The AI Spending Surge: Why Corporate Finance Needs to Treat Tokens Like Infrastructure
Artificial intelligence has officially moved beyond the experimental phase and into the core of corporate operations. However, this transition has brought a major financial headache: AI token spending exploded by 572% year-over-year between June 2025 and June 2026, according to new data from Ramp.
For many companies, the realization is setting in that AI is not a “set-it-and-forget-it” line item. Instead, it functions more like cloud infrastructure—a high-velocity, recurring expense that demands sophisticated financial oversight.
The Problem with Token-Based Economics
Unlike traditional SaaS, which usually relies on predictable per-seat pricing, AI costs are volatile. Expenses fluctuate based on model selection, prompt length, and the volume of API calls. Ramp’s data reveals a striking lack of stability: the median business sees its AI spend swing by 58% month-to-month, with 61% of companies experiencing variations of 40% or more.
This unpredictability is compounded by multi-vendor usage. On average, businesses now utilize nearly three different AI providers, such as Anthropic, OpenAI, and Google. Without a unified view of these disparate invoices, finance teams are often flying blind, unable to discern which projects or departments are driving cost spikes. With 8.4% of companies now spending more than $10,000 monthly, AI has become a budget category that requires formal governance.
Bridging the Gap: Engineering vs. Finance
Ramp suggests that effective cost control requires a two-pronged strategy. While most companies focus on the “infrastructure layer”—the technical implementation where tokens are generated—they often neglect the “finance layer,” which is where sustainable cost reduction policies are enforced.
Key drivers of unexpected cost increases include:
- Model Drift: Teams upgrading from budget models to expensive premium tiers for production without adjusting budgets.
- Long-Context Inflation: Applications that ingest massive document archives or long conversation histories, leading to ballooning token consumption.
- Multi-Model Sprawl: The accumulation of various vendor contracts that lack centralized oversight.
A Framework for AI Financial Governance
To get spending under control, finance teams must bridge the gap between technical token reporting and business accounting. This starts with tracking spending across four dimensions: business unit, project, cost center, and model.
Furthermore, distinguishing between COGS (Cost of Goods Sold) and OpEx is critical. AI tokens that power customer-facing product features should be categorized as COGS, while those used for internal automation fall under OpEx. Mixing these categories distorts gross margins and obscures the true unit economics of AI initiatives.
Strategies for Sustainable Growth
To build a resilient budget, Ramp recommends moving away from aggregate company-wide limits. Instead, organizations should:
- Baseline current usage: Map the last three months of spend to the specific teams generating it.
- Set team-level targets: Establish soft targets to drive accountability before moving to strict enforcement.
- Mandate cost modeling: Require engineering teams to forecast token consumption for new projects before they reach production.
- Prioritize efficiency: Encourage engineering teams to implement AI token cost reduction tactics like prompt caching, batch processing for non-urgent tasks, and intelligent model routing.
By implementing anomaly alerting and chargebacks to specific cost centers, finance teams can turn AI from a mysterious, runaway cost into a manageable, transparent investment. In an era where AI is becoming the backbone of the enterprise, the ability to govern these expenses will be a defining factor in corporate profitability.
