IBM CEO says there is 'no way' spending trillions on AI data centers will pay off at today's infrastructure costs

In the escalating competition for artificial general intelligence (AGI), AI companies are pouring billions into data center infrastructure. Arvind Krishna, the CEO of IBM, has shared his insights on the economic implications of these investments. During a recent earnings call, Meta frequently referenced terms like "capacity" and "AI infrastructure," highlighting the industry's focus on expanding data center capabilities. Google has even announced plans to explore building data centers in space. However, the pressing question remains: can the revenue generated from these data centers ever offset the staggering capital expenditures involved? On the "Decoder" podcast, Krishna expressed skepticism, stating that there is likely "no way" these companies will recover their investments in data centers under current cost structures. He provided a stark example, noting that it costs around $80 billion to equip a one-gigawatt data center. For a company committing to 20 to 30 gigawatts, that translates to a potential capital expenditure of approximately $1.5 trillion. Krishna also highlighted the issue of AI chip depreciation, which poses a significant challenge for companies operating these data centers. "You have to utilize them within five years, or else you’ll need to replace them entirely," he explained. His comments come amid growing concerns from investors, such as Michael Burry, regarding the depreciation of tech companies like Nvidia, contributing to a downturn in AI stock prices. Pointing to the overall commitments in the push for AGI, Krishna estimated that the global investments in this space could reach about 100 gigawatts, equating to an eye-watering $8 trillion in capital expenditures. "To cover the interest on that, you would need around $800 billion in profits," he warned. This monumental financial commitment has led AI companies to seek external support. Recently, OpenAI’s CEO, Sam Altman, proposed adding 100 gigawatts of energy capacity annually in a letter to the White House’s Office of Science and Technology Policy. However, Krishna diverged from Altman’s optimistic outlook, emphasizing that he does not believe the existing technologies will lead to AGI anytime soon. He assigned a likelihood of 0-1% for achieving AGI without further technological advancements. Krishna is not alone in his skepticism; industry leaders like Marc Benioff and Andrew Ng have also expressed doubts about the race toward AGI, with Ng labeling it as "overhyped." Meanwhile, OpenAI cofounder Ilya Sutskever remarked that the era of scaling models is behind us, suggesting a shift back to research rather than mere expansion. Despite these challenges, Krishna remains optimistic about the current AI technologies, stating they hold the potential to unlock trillions in productivity for enterprises. However, he believes that realizing AGI will necessitate new technological breakthroughs beyond the current trajectory of large language models. When asked about the feasibility of achieving AGI, he concluded that he remains a "maybe."