Spend management service provider Ramp today launched its AI token spend management tool, giving finance teams visibility and control over spending across major AI providers. 

AI token spend across Ramp’s customers has climbed 20.7x since June 2025, which illustrates a growing paradox: The cheaper tokens get, the more companies spend, according to Ramp’s release today.

“LLM providers are aggressively undercutting each other, with raw token costs dropping quickly, but it hasn’t triggered a race to the bottom in total vendor revenue,” Calvin Lee, senior director of product management at Ramp, told FinAi News. 

The reason, Lee argues, is behavioral. 

As unit costs fall, enterprises don’t pocket the savings — they use more. 

“Because firms are moving from simple chatbots to continuous, high-volume agentic pipelines that run thousands of background loops a day, their consumption is outstripping the price drops,” he said. “The underlying model is becoming a commodity, but the enterprise appetite for compute is effectively infinite.” 

Managing token cost 

As the demand for AI increases, companies are trying new ways to tame costs while deploying more agentic AI tools, Lee said. 

Ramp’s token spend management tool allows customers to use dashboards to understand how much money has been spent on tokens, cap spending limits for employees and find inefficiencies in AI spend, he said. 

Happen Bank, for example, has deployed a token dashboard to keep tabs on how employees are using tokens and step in when there is waste, the bank previously told FinAi News. 

Venture capital firm AngelList made minor tweaks to its AI agents running continuously, to avoid what could have resulted in a $10,000 token bill for the billing cycle, Lee said, but with Ramp’s spend management tool, “we caught and surfaced that optimization the same day using our detection engines, which we built entirely in-house.” 

Ramp found similar savings while managing its own AI spend, Lee said.  

“Many everyday tasks, such as summarization, classification, reporting and routine automation, don’t necessarily require the intelligence needed to solve the hardest math or scientific problems,” he said. “The goal is to match the model to the job.” 

Ramp’s ML tool also monitors which models are being used for what function, Lee said, adding that’s what companies are paying attention to. 

JPMorgan Chief Financial Officer Jeremy Barnum echoed that sentiment during the bank’s second-quarter earnings call this week. 

“You really don’t need the latest cutting-edge, incredibly expensive model to summarize an analyst report,” Barnum said. “So, the idea is use the right model for the right purpose.” 

Ramp clients include: 

• Stripe; 

• Visa; 

• Defense contractor Anduril Industries; and 

• Credit card issuer BILT. 

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.