Companies Tighten AI Prompts as Token Costs Surge
The rapid adoption of large language models and collaborative AI Agents has introduced a volatile expense for companies: Token consumption. Costs can rise with lengthy prompts, generated responses and repeated calls between agents, making seemingly inexpensive queries significant at scale. The trend is turning AI spending into a governance issue as executives weigh productivity gains against unpredictable operating expenses and uncertain returns on investment.
Companies are responding by training employees to write more precise prompts, restricting access to costly tools and comparing models on both performance and price. As of Aug. 13, 2026, the related report did not identify specific companies or institutions, or disclose spending figures and Token volumes. The latest measures instead center on tighter usage policies designed to curb unnecessary model calls and reduce the risk of budget overruns.
All Coverage
1 original reportsThe Backstory
The history behind this eventRamp Launches Tool to Rein In Surging Corporate AI Costs
Corporate adoption of large language models is reshaping software budgets as predictable per-seat SaaS licenses give way to usage-based charges tied to tokens, models and computing power. Ramp said business spending on AI tokens rose 572% from June 2025 to June 2026, while the median company’s monthly outlay fluctuated about 58%. The volatility matters because invoices are split among providers and rarely map cleanly to teams or projects, limiting finance departments’ ability to forecast costs or assign accountability.
Ramp launched AI Token Spend Management on July 16, 2026, making it available to all Ramp customers and offering free standalone access to other businesses. The dashboard consolidates spending from Anthropic, OpenAI, Gemini and Cursor, flags anomalies and allows limits by user or API key. Ramp said it developed the product alongside more than 1,300 businesses managing over 100 trillion tokens a month. At AngelList, the tool found that missing prompt caching was costing $10,000 monthly; Ramp said potential savings averaged 12% of monthly AI spending.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.