The Power Neine Conundrum
July 24, 2017 Timothy Prickett Morgan
If you take a very liberal interpretation of what the term cognitive systems means, including database, middleware, analytics software and the underlying system hardware and software, then IBM has spent untold tens of billions of dollars – probably hundreds of billions, really – creating its Cognitive Systems stack. We wonder what all of that analytics and machine learning software would say, with Watson’s voice of course, if it was pointed at IBM’s entire financial and technical history.
What is the prognosis, Doctor Watson?
We here at IT Jungle are an optimistic lot, and we realize that IBM, like many of the key IT suppliers, is undergoing what seems like a nearly continuous series of transformations and reorganizations because the IT world is changing at an ever-increasing pace. But it is hard to watch Big Blue turn in 21 straight quarters of revenue declines while at the same time the Power-based server business has, thanks to Moore’s Law performance increases and the shift from RISC/Unix architectures to Linux and Windows Server running on X86 iron, shrunk. We are not sure how much it has shrunk because IBM doesn’t actually provide revenue figures as it should, but the Power Systems business for many years generated well north of $5 billion a year in sales for core servers and operating systems. We think it might be half that now, and maybe we are being generous there.
Since its financial near-death experience back in the early 1990s, IBM has been obsessed by showing earnings per share growth because Wall Street could be wooed by this metric and the compensation for its managers was pegged to this metric. It has been a meaningless measure for years, and now IBM is just trying to figure out how to reconfigure itself for a modern IT world in such a way that it can find sustainable net income with whatever revenues it takes.
This is not the International Business Machines from days gone by, which created markets as much as following them.
IBM has stomached 21 straight quarters of revenue declines, and we suspect that its many business partners are following down the same track. The second quarter of 2017 was nothing to write home about, with revenues down 4.7 percent to $19.29 billion and net income off 6.9 percent to $2.33 billion. IBM used to pride itself, during the height of the “financial engineering” says of former CEOs Lou Gerstner and Sam Palmisano, that it could grow revenues in the high single digits and grow earnings per share in the low single digits year on year. Net income did not always grow that fast, but IBM bought up its own shares on Wall Street to pump up those EPS figures. This, we have always thought, was a perfectly stupid and wasteful use of over $100 billion in cash.
IBM could have done just about anything with that money, including really building a modern, enterprise computing platform that would rival anything that Google, Amazon, or Microsoft could create. But IBM has taken the easy way out, tending a very profitable but ever-shrinking System z mainframe business, buying up application software that none one ever talks about, and trying to whip up enthusiasm for the Power platform through the OpenPower alliance and hoping some hyperscalers would hop on its bandwagon. It was a valiant effort, OpenPower, but as far as we know, Google and Rackspace Hosting have not committed to actually deploying Power9 systems at scale in their datacenters, although they did design systems just in case Intel’s “Skylake” Xeons and their follow-on “Cannonlake” Xeons did not make it to the field in a timely fashion and with the right feeds and speeds. Like everyone else, we expected that the Power9 chip would be out about now, taking on Skylake directly, but as we previously reported, with the exception of the “Summit” and “Sierra” supercomputers being built for the US Department of Energy, IBM is not expected to get machines with Power9 chips that can run IBM i or AIX out the door until early 2018.
The availability of Skylake Xeons from Intel plus X86 server alternatives called Epyc processors from AMD make it a lot harder for the Power9 chips to find their place in the market. IBM needed to beat these two processors to the field, not follow them, and we were absolutely lead to believe that the Power9 chips would ship in 2017 for enterprise customers and this is absolutely a delay in their rolling out even if, technically speaking, by shipping the Power9 in the “Witherspoon” systems that comprise the massively parallel Summit and Sierra machines IBM can say that it made its promise to launch the Power9 in the second half of 2017.
It could be worse, mind you. IBM chief financial officer Martin Schroeter said in a conference call with Wall Street analysts that declines in Power server sales “moderated” and that the declines in mainframe revenues were “consistent with what you would expect at the end of the cycle.” Ditto for Power8 systems, we think, but Schroeter did not say that. IBM has just launched the System Z – note the capital letter, this is the new name for the z14 mainframe. (We will be covering the new System Z and its z14 processor separately, trying to glean some insight from it in how IBM might further developer its Power servers and its Cognitive Systems approach.)
Schroeter said that in the Power Systems line running AIX, high end systems showed revenue growth but that the volumes are driven by the low-end and midrange products; this has been true for a very long time, and is also the pattern with IBM i although Big Blue never mentions IBM i in these Wall Street calls anymore.
“So, to summarize systems, our year-to-year revenue and gross margin trajectory improved with consecutive quarters of revenue growth in storage offset by the expected cycle driven declines in Z systems and Power,” Schroeter explained. “While we face some shifting market dynamics and product transitions, our portfolio remains optimized for cognitive and cloud, and we continue to expand our footprint and add new capabilities. Our recent z14 launch and the upcoming Power refresh beginning later in the year will drive further improvements in second half performance.”
We certainly hope so. The numbers don’t look so good, if you drill down into them:
Every category of every division of IBM declined in the second quarter, and if you do the math on the trailing 12 months as we did in the table above, IBM has shrunk to a $78.44 billion company and is only bringing 14.6 percent of that to the bottom line. This is pretty good net income for a lot of companies, but it is not the level that IBM has historically enjoyed. (We think a company that can bring 5 percent to the bottom line is barely successful, and that 10 percent is pretty average and 15 is pretty above average. It takes an IBM from the 1970s or a Google from the 2000s to deliver 25 percent of revenues as net income.)
As IBM records revenues in its Systems group, revenues were $1.75 billion, down 10.4 percent. Mainframe shops tend to not buy new machines at the tail end of a cycle, but do activate processors in existing machines, which delivers revenues at a very high profit margin because there is no manufacturing or support associated with the sale. (Why IBM doesn’t make all of its servers this way and provide true differentiation in the market is beyond me. Capacity on demand is a good idea for small customers, too.) That hardware figure includes System z and Power servers, whatever networking IBM resells, plus storage, and out guess is that the server portion might be a little north of $900 million of that, with another $292 million coming from operating systems if we teased apart IBM’s figures correctly. That $900 million is probably split halfsies between Z and P, but that is just a guess.
As we have pointed out many times before, IBM’s “real” systems business is much larger than these numbers imply, because all of that middleware, database, support, and other services like financing that gets added on. Here is what we think the underlying, core systems business looks like compared to IBM’s overall sales:
And here is how we think the various parts of the real systems business have seen their revenues change since IBM updated its financial reporting categories in early 2015 and backcast them into 2014:
The question we have is this, Doctor Watson: Have these curves found a local minima that will uptick and find a new, higher, and sustainable level in 2018 and beyond for the Z and P cycles to start all over again, or will this business just flatten out and stay there, or will it just continue to decline as more and more of the industry shifts to distributed systems running on X86 iron?
Damned if we can tell. But we sure hope it starts to grow again, and maybe IBM should focus a bit more on this and work once again to be a vendor of systems, complete and modern systems. The future is P, it sure isn’t Z, and if anything it might be X or A as well. IBM better calculate this equation and get it right. Power9 is late compared to other alternatives, and this is a serious problem. IBM needs to fix this if it hopes to build a server ecosystem that can take on the Xeon hegemony in the datacenter. Either that or just buy Nvidia and AMD, build X86 iron that can run AIX and IBM i, and get on with it.
RELATED STORIES
IBM Commits To Power9 Upgrades For Big Power Systems Shops
IBM i And AIX Won’t Get Power9 Until 2018
Say Sayonara To IBM i 7.1 Next Spring
Clearing The Decks Ahead Of The Power9 Launch
Negotiating The Upgrade Paths To Power8 Enterprise Systems
IBM Rolls Out The Big Power8 Iron
IBM Carves Out Upgrade Paths To New Power7+ Systems
Power7+ Launches In Multi-Chassis Power 770+ And 780+ Systems
Would Oracle buy IBM? Please don’t say we might be owned by China. On the tech front the brains at IBM bet on Java when they should have gone with javascript. And DB2 still does not know JSON. DB2 would be much improved if SQL procedures, functions and triggers could be written in javascript.
IBM power9 are expected for this quarter (or next one at most). Actually, openpower partners already have them in their hands. I do not think that is late. It is a great alternative to intel. A very interesting year for server architectures.