IBM chairperson and chief executive Arvind Krishna, during a guesting on The Decoder podcast on 2 December 2025, warned that technology firms may struggle to recover multi-trillion-dollar investments in AI infrastructure because current data center costs create severe economic pressure on builders committing unmatched power capacity for future AI systems.
The Indian-American business executive noted that a 1-gigawatt data center requires about USD 80 billion to build at current market prices. He added that several key AI developers are pursuing between 20 gigawatts and 30 gigawatts of new capacity. This would translate into about USD 1.5 trillion to 2.4 trillion for a single large-scale expansion program.
He added that global AI infrastructure announcements now approach 100 GW. This is equal to nearly USD 8 trillion in estimated capital expenditure. Krishna argued that such an investment pool would require nearly USD 800 billion in annual profit generation merely to cover interest expenses under contemporary borrowing costs and depreciation patterns.
Depreciation was underscored as a central risk. The IBM chief explained that specialized accelerators, networking equipment, and cooling systems often lose value within 5 years. He then warned that rapid component turnover effectively doubles long-term capital pressure because operators must continually replace equipment even before recovering initial build expenses.
The interview compared his analysis with public remarks by OpenAI CEO Sam Altman. Altman previously encouraged annual additions of nearly 100 GW of power capacity to support advanced model development. Krishna expressed skepticism toward such scaling expectations and raised concerns regarding long-term feasibility without efficiency breakthroughs.
He also doubted the likelihood of current large language model architectures delivering artificial general intelligence. He estimated a probability between 0 percent and 1 percent for AGI emerging directly from present scaling trends, and called for new approaches that integrate additional computational and structural innovations beyond parameter growth.
Other tech leaders, like Andrew Ng, Arthur Mensch, and Marc Benioff, have expressed similar cautions. Several researchers, including Ilya Sutskever, have argued that pure scaling may have reached functional limits, reinforcing industry debates about whether massive capital spending can sustain meaningful performance gains in future AI systems.
Krishna still emphasized significant near-term value for enterprise users. He said AI can unlock trillions in productivity gains through workflow automation, accelerated analysis, and improved operational efficiency. However, he maintained that financial prudence remains essential as firms weigh long-term infrastructure strategies against volatile market conditions.
FURTHER READINGS AND REFERENCES
- Decoder With Nilay Patel and The Verge. 1 December 2025. “IBM CEO Arvind Krishna Says There is No AI Bubble After All.” YouTube. Available online
- Dwarkesh Patel. 26 November 2025. “Ilya Sutskever – We’re Moving From the Age of Scaling to the Age of Research.” YouTube. Available online
- Patel, N. 1 December 2025. “IBM CEO Arvind Krishna Says There is No AI Bubble After All.” The Verge. Available online
Photo Credit: Screen grab from The Decoder podcast published on YouTube and The Verge