Asger Risom
IBM chief executive Arvind Krishna has cast serious doubt on whether the tech industry’s massive spending on AI data centers can ever pay for itself.
IBM chief executive Arvind Krishna has cast serious doubt on whether the tech industry’s massive spending on AI data centers can ever pay for itself. Speaking on the “Decoder” podcast, he said that at current infrastructure costs, the numbers simply don’t work — and that today’s AI hardware is unlikely to take companies anywhere close to true AGI.
His “napkin math” says the returns don’t come close
Krishna walked through the rough economics as he sees them. Building and filling a one-gigawatt data center today costs around $80 billion, he said. So when a single company commits to 20 or 30 gigawatts — something several AI giants now claim they’re planning — that’s up to $1.5 trillion in capital spending.
Looking across the entire AI sector, Krishna said public announcements point to roughly 100 gigawatts of planned compute capacity. Using his same estimate, that’s about $8 trillion in total commitments.
And with that level of spending, Krishna argued, the financial hurdle is enormous.
“Eight trillion dollars of capex means you need roughly $800 billion of profit just to pay for the interest,” he said. “There’s no way you’re going to get a return on that.”
Short equipment lifespans make the math even harder
Krishna also highlighted a factor that investors often worry about: how quickly AI chips lose value. He noted that the hardware inside these data centers depreciates on a roughly five-year cycle — after which the racks need to be replaced.
That aligns with concerns raised recently by investor Michael Burry, who has criticized the fast depreciation of AI chips and shorted Nvidia along the way.
Krishna doesn’t share Silicon Valley’s AGI confidence
When asked about OpenAI CEO Sam Altman’s argument that massive spending will eventually be worthwhile, Krishna was blunt: that’s a belief, not a guarantee.
Krishna said he puts the chances of reaching AGI with current technology at “0 to 1%.” He’s not alone. Leaders including Salesforce’s Marc Benioff, Google Brain founder Andrew Ng, and Mistral CEO Arthur Mensch have all expressed deep skepticism about the industry’s AGI forecasts.
Even inside OpenAI, cofounder Ilya Sutskever has recently said that simply scaling compute won’t deliver the next leap: “It’s back to the age of research again,” he said.
AI still has huge value — just not the trillion-dollar future some are selling
Krishna emphasized he’s not dismissing today’s AI tools. In fact, he expects them to unlock “trillions of dollars” in productivity across businesses. But reaching AGI, he said, will require entirely new breakthroughs — possibly combining large language models with more formal, knowledge-based systems.
Whether those breakthroughs arrive is unclear. “Even then, I’m a ‘maybe,’” he said.
Sources: Decoder podcast, IBM, Business Insider
