IBM CEO Warns Against Unrealistic AI Spending Expectations

IBM's CEO Arvind Krishna has shared his concerns about the AI spending boom in an interview, using numbers to make his case. He estimates that the committed AI buildout globally will require between $6 trillion and $8 trillion in spending, which he believes is not justified by the revenue it would generate.

Krishna's math is based on the cost of semiconductors for 1 gigawatt of data center power, which he estimates to be $60 billion to $80 billion. With companies committed to roughly 100 gigawatts of AI buildout, this translates to a total spending of $6 trillion to $8 trillion.

Krishna assumes that the payback period for these investments would be five to seven years and expects high-margin AI services at 20% to 30% margins. However, he believes that the market is pricing in an extra $1 trillion to $2 trillion in annual revenue, which he thinks is unrealistic.

Krishna also expects that the largest AI models will become commodities with low switching costs, leading to only a few companies surviving long-term. He mentions that Big Tech companies like Amazon, Alphabet, Microsoft, and Meta Platforms are spending heavily on AI infrastructure, with combined spending expected to reach $725 billion in 2026.

Krishna's views contrast with the market's enthusiasm for AI, which he attributes to a lack of understanding about the underlying math. He believes that distribution will determine the outcome, with companies having an existing consumer footprint aligned to AI having a 'pretty good chance' of winning.