IBM’s Power Systems Stalls A Bit As Power8 Wanes
October 31, 2016 Timothy Prickett Morgan
With the Power9 chips slated for some time in the second half of next year, and the Power8 systems not being upgraded with a formal Power8+ variant, it is not surprise to us that sales of Power Systems iron based on the Power8 chip that launched in April 2014 are waning. It is the nature of every product cycle to have its ramp, its peak, and its slide. The slide is just a way to extract profits, with the ease of sales and improving yields on chips compensating for the fact that Big Blue has to discount more to move iron.
In the third quarter, sales of Power Systems were down considerably, but there was some good news with regards to IBM’s concerted efforts to ramp up sales of Power machinery running Linux as the normal cycle of sales of IBM i and AIX systems based on Power8 chips slows.
In a conference call with Wall Street analysts, IBM chief financial officer Martin Schroeter, explained the situation thus, and clearly the focus is on profits, not revenues: “Our Power performance reflects both our performance in a declining Unix market, as well as our growth in a growing Linux market. While our margins were relatively stable at the high end of Power, mid- and low-end margins were down, driving a decline in overall Power margins. We have been shifting our platform to address Linux, and in the third quarter, Linux grew at a double-digit rate, and faster than the market. It now comprises over 15 percent of our Power revenue. Supporting that is our success with HANA, where we are bringing in new clients, and we are replicating this strategy with others.”
Schroeter added that one of the largest hyperscale data centers in China being operated by one of the world’s largest Internet service providers had chosen Power systems, although he did not say who it was, how many they installed, what workloads they were driving, and if IBM itself was the system supplier for the company. (It is not a foregone conclusion that Big Blue was the peddler of the systems, not with Tyan, Wistron, and Supermicro making the Power System LC machines for IBM itself. There are a few other indigenous Power server suppliers in China as well, and it stands to reason that these vendors are the ones that won whatever deal that Schroeter is referring to. (Lenovo, which bought IBM’s System x, BladeCenter, and Pure Systems business, has no inclination so far to create and sell Power-based systems, although it is monkeying around with ARM machinery in a few supercomputing centers in Europe and the United States.)
The new C-Style Power 850C, 870C, and 880C machines, which have been announced in the past month, as well as the updated Power Systems LC models that came out a little earlier, were not out early enough to have an impact on IBM’s Power Systems sales in the third quarter. But there could be an uptick in the fourth quarter as customers who are desperate for capacity opt to get a Power8 machine now rather than try to wait it out and see what Power9 machinery looks like in the second half of 2017. We don’t expect for the Power9 systems to be announced until sometime between September and November, but that is just a guess. We know that the US government needs the initial installations of Power9 nodes for the “Summit” and “Sierra” supercomputers that are being acquired by the Department of Energy as pre-exascale machines. But that does not mean the systems will be commercially available for everyone else, and we suspect that Google and Rackspace Hosting, who are strong OpenPower partners, will be at the front of the line after the US government once Power9 chips become available.
Our point is, there is a lot of time between now and when the Power Systems line will get Power9 chips in machines with one or two sockets. It could even be early 2018, but we certainly hope not. In that case, it would have taken four years to jump between Power chip generations, and that is far too long. To meet Intel’s future “Skylake” Xeon processors, IBM needs to get Power9 into the field sometime in the mid-to-late summer. Earlier will be even better.
The System z mainframe line had margin expansion as companies activated cores on existing machinery but revenues were down because IBM is not selling a lot of iron based on the current z13 processors. We expect for the z14 processors to be launched more or less concurrently with the Power9 chips, and it is reasonable to expect for IBM to unveil the z14 chips either at the International Solid State Circuits Conference in February or the Hot Chips conference in August. IBM i customers don’t care much about what happens to the mainframe, but so long as this remains a profitable business in terms of both hardware and software sales, that is less pressure on the Power Systems line to make money. So, basically, IBM i customers should hope that mainframes remain popular among the few thousand customers that still invest heavily in these throughput-oriented beasts.
IBM does not break out revenues for the System z or Power Systems lines, and it does not even give revenue growth or shrinkage rates for the product lines as it used to. But we play around with the numbers IBM provides and figure out as best we can what systems sales look like. In the third quarter, based on IBM’s statements and some math we do in our model, we estimate that IBM sold $1.2 billion in hardware, down 25 percent year-on-year. This was mostly for servers but also included storage and a smattering of networking gear it resells from other vendors such as Cisco Systems and Lenovo), with gross profits of $440 million. IBM also sold another $434 million in operating system revenues, down 11 percent and mostly from monthly charges for System z mainframe stacks but also including a smattering for AIX and IBM i stacks and Linux resales from Red Hat, Canonical, and SUSE Linux). That operating system revenue stream is very profitable with $385 million in gross profits. Overall, the Systems group had $891 million in gross profits, but pre-tax income was off 45 percent to $136 million.
Basically, as IBM reports its Systems group business, it looks like systems as we know them accounted for a very small portion of the $19.22 billion in revenues and $2.85 billion in net income for the quarter. But don’t get confused. IBM has a much, much larger systems business than it is letting on because it is trying to make it look like it has more cognitive, cloud, and other sales. Here is what we think IBM’s real systems business looks like compared to its overall revenues:
That total for systems includes a bunch of things other than the base operating systems, server, storage, and switching sales. It includes all of the transaction processing software sales that Big Blue has, plus most of its integration software sales (90 percent in our model), and the majority of financing sales (75 percent). If you do that, you get a breakdown of the server business that looks like this since the new classifications used for IBM’s financial reports were backcast to the first quarter of 2014:
Those 2014 revenues for systems included System x, BladeCenter, and Pure Systems revenues for products sold off to Lenovo, so the compares are not exactly apples-to-apples. It would be interesting to try to take out the X86 system sales in the old data to see how the Power Systems and System z lines have done, but what we know for sure is that they are down in 2016, and by a fairly large amount, after both lines grew in 2015. IBM clearly needs a new product cycle to rev up revenues in 2017, and hopefully it comes sooner rather than later. We shall see.
The irony is that most IBM i shops can do quite nice with a Power7+ or a Power8 system, and we would point out that this is actually a problem and not a good thing. Customers who are putting workloads on other platforms should be putting them on Power. If anything, IBM should be paying customers to unplug X86 servers and replace them with Power Systems iron doing the same work. A direct attack on the 90-ish percent of IBM i shops that have Windows Servers running a lot of workloads seems appropriate. But more on that later.