OpenAI CFO: 4 Questions for AI Spend ROI

OpenAI CFO Sarah Friar has proposed a new framework for evaluating the economic value of artificial intelligence investments, moving beyond traditional software metrics. In a blog post, Friar argued that AI success should be measured by the actual work it accomplishes rather than just adoption rates like seats or active users. The core economic question for business leaders, she stated, is whether the value generated by AI-completed work outpaces the cost of its production.

Friar introduced the metric "useful intelligence per dollar," which comprises four critical elements: whether AI is completing important tasks, the cost associated with each successful task, the reliability of AI outputs, and whether each dollar spent yields increasing value as usage grows. To implement this, leaders should track the volume of AI-generated work that meets a specific quality standard, calculate the total cost of completing this work, and then divide by the number of successful tasks to determine the cost per successful task. The subsequent evaluation involves assessing the dependability of AI outputs and observing if high-quality output growth surpasses total cost increases over time, while maintaining or improving quality.

Friar emphasized that for AI providers like OpenAI, compute is a fundamental strategic asset, not merely a technological expense. While OpenAI, as a private entity, does not issue formal capital expenditure guidance, its "Stargate" initiative, announced in January 2025, details a plan to invest up to $500 billion over approximately four years. This investment is earmarked for developing large-scale AI infrastructure within the United States. Friar noted that OpenAI's objective is to continuously improve the AI equation with each new generation of models, aiming for increased capability, faster and more reliable results, and reduced costs for customer-required tasks.