In July 1993, Lou Gerstner had been running IBM for three months and was already being asked where he planned to take the company. The situation was bad: IBM had lost more than US$8 billion in a single year, and its mainframe business was in decline. Gerstner left behind a line that would become closely associated with his tenure: “The last thing IBM needs right now is a vision.” He did not believe the company lacked ideas about the future. IBM had spent years anticipating technological change and had often been right. The problem was different: deciding what to do with each business and then executing those decisions.
Thirty-three years later, the mainframe once again caused problems for IBM in a quarter. Krishna chose to look further ahead, talking about demonstrating quantum advantage before the end of the year, having a fault-tolerant machine in 2029 and a market that, by his estimate, could be worth a trillion dollars by the end of the 2030s.
Gerstner had come from outside. Before IBM, he had been at RJR Nabisco, in the world of cookies and cigarettes, and knew relatively little about the company from the inside. Krishna took the opposite path. He joined IBM in 1990 and never left. Born in 1962 in West Godavari, Andhra Pradesh, the son of an Indian Army general, he studied at IIT Kanpur, earned his doctorate in Illinois and began his career as a researcher at the Thomas J. Watson Research Center. Thirty-five years later, he ended up running IBM.
And now the problems in front of him move on completely different timelines.
The most immediate one is weighing on him today. In June, with the quarter almost over, several of IBM’s largest customers —“basically the Fortune 100, and not even all of it,” he says— had to reshuffle their budgets. More money went into servers, storage and memory. And the increase was substantial: memory now costs three or four times what it did less than eighteen months ago, while networking infrastructure has risen between 60% and 80%. That money had to come from somewhere, so other purchases were pushed back.
Krishna insists those deals were not lost, only delayed, and says a third of them have already returned. He is now watching how much more comes back to determine whether this was a temporary issue or something more serious.
He believes September will provide the answer. That is why IBM lowered its annual guidance from “more than 5%” to between 4% and 5%. Where it ends up within that range will depend on how much of that business returns.
The Starbucks case also came up. The company is replacing an IBM tool with software developed in-house, and Krishna said the decision did not surprise him. The version Starbucks had been using is almost ten years old and, in his view, software whose main selling point is ease of use is among the first categories likely to be exposed to AI agents.
Even so, he expects IBM to sell more to Starbucks next year. And with AI, he acknowledges another problem: when it comes to productivity tools, it is still difficult to show in hard numbers how much they actually improve a business. If that impact does not show up in the results, economists and CFOs begin to question it.
That is where the comparison with 1993 becomes most uncomfortable: the mainframe. A journalist put the question to him directly. Back then, IBM also blamed the decline on the economy and temporary problems. Eventually, it became clear that something deeper was happening. Why should this time be different?
Krishna explains it with a fairly simple calculation. In 1993, Unix offered a cheaper alternative and many workloads moved there. Today, he says, the economics are different: for systems that require high resilience, security and the ability to handle spikes in demand, keeping them on the mainframe costs between five and fifteen times less than moving them to another platform.
He also acknowledges where that advantage ends. If a cheaper option appears for a particular workload, sooner or later that workload will move, even if the process takes five to fifteen years. In the end, cost decides.
The figures he uses to defend the mainframe tell another part of the story. The Z17, launched in May 2025, has already reached 130% of the capacity of the previous generation, and 85% of customers are adding capacity. These transitions also take time: nobody replaces a system that authorizes credit card payments in a matter of months.
The 2029 clock
Here, the timelines become much longer. Krishna often explains it through an example from molecular simulation. In June 2025, his team could barely work with a five-atom molecule —“you can do that on paper if you’re smart enough,” he says. By November, it had reached 300 atoms; by April, 12,000, roughly half a protein. The next target is 30,000.
IBM plans to invest more than US$10 billion over five years and has already put some dates on the calendar. Starling, IBM’s first fault-tolerant machine, is expected in 2029. Blue Jay, a larger system, would follow in 2033.
There is also Anderon, a quantum wafer fabrication plant to be built in Albany. In May, the U.S. Department of Commerce announced plans to commit US$1 billion from the CHIPS Act to the project. IBM will contribute the same amount. In July, the company also acquired HRL Laboratories, the Malibu laboratory formerly owned by Boeing and General Motors, adding another qubit technology. And in September, it expects to bring online in Amaravati —capital of the state where Krishna was born— one of India’s first physical quantum computers.
His explanation for choosing Amaravati is fairly practical: he needs large numbers of people who understand mathematics and physics and can translate real problems into the language of these machines. “Where are you going to find enough people…?” he asks. When it comes to China, however, he admits he is not relaxed and is closely watching what it is doing. “I’m paid to be paranoid, so I wouldn’t say I’m not worried.” He believes the United States is still ahead, but not by much: perhaps a couple of years.
Two years ago, Krishna promised to demonstrate quantum advantage using IBM hardware before the end of 2026. A few weeks ago, he was asked whether he still stood by that deadline. He answered with one word: “Absolutely.” He has four and a half months left.
And that is where one of the tensions in his management becomes clear. He has put a date on a technology that moves on much longer timelines, just as the mainframe once again hurt IBM’s quarterly results. Gerstner chose a different route: for a while, he stopped talking about the future and concentrated on putting IBM back in order.
Krishna faces a different challenge: protect the business that works today, transform the company and sustain a long-term bet, while the market checks his numbers every three months.
