IBM Chairman and Chief Executive Officer Arvind Krishna has questioned whether the economics of the current artificial intelligence infrastructure race can justify the scale of spending being committed by major technology companies, saying the capital required to build data centres for artificial general intelligence may be difficult to recover under current assumptions.

Speaking on The Verge's Decoder podcast with editor-in-chief Nilay Patel, Krishna said it costs about $80 billion to fill a one-gigawatt data centre at today's prices. A company committing 20-30 gigawatts of capacity would be looking at about $1.5 trillion of capital expenditure, he said.

Krishna said announced commitments across the industry for AI infrastructure aimed at artificial general intelligence, or AGI, appeared to add up to about 100 gigawatts. At current costs, he estimated that would translate into about $8 trillion of capital expenditure. "It's my view that there's no way you're going to get a return on that," Krishna said, adding that $8 trillion of capital expenditure would require roughly $800 billion of profit "just to pay for the interest."

Krishna's comments come as OpenAI and other AI companies continue to sign large infrastructure and chip agreements to expand computing capacity for training and running advanced models. OpenAI has defined AGI in its charter as highly autonomous systems that outperform humans at most economically valuable work.

In September 2025, OpenAI and Nvidia announced a strategic partnership to deploy at least 10 gigawatts of Nvidia systems for OpenAI's next-generation AI infrastructure. Nvidia said it intended to invest up to $100 billion in OpenAI progressively as each gigawatt is deployed.

In October 2025, AMD and OpenAI announced a multi-year agreement under which OpenAI would deploy six gigawatts of AMD GPUs, starting with an initial one-gigawatt deployment in the second half of 2026. AMD also issued OpenAI a warrant for up to 160 million AMD shares, vesting as deployment milestones are met.

OpenAI has also been linked to large cloud and data centre commitments with Oracle. Reuters, citing a Wall Street Journal report in September 2025, said OpenAI had signed a roughly $300 billion agreement to buy computing power from Oracle over about five years.

The spending plans have intensified investor scrutiny over the financial model behind generative AI. Reuters reported in February 2026, citing a person familiar with the matter, that OpenAI was targeting roughly $600 billion in compute spend through 2030 and expected more than $280 billion in revenue by the end of the decade. The report also said OpenAI CEO Sam Altman had previously spoken of spending $1.4 trillion to build 30 gigawatts of compute infrastructure.

Krishna did not dismiss generative AI's commercial use. He said the technology was "incredibly useful for enterprise" and could unlock productivity gains. His objection was narrower: he gave "really low odds" to the current set of known technologies leading to AGI.

Krishna said AGI would require more than the present large-language-model path. He said the industry would need a way to fuse "knowledge with LLMs", adding that even if that happens, he would still remain uncertain about the outcome.

IBM's own AI strategy has focused on enterprise adoption through products such as watsonx, hybrid cloud tools and consulting services. In May 2025, IBM said a CEO study showed business leaders expected the growth rate of AI investments to more than double over the next two years, but only 25 percent of AI initiatives had achieved the return on investment expected.

IBM has also had its own earlier experience with high-profile AI bets. The company's Watson system won Jeopardy! in 2011, but Krishna told The Verge that IBM's early attempt to push Watson into healthcare was "inappropriate" and that the company's earlier AI approach had been too monolithic.