Power Systems Has A Great Quarter; System Z, Not So Much

July 27, 2026 Timothy Prickett Morgan

We have been waiting for a long time for Big Blue’s Power Systems business to get a big bump, and it looks like it has finally happened. And a handful of Fortune 100-level companies deferred their System z17 mainframe purchases as they allocated budget to buying servers for AI and generic – generally X86 machines, and certainly not from IBM – as CPU, GPU, DRAM memory, and flash storage prices are all skyrocketing. And last week, when IBM warned about this push out of System z sales, Wall Street predictably freaked all the way out, sending IBM’s stock down 25 percent at one point.

That stock price drop is an overreaction, but then again, all stocks seem to be overvalued (as they often are) and investors and algorithms both are jumpy.

IBM’s core “real” systems business actually took the mainframe hit pretty well, and both last week and this week, Arvind Krishna, the company’s chief executive officer, and Jim Kavanaugh, its chief financial officer, did their best to get everyone to calm the heck down.

First of all, on the System z front, if you look at the first five quarters of revenues for machines using the “Telum” z16 and “Telum II” z17 processors, the z17 revenue stream is actually 30 percent higher than the z16 stream averaged across those quarters. Some revenue bump is memory and flash price increases, which as you know from reading The Four Hundred have been crazy since the beginning of 2026, but a lot of it is driven by capacity increases. The z17 line had a very strong start last summer, and it is weaker now. System z sales were off 42 percent year on year, which was a shortfall of several hundred million dollars and which has a drag-along effect on database and middleware sales on the mainframe. (IBM charges by capacity for monthly rentals of its mainframe software, so when there is not a MIPS increase, there is not a software bill increase.) By the way, one third of the revenue that got pushed out from Q2 2026 has already been fulfilled in the first few weeks of Q3 2026.

Everybody has to take a chill pill. . . .

Kavanaugh said as much here: “We see no evidence of clients moving off the mainframe. Clients continue to invest in IBM Z to modernize mission critical workloads, with a focus on resiliency, security, and increasingly enabling AI on the platform. AI is driving incremental capacity growth and new workloads as clients look to run AI closer to their most sensitive data. We are seeing strong early adoption of our AI innovations, with nearly 50 percent of z17 customers investing in AI capabilities with the Spyre accelerator. And clients deploying watsonx Code Assistant for Z are growing MIPS capacity three times faster than those who are not.”

The mainframe business might have had a hiccup, but the Distributed Infrastructure segment, which includes sales of Power Systems servers and capacity on the PowerVS cloud as well as all disk, flash, and tape storage products, had a 37 percent revenue increase and, significantly, has a $500 million backlog right now.

IBM has never given out backlog data for this business, so we don’t have a “compared to what” to reckon this against. But a half billion dollars sure does sound like a reasonable number in its own right. It is not clear why IBM has such a big backlog, but getting memory and flash is as tricky for IT system makers as it is for customer right now.

We do not know the breakdown of System z revenues from Power Systems revenues from Storage revenues because IBM deliberately wants to obfuscate. It makes no sense to merge Power Systems and Storage, except perhaps for the Power-based storage products such as the DS8000 and DS9000 arrays that are based on Power. (Which I estimate sales of in my own model and which I will get into in a minute.)

What we do know is that IBM’s sales for the Infrastructure division. This brings together all server and storage sales plus Power Systems utility rentals in the cloud, plus operating systems for the Z and P machines – if Big Blue is now calling the mainframe IBM Z, which we do not, why not be consistent and call the Power-based line IBM P? What is wrong with being consistent and making sense? – and tech support for the whole shebang. In Q2 2026, IBM’s Infrastructure division had $3.84 billion, down 7.4 percent year on year but up 15.3 percent sequentially. Gross profits for Infrastructure were down 12.1 percent to $2.24 billion, representing 58.4 percent of revenues. This is about average, which is what matters. But pretax income also followed aggregate revenues down, sinking 13.5 percent to $835 million. The Power Systems business is just nowhere as profitable as the System z business, and never has been.

Within this, Hybrid Infrastructure posted $2.58 billion in revenues, down 10.3 percent in the quarter, while Infrastructure Support fell by a single point to $1.26 billion.

As best I can figure, this was the best quarter for Power Systems revenues in several years. Power Systems revenues based on sales to customers and distribution in the channel came to $539 million in our model, up 41 percent. An additional $94 million was generated selling Power iron to the Storage division, up 40 percent. So overall Power Systems revenues (including PowerVS subscriptions) were $633 million, up 40.9 percent year on year.

The trend, as you can see above, has been positive for Power Systems, but a lot of the sales in Q2 might have been driven by customers trying to beat out future and higher DRAM, flash, and CPU costs and ordering now instead of later, as they have been warned by Big Blue to do. Of course, some of the revenue bump is due to higher prices for these and other components.

When you look at this chart below, you can see why there was a steep decline in System z revenues in Q2 2026:

Look at that spike two quarters ago, which was well after Power11 and z17 systems had been launched and were ramping nicely. Those revenues for Hybrid Infrastructure were the highest we have seen in a long time.

If you add all those divisions up, IBM’s overall revenues were up 1 percent to $17.16 billion, with gross profits down seven-tenths of a point to $9.91 billion, pre-tax income off 4.5 percent to $2.48 billion, and net income down 1.3 percent to $2.17 billion.

IBM is burning cash a little faster than it probably likes, as you can see, with the company only having $8.13 billion in the bank as the quarter came to a close. As far as we know, IBM thinks it will have an extra $1 billion over forecasts for all of 2026, and there are no indications yet that IBM will cut employees faster than its normal rate of attrition much less cut its dividends and share buybacks. IBM’s AI efforts have removed $4.5 billion in costs to date; we do not know how many people it has removed and how that is reckoned in those cost savings.

I like to look at the core “real” IBM systems business as a whole, including middleware but not databases and application software, to get a sense of how the company is doing. In my model, I reckon that the core real IBM systems business, including the Red Hat stack of server software but not including any development tools or databases in RHEL, came to $9.07 billion, up a smidgen year on year, with a pre-tax income of $4.27 billion. Red Hat grew 11 percent to $2.21 billion, and of this, I believe that $1.55 billion was for datacenter systems software, also up 11 percent, with a pre-tax income of maybe $242 million. If you take out Red Hat to get the indigenous IBM systems only, that comes to $7.52 billion, down 1.8 percent year on year, and I estimate that pre-tax income for this business was a tad over $4 billion, representing 53.6 percent of that total.

This core IBM systems business, with and without Red Hat, is doing just fine and is having perfectly reasonable pressures given the complexity of the situation caused by the GenAI boom.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search