Fintech company Ramp burned through $1.5 million on AI tokens in a single week, a figure that underscores the runaway costs companies face when large language models move from experiments to everyday tools. On Wednesday, Ramp launched Glass, an AI workspace built to monitor and optimize that spending.
The product is Ramp's attempt to apply its core competency—tracking and controlling corporate spend—to the fast-growing category of generative AI. Instead of just issuing cards and scanning receipts, Glass gives finance teams a view into how employees use AI, how many tokens those interactions consume, and where the bill is coming from.
Why $1.5 million in seven days is a warning
Ramp's own token bill is the clearest argument for the product. In a week, the company spent $1.5 million on tokens—the units that AI models process for input and output. That's not an annual budget or a pilot program. It's seven days of usage. For a company of Ramp's size, it's a burn rate that can't scale without guardrails.
The problem isn't unique to Ramp. As AI assistants get embedded in sales, support, coding, and internal search, token consumption grows quietly. A single employee using a chatbot for research might cost a few dollars a month. Thousands of employees doing the same, with agentic workflows that call models repeatedly, can push costs into the millions. Traditional expense software wasn't built for that kind of granular, usage-based billing.
Glass is designed to sit between the AI tools and the finance department. It tracks token usage across providers, attributes costs to teams or projects, and flags anomalies before they become line items on a quarterly report. The pitch is straightforward: you can't manage what you can't measure, and most companies currently have no idea what they're spending on AI until the invoice arrives.
What Glass actually does
Ramp hasn't published a full feature list, but the workspace is meant to be a central dashboard for AI operations. That includes monitoring consumption, setting spending limits, and optimizing which models get used for which tasks. A customer support bot might not need a frontier model to answer basic questions; Glass could route simpler queries to cheaper models, cutting costs without hurting the user experience.
The product also fits Ramp's broader strategy of inserting itself into every corporate transaction. Ramp started with corporate cards, moved into bill pay and accounting automation, and now wants to be the control layer for AI spending. If it works, Glass could become the default way finance teams audit their AI vendors—a position that gives Ramp leverage over both buyers and sellers.
Competitors are circling the same problem. Cloud providers offer cost management tools, and several startups are building AI observability platforms. But Ramp's advantage is its existing customer base and the fact that it already handles the money. Finance teams don't want another dashboard; they want fewer places to check. Glass promises to fold AI costs into the same system that tracks travel, software subscriptions, and office supplies.
The token economy gets a budget line
Token costs are becoming a board-level concern. Companies that embraced AI early are now discovering that usage-based pricing can spiral in ways that per-seat software never did. A single rogue process—an agent that loops, a developer who leaves a script running—can generate thousands of dollars in charges overnight. Ramp's $1.5 million week is an extreme example, but it points to a structural issue: AI is cheap to start and expensive to scale.
Glass is Ramp's bet that the next phase of enterprise AI won't be about access. It'll be about control. The company hasn't said how much Glass costs or when it will be generally available. It also hasn't disclosed whether the $1.5 million token spend was a one-time spike or a recurring run rate. Those details matter for customers trying to model their own budgets.
Ramp's launch comes as CFOs are under pressure to justify every software line item. AI tools that once sailed through procurement on the promise of productivity are now facing the same scrutiny as any other vendor. A workspace that can show exactly which teams are burning tokens—and how to cut that burn—is a timely sell. The question is whether Ramp can convince companies to let it police their AI usage before the next $1.5 million invoice lands.
