IBM CEO Arvind Krishna has cast serious doubt on the financial logic behind the massive data-center spending spree fueling the global race to artificial general intelligence (AGI). Speaking on The Verge’s Decoder podcast, Krishna said the economics simply don’t add up at today’s infrastructure and energy costs.
Krishna walked through what he called “napkin math” to illustrate the scale of the problem. Building and powering a single one-gigawatt AI data center, he said, now costs roughly $80 billion. If a single tech giant plans to deploy 20 to 30 gigawatts of capacity — a figure several companies are targeting — the capital expenditure skyrockets to $1.5 trillion for just one player.
“And if the industry ends up spending $8 trillion in CapEx,” Krishna added, “you need $800 billion in annual profit just to service the interest. There is no way that’s profitable at today’s economics.”
His skepticism extends to the technology itself. While companies like Meta, Google, Microsoft, and OpenAI are pouring billions into AI infrastructure — Meta repeatedly emphasizing “capacity” and Google even exploring space-based data centers — Krishna believes the current trajectory is unsustainable. He also assigned a 0–1% chance that today’s AI architectures will reach true AGI.
Despite soaring expectations, he argued that current AI infrastructure costs, energy demands, and compute requirements make long-term returns unlikely unless breakthroughs radically reduce expenses.
Krishna’s warning lands as tech giants escalate investments in GPU clusters, hyperscale data centers, and advanced cooling systems. Yet his message is blunt: without major technological shifts, the trillion-dollar AI arms race may never produce the financial returns investors expect.
