What smart people are saying about the AI ​​ROI crisis

CIOs are concerned that AI budgets are starting to reach their limits and are not justifying the costs incurred.

Uber Chief Operating Officer Andrew MacDonald voiced growing concerns, saying he hasn’t yet seen direct improvements from AI spending.

The rise in budget consciousness stands in stark contrast to the freewheeling “token max” culture that some companies have begun to employ as AI agents. Although many companies did not adopt the leaderboards that popularized the first wave of TokenMax, AI costs are rising across the board.

Here’s what tech executives, engineers, and analysts have to say about the growing backlash:

Andrew McDonald, Uber COO

Uber Chief Operating Officer Andrew MacDonald opened up a discussion about the potential math around AI costs, saying the ride-hailing giant hasn’t realized productivity gains relative to overall billings.

“We don’t have that link yet, right?” he told Rapid Response. “Maybe implicitly, we’re shipping more products, but it’s very difficult to draw the line between one of those statistics and, ‘Okay, we’re actually producing 25% more useful features for the consumer.'”

McDonald’s comments came after Uber Chief Technology Officer (CTO) Praveen Nepali Naga made headlines when she said the company had exhausted this year’s Claude Code budget by April.

Sam Altman, OpenAI CEO

OpenAI CEO Sam Altman said that when he talks to business leaders, cost is at the forefront of their minds.

“We hear a lot of great stories from businesses, but the negative story is, ‘We’re spending more and more, people feel like they’re being very productive… but where’s the revenue, where’s the real productivity growth?'” Altman said during a virtual appearance at a recent AI event in Australia, according to Forbes Australia.

Altman said AI is so new that companies are still considering its implementation.

“My best answer to that is that it’s all still very new and it’s going to take a little more time for companies to figure out how to actually operate effectively and create these great new products,” he said, according to CommBank, which organized the event. “But I would be more concerned if we were still talking about the same issues a year from now.”

mark cuban

Mark Cuban said the issue is not token spending.

“Companies have long failed to integrate new technology,” Cuban wrote in X in response to an article about CEOs cutting token spending.

Rather, companies need to ask themselves, “As an AI-native company, what can we do to disintermediate our startup competitors?”

“That’s the innovator’s AI dilemma,” he wrote.

Jason Lemkin, SaaStr Founder

Jason Lemkin, sometimes referred to as the “godfather of SaaS,” said not all companies face the potential cash crunch equally.

“I think we’re going to go into a world where things are going to become more and more polarized, and there’s going to be people who are going to benefit more and more from AI spending,” Lemkin said on a May 28 episode of the 20VC podcast.

Lemkin said companies with “revenues well over $1 million per employee” will continue to use tokenmaxx “until there’s no tomorrow.”

“If you’re already very efficient, you’ll find more ways to leverage AI,” he says. “I think people who are less efficient, more organized, more traditional will become more skeptical as the years go on, especially if prices go up.”

Lemkin is the founder of SaaStr, the world’s largest community of B2B founders.

Shruti Gandhi, General Partner, Array Ventures

Shruti Gandhi, general partner at Array Ventures, said “token maxing” companies sound like companies that measure productivity by turning on all the lights.

“Spending more does not mean producing more results,” she wrote in X. “Most companies distributing access to AI to their employees have no idea what the AI ​​is being used for, whether it has changed the results in any way, or whether those results would have been the same in any case.”

He said some companies will “get serious about understanding ROI,” while others will realize too late that they should have paid more attention.

“The bigger question is: What tools do we actually need to figure this out? Because right now, no one has a good answer,” she wrote.

Richard Socher, CEO of Recursive Superintelligence and You.com

Veteran AI researcher Richard Sorcher said his company, Recursive Superintelligence, has outdone itself thanks to its massive token spend.

“It feels like our small team is operating at 10, 20 times the size,” Socher said on a May episode of the podcast “This Week in AI.” “The results we’re getting are expected to come from a team of maybe a few hundred people, but we’re a team of just a few dozen people.”

Socher, who previously served as head of AI research at Salesforce, said he expects to see more startups with budgets like his, where spending on AI computing exceeds spending on actual employees.

Recursive SuperIntelligence is named after the idea that AI might eventually be able to improve itself. Socher said startup costs are real if the company wants to drive the future of AI.

Gary Marcus, AI researcher and professor emeritus at New York University

AI critic Gary Marcus said “agent hysteria” drove much of the initial demand for AI.

“The key variable now is whether these expensive drugs offer a benefit,” Marcus wrote about X in May. “The Uber and Microsoft lawsuits, the dissatisfaction of[fully participating]Geohots, etc., may indicate that reliability and rewards may be lacking for many customers.”

Marcus said the AI ​​industry would be in trouble if other companies had the same experience as Uber.

“If enough other companies report the same thing, the bubble will burst.🫧,” he wrote to X.

Michael Burley, famous investor for “The Big Short”

Investor Michael Barry, who is often skeptical of Nvidia’s valuation, said the “token maxing” trend won’t last long.

“Tokenmaxxing is not just a heavy use of AI, and it is by no means a sustainable use of AI,” he wrote in a May post on Substack. “It’s quota-driven, leaderboard-driven, executive-mandated overconsumption.”

Burley called unlimited AI spending “a crazy, rushed, temporary phase.” He said that is completely unsustainable for the AI ​​industry.

“Everyone is already using it, and many are using it as desperately and as expensively as possible during this massive training phase.”

Known for his huge bets on the housing bubble in the mid-2000s that later led to his Hollywood career, Barry is a frequent commentator on Substack, where he shares his thoughts on the AI ​​bubble. Nvidia has felt the need to retract his comments in the past.

Akshat Bubna, Modal Co-Founder and CTO

Modal CTO Akshat Bubna said it is difficult for companies to determine which tokens are just a waste.

“It is true that 50% of internal token spending is completely wasteful, but it is difficult to know which 50% at this point,” Bubna wrote to X.

Bubna said he would like to see a better way to track usage.

“As an administrator, I want a dashboard that breaks down each person’s spending into summarized clusters,” he writes. “If you can draw a clear line around your values, it’s much easier to increase your spending.”

Sundar Pichai, Google CEO

Google CEO Sundar Pichai said he has heard the concerns directly.

“I’ve heard anecdotes from a lot of CIOs who are very concerned about how much their company’s budget is being blown,” Pichai said recently at Google’s flagship developer conference, I/O. “You can feel that as you talk to them. And I think this problem will only get worse as this year goes on.”

Pichai’s response is not to dismiss bean counters, but to tout more efficient AI models. He touted Google’s Gemini 3.5 Flash, a model the company says is “faster, lower cost and optimized for real-world tasks.”

“Flash is incredibly cost-effective, even considering the use of tokens,” he said.