IBM、衝撃的な四半期決算後も「AIはメインフレームを殺さない」と主張After shocking quarter, IBM insists that AI isn’t killing the mainframe

IBM used its latest quarterly earnings call to counter a growing perception that generative AI is eroding demand for its mainframe business, arguing that the two technologies reinforce one another rather than compete. The question matters because the mainframe remains one of IBM's most durable and profitable franchises, underpinning transaction processing at banks, insurers, airlines, retailers and government agencies. Any suggestion that the current AI wave could accelerate a shift away from that hardware carries weight for both the company's outlook and the enterprises that depend on it.

The rebuttal came alongside a results report that appears to have disappointed investors, and management used the moment to address the narrative directly. Rather than treating large language models and cloud-native platforms as substitutes for mainframe capacity, IBM framed them as complementary layers. The core of that argument is data gravity: a large share of the world's most sensitive and high-volume transactional data already lives on mainframes, and moving it wholesale to other environments is costly, risky and often unnecessary. In IBM's telling, AI workloads increasingly need to run close to where that data resides, which is likely to sustain rather than undermine mainframe relevance.

That positioning aligns with the direction of IBM's hardware roadmap. Recent generations of its zSystems machines have integrated AI acceleration directly onto the processor. The Telum chip introduced on-chip inference capabilities aimed at scoring transactions for fraud in real time, and IBM has continued to expand that approach with newer silicon and dedicated accelerator options designed to handle heavier AI models alongside conventional workloads. The pitch is that a bank can run an inference request within the same system that processes a payment, avoiding the latency and data-movement overhead of shipping information to a separate AI cluster. IBM has also promoted tooling intended to help modernize aging COBOL codebases, presenting generative AI as a way to make mainframe applications easier to maintain rather than a reason to abandon them.

The financial context helps explain the market's reaction. IBM's infrastructure segment, which houses the mainframe, is notoriously cyclical, with revenue tending to spike after a new machine ships and then taper as the product ages. That pattern can make a single quarter look weak even when the underlying installed base is stable, and it complicates any attempt to read AI-driven displacement into the numbers. IBM's broader business has shifted over the past several years toward software and services, anchored by its Red Hat acquisition and its hybrid-cloud strategy, so the mainframe is now one part of a larger portfolio that also includes the watsonx family of AI products and consulting work tied to AI adoption.

The debate sits within a long-running discussion about whether the mainframe is a legacy platform in decline or a specialized tool with continued staying power. Skeptics have predicted the format's demise for decades, pointing to the rise of commodity servers, public cloud and now AI as forces that should erode it. Defenders counter that regulated industries value the platform's reliability, security features and throughput for structured transactions, and that migration projects are frequently expensive enough to stall. The reality has tended to fall between those poles, with many organizations running hybrid setups that keep systems of record on the mainframe while building new customer-facing and analytical services elsewhere.

For customers, the practical takeaway is that IBM is betting AI demand can be captured on the mainframe itself rather than lost to competing infrastructure. Whether that thesis holds will depend on how enterprises choose to architect AI applications, how competitive the on-chip acceleration proves against dedicated GPU systems from vendors such as Nvidia, and how aggressively hyperscale cloud providers court the same workloads. IBM's argument is plausible given the entrenchment of mainframe data, but it remains a claim the company has a clear incentive to make. Investors and IT buyers are likely to watch upcoming quarters, and the reception of IBM's newest hardware, for firmer evidence of whether AI is expanding the mainframe's role or quietly narrowing it.