IBM CEO: rationing AI is a slow way to lose
Reported by Fortune1 source ↓
IBM CEO Arvind Krishna says businesses should address AI costs through engineering, model choice and spending controls rather than restricting access. Fortune reports that global spending on running AI systems will exceed spending on training them for the first time this year.
Key points
- Krishna argues that companies should fund selected AI projects to scale instead of rationing access
- IBM reports caching and other optimizations let the same hardware handle seven times more work
- In an IBM test, an open-weight model delivered a 52-fold design speedup for $26; a proprietary model delivered 204-fold for $299
- IBM says its systems had 98% of needed AI data in cache
- Krishna says about four-fifths of enterprise data is unstructured and largely untouched
Details
Fortune reports that IBM CEO Arvind Krishna argues production AI costs are prompting businesses to cap usage, meter spending and limit access to more capable models. He says companies should choose a small number of projects and fund them properly, rather than underfunding work they have already selected.
Krishna attributes costs in part to AI agents repeatedly reading company documents. The software is open source and available for companies to use, according to Krishna.
In an IBM test involving 3,000 AI agents, an open-weight model completed a quantum error-correction hardware-design task with a 52-fold design speedup for $26 in inference costs. A leading proprietary model produced a 204-fold speedup for $299, even with cached-data discounts, Krishna says. He argues that companies should match models to tasks rather than use one model for every workload.
He estimates that roughly four-fifths of enterprise data is unstructured and largely untouched, and argues that companies should keep control of the infrastructure and operating layer used to apply AI to that data.
Written from coverage as of
Images
- Fortune
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh