In a recent interview, Arvind Krishna — CEO of IBM — voiced strong skepticism over the massive capital expenditures many tech firms are making to build data centers dedicated to artificial intelligence (AI). According to Krishna, at current cost levels it takes roughly US$ 80 billion to build a one‑gigawatt data center. Given some industry estimates suggest companies may aim for 20–30 gigawatts (or even more), that translates into a potential US$ 1.5–8 trillion investment, depending on the scale.
Krishna warned that to justify such spending, firms would need “roughly US$ 800 billion” in annual profits just to cover interest payments — a tall order by any measure. He also pointed out the rapid depreciation of AI hardware, noting that the specialized chips used in these centers typically must be replaced every five years — which further raises the ongoing cost and reduces the likelihood of a favorable return on investment.
On the broader question of whether current AI efforts will lead to the much‑hyped goal of artificial general intelligence (AGI), Krishna remained unconvinced. He pegged the chances of achieving AGI — under current technological paths like scaling large language models (LLMs) — at a mere “0–1%.”He suggested that reaching AGI might require integrating “hard knowledge” with LLMs and possibly new technological breakthroughs — a path he described as uncertain, calling success only a “maybe.”
Despite his skepticism on data‑center economics and AGI, Krishna acknowledged that today’s AI tools can still unlock “trillions of dollars of productivity” for enterprises. He argued that while the AGI dream may be far‑fetched under current approaches, more grounded applications of AI — particularly in enterprise settings — remain highly valuable and likely to deliver real returns.
