IBM CEO Warns Against Rationing AI Tokens

Arvind Krishna, CEO of IBM, argues that businesses are hindering their own productivity by rationing AI tokens and capping spending to manage the high operating costs of running AI systems at scale. He asserts that such restrictions are a "slow way to lose" for enterprises.

Krishna advocates for engineering optimizations to improve the economics of enterprise AI. He specifically highlights caching—a technology IBM developed through an open-source project with Red Hat and Google—to prevent the repetitive reading of data and reduce costs.

To maintain data sovereignty and avoid reliance on proprietary pricing, Krishna recommends using open-weight models on company-controlled infrastructure. He emphasizes that organizations must own their AI operating layer to govern workloads and ensure predictable costs rather than relying on vendor-controlled meters.