A Rising Tide
December 7, 2015 Timothy Prickett Morgan
The system business is roaring along in 2015, and it looks like the industry might devour a historic number of machines this year, if the first three quarters are any indication. The amount of capacity that companies are consuming is staggering, and not something the analysts at IDC or Gartner talk about when they put out their quarterly figures. But we like to think these things through.
It looks like Power Systems sales are on the rise, too. So maybe, just maybe, a tide is rising that will lift all boats a little. (Provided interest rate increases or some other crisis doesn’t stall the global economy–always a worry.)
With the Power8-based machines being more or less competitive with Xeon iron, we would like to think that the rising tide in server sales is also lifting the Power Systems market, that the same pressures that are forcing companies to buy more–and more powerful–Xeon machines are also compelling companies to buy bigger and faster Power Systems iron to run a widening suite of applications.
This is clearly IBM‘s goal, with its efforts to create cost-effective Linux platforms based on Power8 platforms. But those efforts are really aimed at hyperscalers and HPC centers looking for leverage over Intel at the very least and often actual features, such as higher memory bandwidth, that their applications require and the Power processors can offer over Xeons in many cases. But for the most part, we are operating with very little knowledge of how well or poorly IBM’s strategy of reinvigorating the Power Systems platform is going. The way we are forced to look at it, if the System z and Power Systems lines together are holding their own in terms of market share, then IBM will remain committed to its platform business.
In the third quarter ended in September, Gartner says that companies consumed 2.76 million servers, an increase of 9.2 percent compared to the year ago period, and that this generated $13.5 billion in revenues, up 7.5 percent year-on-year. We don’t know precisely how many Power Systems machines IBM sold, but Gartner reckons that for Power-based machines running AIX, IBM had $613.5 million in sales, down 7.5 percent, but shipments were up 3 percent to 10,688 boxes. It is hard to guess how many Linux-based Power Systems machines Big Blue could have sold, or how many were sold that ran IBM i, but it is hard to imagine the number being more than twice that amount. (IBM doesn’t make the top five shipper cut anymore, so we can’t do the math to figure out how many machines IBM sold that were not based on X86 processors or were AIX on Power.)
We think that the trend is probably up for both IBM i and Linux on Power, by the way, but it would be nice if IBM actually told us so we would know for sure. Gartner did provide some clues, adding that for all of its RISC platforms, sales were only off 3.1 percent, which implies that the combination of IBM i and Linux on Power were sufficient to stop the revenue decline and, presumably, boost overall Power Systems shipments by more than is evident in the AIX-only figures. IBM’s System z mainframe sales were up 15 percent in the quarter, according to Gartner, and that meant overall, IBM’s server business rose by 5.1 percent.
What we are saying it that it is important to get beyond the hard, cold numbers that show IBM’s server revenues fell by 42.8 percent in the third quarter, to $1.33 billion, thanks to the sale of its System x division to Lenovo last October. It has taken a year to go through the transition, and there will still be some moderately annoying compares for the fourth quarter of 2015 and in the first quarter of 2016, when IBM was still selling System x gear, but soon we will be left with compares that do not include any System x iron.
IBM is interested in profits as much or more than revenues, and we have very little sense from the Gartner and IDC how much profits each vendor and type of system throw off each quarter. Based on our own analysis of IBM’s financial results for the third quarter, IBM reckons it had $1.49 billion in revenues in the third quarter with another $121 million in sales and posted a pre-tax loss of $24 million. Gartner is looking at the sales directly from IBM and through the channel, hence the difference in the numbers. To one way of thinking, the channel numbers are what matters, but IBM’s own numbers are the ones it uses to justify product lines and hence it would be good if the combination of Power Systems and System z hardware sales were profitable. And not because that matters, but because of the silly way IBM carves up its products into parts that it casts out to fiefdoms–Systems Group, Software Group, and Global Services–and then judges them individually instead of as a whole. As we pointed out back in late October, we think IBM had a $6.7 billion systems business in the third quarter, with servers, storage, and operating systems driving about $2.1 billion of that, Across this core systems business, including support and financing and other stuff, that $6.7 billion business has maybe 60 percent gross margins.
This is not a small or weak business. It is just one that IBM describes poorly.
That is not to say that it does not have its issues. IBM needs to crank up the volumes on Power-based system sales, and frankly, until the Power Systems LC machines were announced in the fall, IBM did not have the right boxes to go after Xeon-based machines. And IBM better have some good stuff coming with the Power8+ machines next year so it can keep pace with the “Broadwell” Xeon E5 processors coming in the spring from Intel. There is no resting in this market.
We all want a healthy and vibrant Power server business, but we specifically want the IBM i base to be engaged and growing their workloads on IBM i. There are hints that this is happening, but we would like to have more concrete figures to prove it. Such facts would help build some momentum and enthusiasm among the 125,000 IBM i faithful. If anything, IBM needs for Power Systems sales to outgrow the market at large for an extended period of time. Its OpenPower Foundation efforts will help, if some big hyperscalers and large enterprises in Asia adopt local processors and systems, and the trick will be to bring together the stats for the whole OpenPower community to show its growth. IBM needs to make this happen because IDC and Gartner will not provide such data publicly unless someone pays for it.
As we have previously reported, Tom Rosamilia, who runs IBM’s Systems Group, said that in the hyperscale market IBM had a goal of achieving something on the order of a 10 percent to 20 percent share, and similarly for the high performance computing and supercomputing space. He did not say when, so we have to make out own guesses there. So let’s do that.
The hyperscalers consume about 20 percent of shipments these days according to Gartner, and overall server shipments will probably break through 10 million this year, and assuming 5 percent shipment growth per year, server shipments will be around 12.7 million units in 2020. Ah, but the share of servers acquired by hyperscalers keeps growing, and it will probably be closer to 40 percent by 2020 (some say as high as 50 percent), and that means IBM and its OpenPower partners will have to sell between 500,000 and 1 million servers to meet that 10 percent to 20 percent share goal. There could be another 100,000 to 200,000 nodes for the HPC shops by 2020, and let’s be generous and say another 200,000 machines for enterprises.
Add that up, and we have somewhere between 800,000 and 1.4 million machines running a Power chip some years hence, if Rosamilia and the OpenPower partners have their way. That would represent about a 6 to 11 percent overall share of the worldwide server shipments, and assuming revenues stay flat for the market at large (as they have for the past five years), average selling prices keep trending down, and a Power machine costs about the same as a Xeon machine, that is somewhere between $3.3 billion and $6 billion in server revenues as Gartner reckons them. And all of that revenue will not come to Big Blue–in fact, maybe only half of it does and the rest is garnered by its OpenPower allies.
What that means is that IBM will have, through its partnerships and its own efforts, increased its Power server volumes by something like 20X to 35X, but the revenue stream will only increase by 30 percent at the low end of the shipment range to a high of 2.4X at the high end of the shipment range. This may not be a business that anyone would want to get into, but it is the one that IBM is in.
The rising tide requires us all to paddle faster, Big Blue included. It beats drowning. And we think, as IBM must, that this increased Power Systems footprint will drive more sales of software and services, and that its overall systems business will grow and be more profitable. For all we know, it could grow enough to fill in the System x gap, mostly due to Linux. And AIX and IBM i will benefit from that, both directly and indirectly.