Is AI delivering results? Today, OpenAI CFO Sarah Friar shared the scorecard you use to determine whether you’re actually getting economic value from your AI spending.
For years, Friar said, software success has been measured by adoption numbers: seats, active users, and updates. She argues that AI is different. AI needs to be measured by the work it actually performs.
“The fundamental economic question facing CFOs and other business leaders is whether the value of work completed by AI will grow faster than the cost of producing it,” Frier wrote in a blog post. To answer that question, she says, we need to dig deeper than simple metrics like cost per token.
She argues that the key metric for AI is what she calls “useful intelligence per dollar.” There are four factors: Is the AI completing the important work? What is the cost of each successful task? Can people rely on the results? And as usage increases, will the value per dollar increase?
In practice, this means that leaders need to track the amount of work completed by AI that meets defined quality criteria, add up the total cost of completing that work, and divide by the number of successful tasks to get the cost per successful task. From there, the test is whether people can reliably rely on the output and whether, over time, high-quality finished work increases faster than the total cost, with quality maintained or improved. Then, Friar explains, each dollar of AI will create more value, and computing will be at the center of that equation.
“Our job is to improve that equation with each generation: higher-performing models, faster and more reliable results, and lower costs for the work our customers need to do,” she wrote.
For hyperscaler OpenAI, compute is more than just a technology expense; it’s a strategic asset. As a private company, the company has not released formal capital investment guidance, but its Stargate initiative, announced in January 2025, outlines plans to invest up to $500 billion over about four years to build large-scale AI infrastructure in the U.S., with an initial target of about $100 billion and widespread build-out accelerating toward a capacity goal of 10 gigawatts in the U.S. by 2029. Just over a year later, demand has already surpassed that milestone. Because AI continues to accelerate. According to reports, OpenAI’s IPO could occur as early as this summer or as late as 2027. The company is already valued at $852 billion, approaching the $1 trillion range.
Finance chiefs have long led capital allocation and investor communications, but they are increasingly expected to contribute alongside CEOs to strategic decisions such as where to place their biggest long-term bets, such as AI investments.
McKinsey recently held its 24th annual Global CFO Forum. The forum brought together approximately 100 finance executives from more than 30 countries, representing some of the world’s largest organizations. said Andy West, a senior partner at McKinsey & Co. and global co-leader of the firm’s strategy and corporate finance practice. luck He conducted an informal poll asking CFOs whether their strategy departments currently report to them. About two-thirds raised their hands. Five years ago, it would have been less than a third, he says.
“We’ve been talking about AI at this conference for several years now,” West said. Last year, financial industry leaders were still experimenting with AI. This year, he said, the conversation has decisively shifted to company-wide transformation.
