A new study shows that 62% of organizations said an unexpected AI expense materially altered a business decision over the past year, with 40% requiring board-level escalation, 33% implementing emergency spending freezes, and 25% delaying or canceling an AI initiative. The high costs stem from AI coding agents like Claude Code and Codex, which can work for hours, repeatedly calling frontier models and racking up millions of tokens. In response, AI model routers—software that lets organizations choose the right model for each task at the right cost—have become a hot enterprise tech area. Intelligent model routing can reduce inference costs by double-digit percentages, in some cases up to 30%. Startups like OpenRouter, reportedly in acquisition talks with Stripe at a valuation of up to $10 billion, provide a marketplace and unified gateway to hundreds of models, while Not Diamond automatically routes requests, and LiteLLM lets enterprises build their own infrastructure. Salesforce, Databricks, Cursor, Ramp, Meta, and Runway are also developing routers. OpenRouter co-founder Chris Clark said demand is driven by token-hungry agents like Anthropic’s Claude Code, released in mid-2025, noting that companies had no budgets in place for this maximalist AI adoption.