IBM chief executive Arvind Krishna has a lot on his mind. “Executive orders. A dynamic market. Where’s the AI trade? What do clients want to do? The end of the quarter, product innovation. Name it,” he says.
Krishna became the company’s CEO and chairman in 2020, just in time for the AI boom. To him, steering a 115-year-old technology giant through that upheaval requires being bold. “I actually think that trying to muddle along without taking risk is the riskiest path,” he says. “It leads to eventual falling off a cliff.”
IBM trains small, efficient models of its own—some of which run on its mainframe computers that many top banks and other enterprise clients still rely on. If a client’s application needs a larger model, IBM rents one from Google, OpenAI, or Anthropic. Krishna’s central bet is that, in the end, models will become commodities. That doesn’t mean they’re not valuable. “What I mean is that you can use any one of them, because they’re close enough,” he says. Krishna believes that 90% of the value will eventually accrue not to the companies training large AI models, but companies like IBM that build applications on top.
His prediction is yet to fully pan out. Though IBM’s shares have nearly doubled since Krishna took the reins, they tumbled 25% in one day in July as customers shifted spending from mainframes to AI infrastructure and have yet to recover to their highs. “We did not adapt and move quickly enough,” Krishna wrote to investors. Meanwhile, leading model builders have swallowed some of the most successful AI-native application startups, like Cursor and Windsurf.
“We are not quite there yet,” Krishna says. But he believes proprietary data will be a differentiating factor rather than raw model-power, in which case trust becomes key. He believes IBM’s reputation working with enterprises in highly regulated sectors will be an advantage.