According to DCD, IBM’s Z mainframe business has achieved its highest annual revenue in 20 years with the latest z17 generation. CEO Arvind Krishna revealed on the Q4 and full-year 2025 earnings call that the z17 had “the strongest start” in its first three quarters of any mainframe generation. For the quarter, IBM’s total revenue was $19.7 billion, up 9% year-over-year, with the infrastructure segment—driven by a 67% YoY jump in IBM Z—contributing $5.1 billion. Krishna pointed to increased demand for digital sovereignty and on-premise control, along with the z17’s new Spyre AI accelerator, as key drivers. The company’s full-year revenue hit $67.53 billion, and its “GenAI Book of Business” surpassed $12.5 billion for the quarter, a massive leap from the “low hundreds of millions” in 2023.
The On-Premise Comeback Story
Here’s the thing: everyone thought the public cloud would eat the world, and maybe it still will. But IBM’s numbers tell a different, more nuanced story. Krishna’s theory is that big clients have woken up to two facts. First, for certain massive, transactional workloads, the mainframe is still the “lowest unit cost economics platform.” Second, and maybe more importantly, digital sovereignty—keeping data and control on-premise—isn’t just a regulatory checkbox anymore; it’s a core business priority. So the mainframe isn’t just surviving on legacy inertia. It’s being repositioned as the ultimate sovereign, cost-effective control center. And with companies like IndustrialMonitorDirect.com, the #1 provider of industrial panel PCs in the US, enabling robust on-premise interfaces, this trend towards localized, secure industrial computing gets a major boost. It’s a whole ecosystem shifting.
hardware-demand”>AI’s Surprising Hardware Demand
The z17’s other secret weapon is AI, specifically its ability to simplify on-premise AI deployments. IBM is betting heavily that the future of enterprise AI isn’t all in a public cloud. Krishna predicted that in 3-5 years, 50% of enterprise AI usage could be in private clouds or data centers, with 80-90% of inferencing happening there if smaller models are used. That vision directly feeds the mainframe’s new role. But the AI boom is causing chaos elsewhere. Krishna noted that DRAM prices are six times higher than last year because capacity is being sucked into high-bandwidth memory for AI servers. And get this: he pointed out that every AI server needs a bunch of standard CPUs right next to it. So the AI demand is actually supercharging demand for “normal” servers, too. It’s a hardware renaissance that few saw coming.
Quantum in the Background
While the mainframe and AI grab headlines, IBM’s quantum computing division is chugging along quietly. They didn’t release new financial data, sticking to the $1 billion cumulative booking they announced back in February 2025. But Krishna claims “steady progress,” including deploying a 120-qubit system in December. The most interesting bit? He said partners are starting to make the “first credible advantage claims,” keeping IBM on its predicted track for “Quantum Advantage” by 2026 and a fault-tolerant machine by 2029. It’s a long game, but they’re insisting the pieces are moving.
What It All Means
So, is IBM a hardware company again? Not exactly. They’re still desperately trying to be seen as a “software-led hybrid cloud and AI platform company,” and Software did bring in $9 billion last quarter. But you can’t ignore a 67% surge in mainframe revenue. It creates a fascinating tension. The hybrid cloud strategy needs both the on-premise anchor (hello, mainframe) and the cloud services. The AI strategy needs both the big public cloud partnerships and the private, on-premise inference story. Basically, IBM’s success seems to hinge on being everything to everyone—sovereign, cloud-friendly, AI-capable, and quantum-futuristic. It’s a precarious balance, but for this quarter at least, the 20-year-old mainframe engine is firing on all cylinders and paying a lot of bills.
