IBM's Z mainframe business has achieved the highest annual revenue with its latest generation for the past 20 years.

Revealed during the company's Q4 2025 and full-year earnings call this week, CEO Arvind Krishna revealed that the z17 system has "the strongest start" in the three quarters since it launched of any generation of IBM's mainframe systems.

Total revenue for the quarter was $19.7 billion, up nine percent Year-on-Year (YoY), and up from the previous quarter's $16.3bn. Of that, the infrastructure segment contributed $5.1bn, up 21 percent. IBM Z increased 67 percent, YoY.

Speaking on the mainframe's success, Krishna noted that each generation of mainframe has performed better than the last, and this is "continuing improvement."

He theorized that the success of the z17 comes in part from an increased interest in digital sovereignty or "on-premise control," along with economic value. "I think more and more clients have woken up to that for certain workloads, the mainframe is actually the lowest unit cost economics platform, and that is really important."

He added that the latest iteration also simplifies AI deployments for customers, including the addition of the Spyre AI accelerator, which was made available on the z17 in October 2025.

Revenue for the full year was $67.53bn, of which infrastructure contributed $15.72bn. This is up from $62.75bn and $14bn the year prior, respectively.

Despite the strength in the mainframe business, the company remains keen to continue its strategy of positioning IBM as a "software-led hybrid cloud and AI platform company," rather than a hardware and infrastructure focused business. Within this, Software remained the largest revenue contributor for IBM, bringing in $9 billion in the quarter.

The company's "GenAI Book of Business" surpassed $12.5bn for the quarter, which CFO Jim Kavanaugh noted was a growth from being in the "low hundreds of millions of dollars" in 2023 when the company first began reporting on this.

Elaborating on how IBM sees generative AI playing out for the company, Krishna hypothesized that in the next three to five years, 50 percent of enterprise AI usage will be in either a private cloud or their own data centers.

"There's also an efficiency question. So if what's being used on-premise is smaller models, then actually, it could be that 80 to 90 percent of all the inferencing is really in a private cloud or on-premise," he said. Because of this, the company is leaning on increasing its support for these workloads, including establishing a "sovereign core offering" this month - a software offering that can increase the control over cloud deployments.

This expectation for more on-premise AI also leans into the z17 mainframe with its expanded AI capabilities from previous mainframe generations.

Also noted during the questions and answers segment of the earnings call was the "volatile" server dynamics. Krishna said that memory DRAM prices are currently six times that of last year, and that is because "a lot of the capacity is moving over to HBM - or high bandwidth memory - which is required for AI servers."

"I personally believe as long as that dynamic is there, those pricing issues are going to be there through the year. Now the demand side. There is no AI server without a bunch of CPUs right next to it. So the reality becomes that the AI demand also drives demand for normal servers that in turn feed and load up those servers," he added.

Despite this, Krishna said that he believed this might be "a little bit of an issue" but is unlikely to be any real "headwind to us" on their hybrid cloud offerings.

No data was offered on the company's quantum computing business, as has become standard practice since it revealed it had booked $1 billion worth of cumulative quantum business in February 2025.

Krishna, however, said the company was making "steady progress" having "advanced our development roadmap, improved error correction capabilities, and expanded ecosystem partnerships."

He added: "In December, we deployed our first 120-qubit IBM Quantum Nighthawk-based system for use by our clients. Back in 2024, we predicted that we'd see Quantum Advantage by 2026, and with the help of IBM hardware, software, and rapid cycles of learning, our partners in the scientific computing community are starting to make the first credible advantage claims. We remain on track to deliver the first large-scale fault-tolerant quantum computer by 2029."

Other key metrics for the quarter include IBM's adjusted EBITDA, which grew 17 percent YoY, up $2.8bn, gross profit margin was 60.6 percent, and operating margins 91.8 percent.

For the full year, profit margins were 58.2 percent, and operating margins 59.5 percent. IBM finished the year with a free cash flow of $14.7bn.