IBM CEO Arvind Krishna has expressed serious doubts about whether the massive spending by Big Tech on data centers for AI will generate a return on investment. Speaking on the "Decoder" podcast, Krishna questioned the financial logic behind the industry’s push to build out vast computing infrastructure in the race toward artificial general intelligence, or AGI. He noted that today’s data center costs are extremely high, estimating that filling a one-gigawatt facility requires about $80 billion in capital expenditure. With major companies committing to 20 to 30 gigawatts of capacity, the total capex could reach $1.5 trillion for a single company. When scaled across the entire industry, with global commitments reaching around 100 gigawatts, the total investment could hit roughly $8 trillion. Krishna argued that such a level of spending would require $800 billion in annual profits just to cover the interest on the debt, making a profitable return seem nearly impossible. He also highlighted the rapid depreciation of AI chips, which he said must be replaced every five years, further reducing the long-term value of the investment. While companies like Meta and Google are ramping up infrastructure plans—Google even suggesting space-based data centers—Krishna remains unconvinced. He contrasted this with OpenAI CEO Sam Altman’s call for the U.S. to add 100 gigawatts of energy capacity annually, a plan that would require trillions in investment. Altman believes OpenAI can generate returns on such spending, but Krishna said that’s a belief, not a proven strategy. Krishna also cast doubt on the likelihood of achieving AGI using current large language model technology, estimating the chance of success without a major breakthrough at just 0% to 1%. He joined a growing list of industry leaders skeptical of the AGI narrative, including Marc Benioff, who likened the pursuit to hypnosis, and Google Brain founder Andrew Ng, who called AGI overhyped. Mistral CEO Arthur Mensch described it as a marketing tactic, while OpenAI’s Ilya Sutskever recently stated that the era of scaling AI models is over, and that true progress now requires deeper research. Despite his skepticism about AGI, Krishna praised the real-world impact of today’s AI tools, saying they will unlock trillions in enterprise productivity. He believes that reaching AGI will require more than just bigger models and more compute—likely a fusion of LLMs with structured, hard-coded knowledge. But even that path, he said, is uncertain. “Even then, I'm a 'maybe,'” he admitted.