Companies are facing significant increases in their AI spending, with enterprise AI bills tripling despite a 98% reduction in per-token prices since late 2022. This paradox is driven by a massive surge in token consumption, fueled by the widespread adoption of autonomous AI agents. For example, a simple linear workflow in 2023 cost about $0.04 per interaction, but an orchestrated agentic system in 2026 costs roughly $1.20, about 30 times more.
Several high-profile instances highlight this issue: Uber reportedly exhausted its annual AI budget in four months before capping consumption at $1,500 per employee per month for agentic coding tools. Microsoft revoked licenses for Anthropic's Claude Code for some internal users, redirecting them to GitHub Copilot CLI. One company reportedly accrued a $500 million Claude bill in a single month due to a lack of usage limits, and Priceline saw a routine Cursor contract renewal jump by four to five times.
Analysts emphasize that while per-token prices have fallen dramatically (GPT-4 equivalent performance now costs approximately $0.40 per million tokens, down from $20 per million in late 2022), the sheer volume of tokens consumed by complex agentic workflows has caused overall costs to skyrocket. Forrester analyst Biswajeet Mahapatra noted that generative AI models can require up to 100 times more computing power than traditional AI systems. Gartner analyst Anushree Verma estimates that by 2027, 80% of enterprises will face significant budget overruns, with software tool costs potentially exceeding human development costs.
This "spending reckoning" has led companies to implement stricter controls. Recommendations include aligning licenses with actual employee needs, monitoring token consumption, introducing model selection controls, and choosing vendors that do not encourage excessive back-and-forth interactions. While some, like Palo Alto Networks, are implementing team-level budgets and centralized governance, the consensus is shifting towards managing consumption and tightly scoping use cases rather than just negotiating rates, as the market enters a phase of higher adoption, cheaper models, and stricter scrutiny of how AI is actually being used to deliver value.
