Key Takeaways

  • Ramp passed $2 billion in annualized revenue over the summer, growing more than 100% year over year alongside a $1.85 billion fundraise.
  • Usage-based token pricing turned software from a predictable fixed operating expense into a variable cost with direct marginal production costs.
  • Corporate capital allocation now spans three buckets instead of two: payroll for employees, contracts for vendors, and agentic compute for rented intelligence.
  • Ramp is deploying tools called Router and Glass to help CFOs track token consumption against direct revenue outcomes.

Software Carries a Marginal Cost Again

For thirty years, enterprise software ran on a simple economic model. A vendor built code once, copied it infinitely for zero marginal cost, and sold it to companies via annual per-seat licenses. Finance teams loved the predictability. They budgeted for headcount, multiplied it by seat prices, and locked in their software spend.

AI broke that arithmetic. As Ramp co-founder Eric Glyman explained after announcing the company passed $2 billion in annualized revenue: “The minute that it became such that if you used more tokens your bill went up it meant that software suddenly had marginal costs. In a way, software became money.”

When your product relies on API calls to model providers, customer usage directly drains cash. If an engineering team spins up autonomous agents to test internal code or scrape research, every workflow costs money in real time. Glyman notes that organizations historically managed two primary outflows: “You could hire people or you could hire vendors to go take care of stuff for you. Software and AI spend has become this third way to buy things.” Treating this third category like traditional software procurement blinds finance teams to ballooning bills that scale with bot activity rather than team headcount.

Autonomous Agents Require a New Ledger

The problem deepens when autonomous software begins buying services from other autonomous software. Glyman pointed out that an expanding share of corporate transactions is machine-directed: “Most spend is actually agentic in nature if you really look at how these things are growing, and yet you're renting intelligence.”

Companies already maintain disciplined systems for the first two spending buckets. Payroll tracks employee wages, while procurement platforms track supplier contracts. Yet when agents trigger compute chains across third-party models, finance departments rarely know which customer or business unit generated the expense.

Ramp is building products like Router and Glass specifically to address this governance gap. Instead of lumping model API bills into a general cloud infrastructure line item, these tools let finance teams trace every token query back to the commercial outcome it produced. As Glyman framed the mandate: “For every dollar that's going out, what is the return on intelligence? How does it go into your books and records? What revenue can you see that's directly connected?” If you run an engineering org where agents query external models all day, you cannot evaluate unit economics without tying raw token burn directly to collected gross profit.

What to Do With This

Audit your model consumption across your engineering and product stacks tomorrow morning. Export the last 30 days of API bills from OpenAI, Anthropic, or cloud model hosts, then map each API key to a specific customer contract, internal project, or paid feature tier. If more than 20% of your token spend sits in an untracked shared pool without direct revenue attribution, split those keys immediately by project before next month's billing cycle.