ramp says it now pulls token-level usage data from providers like OpenAI and Anthropic into the same place finance teams already watch cards, bills, and software spend. (ramp.com) That is a shift in what counts as “spend.” Ramp’s new product page says customers can track artificial intelligence costs by provider, model, team, and project, which turns model usage into something closer to a cloud bill than a vague research budget. (ramp.com) Ramp is not a niche analytics startup trying to wedge into finance. It is the New York company best known for corporate cards and expense management, and Fast Company reported in March 2026 that more than 50,000 businesses use its platform. (fastcompany.com) The reason this product exists is that artificial intelligence is billed in tiny units that do not feel like money when engineers are using them. A token is a chunk of text a model reads or writes, so a long prompt, a long answer, and repeated automated calls can quietly stack into a real invoice. (cnbc.com) That billing model breaks the habits finance teams built for software subscriptions. Ramp’s help center says its tool is meant to replace the work of reconciling invoices across separate provider dashboards and instead show costs and trends in one view. (ramp.com) The spending is no longer small enough to ignore. Ramp says median monthly artificial intelligence spend in its transaction data grew 4 times from February 2025 to February 2026, and its research page cites surveys showing 78 percent of information technology leaders saw unexpected artificial intelligence charges. (ramp.com) That helps explain why finance software is starting to treat model usage like cloud infrastructure. Deloitte wrote this week that enterprises are mixing software subscriptions, application programming interfaces, and self-hosted systems for artificial intelligence, which gives one company several different cost curves to monitor at once. (deloitte.com) Ramp’s launch is narrow on purpose. Its public materials name Anthropic and OpenAI first, and The New Stack reported that the company is pulling token-level usage directly from providers so finance teams can see where spending is generated and who is consuming it. (thenewstack.io) Once that data sits next to budgets and approvals, the conversation changes from “Are we experimenting with artificial intelligence?” to “Why did one workflow switch from a cheaper model to a pricier one on Tuesday?” That is the kind of question procurement teams already ask about cloud servers and software licenses, and Ramp is betting they will now ask it about tokens too. (ramp.com)