Enterprises Face Token Cost Shock as AI Agents Multiply Across Industries
Companies deploying AI agents at scale are running into unexpected costs as token consumption rises sharply. Goldman Sachs projects a 24-fold increase in token consumption over the next four years, and a new market for AI evaluation tools is emerging to help businesses manage the expense.
Companies that deployed AI agents expecting efficiency gains are now confronting a new problem: the cost of running those agents is rising fast.
Goldman Sachs projects a 24-fold increase in token consumption over the next four years as AI agents proliferate across industries. Tokens are the units of text that large language models process, and each interaction with an agent consumes them. As agents handle more tasks and interact with each other, the bills add up quickly.
The phenomenon is being called token cost shock. Enterprises that ran small pilots found the economics manageable. Scaling those pilots to production has revealed a different picture.
The problem has created a market for AI evaluation and stress-testing tools. Patronus AI, which builds tools to help companies measure and manage AI agent performance and cost, raised $50 million in a recent funding round. Investors are betting that businesses will pay for help understanding what their AI systems are actually doing and what it costs.
The broader technology sector is also watching a shift toward what industry observers call physical AI. Onsemi acquired Synaptics for $7 billion to integrate AI into industrial and automotive hardware. Startups like Six Robotics are scaling coordination software for unmanned systems. Researchers have developed neuromorphic artificial skin for humanoid robots.
Meanwhile, the Trump administration has implemented staggered release protocols for advanced AI models, including OpenAI's GPT 5.6 and Anthropic's Mythos and Fable. The protocols require customer-by-customer approval for safety reasons, slowing deployment for some enterprise users.
A widening pay gap has emerged between AI-focused roles and traditional tech positions, particularly in San Francisco, where high-earning professionals in frontier AI firms report that even large salaries struggle to keep pace with the city's cost of living.
