Taking At Stab At Modeling The Power Systems Business
January 28, 2019 Timothy Prickett Morgan
It is incredibly difficult to try to get a handle on how IBM’s overall systems business and then its Power Systems portion of that business is doing, something that I voiced frustration about last summer when talking about Big Blue’s financial results for the second quarter of 2018. At the time, I told you I would take a whack at trying to build a model of Power Systems sales to give us a sense of how the hardware platform is doing, and I took the first stab at this in the wake of IBM announcing its financial results for the fourth quarter of last year.
I didn’t intend to do this at this time, but I started to make a summary table of the growth and decline rates for Power Systems revenues for machines sold externally to customers and resellers – it was going to be a simple table, just something to spice up the coverage on the final quarter of last year. But as often happens with me, my curiosity took over and I just kept building the table out all the way back to 2010 with the growth and decline rates of Power Systems hardware sales each quarter both as reported and at constant currency (meaning exchange rate neutral year-on-year). Back in 2009 and 2010, IBM actually gave figures for both, but in 2015 the company only talked about growth rates for its products in constant currency. But you can work it backwards, sort of, by seeing the difference between IBM’s reported growth and constant currency growth overall and applying the difference to the constant currency Power Systems rates. If you do that, you get a chart that looks like this:
Can you tell – really tell – how the Power Systems business is doing if you just take the quarterly percent changes that IBM supplies and plot them out? Maybe you can, but I sure as heck can’t. I mean, I get a sense each quarter of how the Power Systems line is going – in constant currency, up 3 percent in Q1, up 4 percent in Q2, up 17 percent in Q3, and up 10 percent in Q4 is what IBM has told Wall Street in its last four quarterly reports. Up is always good. And I can sort of adjust that to get an as-reported growth rate (which IBM used to supply but has not since 2015); in some cases, currency exchange seems to help and in others, as with the past year, it hurts.
When you average it all out, from my model at least, it looks like the Power Systems hardware business – not including operating systems and integrated databases like the IBM i platform has bundled with it – grew by 8.8 percent in 2018 to just hair over $1.6 billion. ( I will explain how I got there in a second.) This is not too shabby. It’s not the 35-ish percent that the overall server market might grow in 2018 – the fourth quarter figures from IDC and Gartner won’t be out for several weeks, so we don’t know as yet – but it sure is better than the kicks in the teeth that Power hardware took in years gone by.
Still, I like real numbers. And in the absence of them, I feel compelled to estimate them. So if we have the constant currency growth rates each quarter for Power Systems from Big Blue itself, all we need is a starting point for revenues in any given year and then we can apply the estimated actual quarterly growth or decline rates (pulled from the overall currency effects that IBM as a whole experienced) to this data and cascade it forward. I looked at the RISC/Itanium Unix server revenue figures for 2010 from Gartner, took my best guess at how much of this was hardware and how much was operating system software, backed out the software, and then added in how much sales I thought were coming in from Linux (next to zero) and IBM i (about 15 percent of the overall sales, but maybe 50 percent of shipments, I reckon) in 2010 to get that starting point.
I don’t claim this is perfect, but it allows us to see the ebb and flow of revenues rather than just a bunch of percent changes. The shape of the data is mostly right (it depends on how closely currency effects on Power Systems sales hew to the overall currency effects on IBM’s entire product portfolio each quarter), but I believe this is a few points at best and the shape I agree that the height of the bars could raise or fall depending on those initial 2010 revenue assumptions. But the shape is not going to change that much.
So, if you do all of that math for the quarterly numbers all the way back to 2009, the belly of the Great Recession, then this is the kind of chart you get:
I tried to make sure that the revenues didn’t violate my sensibilities about how these sales compared to IBM System z and storage hardware sales. I think the chart above has the right shape, and probably close to the right magnitude of sales. I can adjust it as I get more data. Some of the fourth quarters were quite good, as you can see, even in 2009, 2010, and 2011, and that was because large ERP systems were largely insulated from the effects of the Great Recession until a few years later. You can sort of pick out the general trend, and it sure looks like a rebound is underway here in 2017 and 2018.
The magnitude of this uptick – and how far revenues of Power Systems have fallen off in recent years as Moore’s Law increases give people 2X the performance in each jump for the same revenue – becomes a bit more obvious if you annualize the data, like in the chart below, which is probably more fair given the lumpiness of sales of Power Systems and the less-than-regular timing of machine rollouts during some of these years. Here is the estimated annual sales of Power Systems iron based on my model:
For those of you who like to see the numbers and the growth rates, here is a table:
The last time that IBM seems to have seen annual growth in Power Systems was way back in 2011, and the declines in 2013 and 2016 were quite bad. Hopefully IBM can get more traction with Linux on Power and boost revenues even more than it has, and provide a big foundation for AIX and IBM i platforms to continue to run upon. That’s the bet.