IBM’s Own Positioning Of Power Systems Revealed
September 24, 2018 Timothy Prickett Morgan
We don’t get a lot of data about how the Power Systems business is doing out of Big Blue these days, and we get even less, in terms of specifics, about the markets that IBM is chasing and how well or poorly it thinks it is doing in this regard. We get a snippet here and there from the quarterly results, we get some insight from the quarterly market trackers from IDC and Gartner, but not much else.
Digging around through some documents relating to the Power9 rollout, we stumbled across a few interesting charts that gives us at least some insight into how IBM sees the systems world and how it is chasing these markets with particular parts of the Power Systems line and related products such as the collection of Spectrum Storage software that is based on Platform Computing’s LSF and other workload management tools and the General Parallel File System (GPFS) that has been given this new brand, as well as the legacy IBM i and AIX platforms running databases and other enterprise applications.
Here is how IBM reckons the total addressable markets stood as 2017 was finishing and looking ahead at growth rates between 2017 and 2019, inclusive. Take a look:
In this presentation, IBM says that its goal is to increase its market share across every one of those five key segments, although this chart does not actually show the market share that IBM holds in each of these areas. Let’s walk through these categories. The total addressable market, or TAM for short in marketing speak, for core back-end systems running databases and applications – minus mainframes and X86 systems – is $4.5 billion, according to IBM’s research. The compound annual growth rate from 2017 through 2019 is actually a decrease of 3 percent per year, so this is not a growing market. We will remind you that the overall midrange, as defined by IDC, is growing at the moment – something we talked about earlier this month – and it certainly includes X86 iron and is dominated by X86 iron it terms of shipments and rivals RISC iron in terms of revenues. So this TAM is a little weird, I think. But here’s the neat bit. In this way of looking at the market. AIX probably has around 50 percent market share by revenues of the RISC/Unix market and IBM i has 100 percent market share of the IBM i market, so IBM therefore has probably somewhere around 60 percent of the overall TAM, or about $2.7 billion in revenues. IBM had $6.5 billion in system hardware revenues in 2017 (including storage) and another $1.7 billion in operating system sales, so call it around $6 billion in overall revenues once storage is pulled back out. This core enterprise Power systems market therefore is, we guesstimate, about 45 percent of IBM’s overall system hardware revenues; AIX and IBM i have a much smaller proportion of that operating systems revenues, because this is where System z rings up the dough.
The enterprise Linux server market, which is the middle column has a $2.8 billion TAM according to IBM’s research, and is growing at a 10 percent CAGR between 2017 and 2019. Linux, as we have pointed out before, is the only operating system in the datacenter that is showing revenue and shipment growth; all other operating systems, including Microsoft’s Windows Server, are declining even as the system market is exploding. The buildout of public clouds and hyperscale infrastructure that provides services such as those provided by Google, Facebook, Alibaba, and others are helping to drive Linux, and open source software applications for analytics and machine learning are helping to push Linux, too. But this column is just about traditional infrastructure jobs such as web serving and application serving and, oddly enough, the SAP HANA in-memory database. It is hard to say what share Power Systems have of this $2.8 billion TAM, but it is probably well under 5 percent and maybe only 1 or 2 percent share is for generic Linux infrastructure while IBM could have as much as a fifth or a quarter of the SAP HANA market, which is a lot smaller than the generic Linux market.
The column to the right of this is called HSDC, which is short for hyperscale datacenter and which includes machines sold into managed service providers (MSPs) and cloud service providers (CSPs), which IBM pegged at a $2.1 billion TAM in 2017 with a 19 percent CAGR. For all we know, Google and Rackspace Hosting have deployed lots of Power9 machines in here, but that revenue would be noise in all of the X86 sales data here. So the TAM is juicy but the share is currently low.
That leaves high performance computing, high performance data analytics, machine learning and other adjacent modern database workloads that tend to come out of hyperscalers and cloud builders as open source tools and are being adopted This cognitive infrastructure market is pegged with a TAM of $6.4 billion. After selling off the System x business to Lenovo, Big Blue gave up a big piece of this TAM with the hopes of rebuilding from not-quite-scratch on Power iron. This market is growing at 20 percent a year at CAGR, and there is every chance IBM could attack this business profitably because of its own software stack and the considerable performance advantages that Power Systems have for these workloads.
That leaves the far right column in the chart, which is the custom HPC system part of the market, for real high capacity systems such as the “Summit” supercomputer at Oak Ridge National Laboratory (that’s the CORAL procurement reference) and the “Sierra” supercomputer at Lawrence Livermore National Laboratory (that’s the APEX procurement reference). IBM reckons that this custom HPC system market had a $3.7 billion TAM in 2017 and is growing at a meager 3 percent. Big Blue probably has a decent share of this market, but there are a lot of vendors in a lot of different nations that also have their nationalistic shares. Hyperion Research reckons that IBM had $575 million in HPC server revenues in 2017, and a lot of this was in this HPC custom category. Call it 50-50 share regular HPC and custom HPC, and either way, IBM’s share is probably somewhere between 5 percent and 10 percent in these segments.
Across all of these segments, you can see how Power Systems averages only a few points per share of the overall systems market – put a gun to my head, and I would say somewhere between 3 percent and 5 percent, including systems software. (There is so much hyperscale iron running homegrown Linux that it brings the system average down, unless you counted the cost of all the many tens of thousands of software engineers who create this software. It would be fascinating to figure out if such “free” software is more expensive than commercial software with perpetual licenses and support or annual subscriptions.)
Given these market segments, this is how IBM carves up the software partnerships for each segment:
Knowing what you now know about the “ZZ” Power9 entry systems announced back in February, the “Boston” entry machines announced in May, and the midrange “Zeppelin” and high-end “Fleetwood” machine announced in August, you can see what partners of the market that IBM is trying hit with specific classes of machines and what partners, if any, it is leveraging to try to do it.
The #FromHereToAI hashtag is interesting in the chart above, starting with the “Maintaining the legacy” column for AIX and IBM i and presumably showing how companies will evolve to use SAP HANA in-memory databases, open source databases and datastores and batch analytics and eventually deploy machine learning and analytics. There are probably not many IBM i shops who are on this particular journey as described. Not to be cynical. But like IBM, we believe that companies that do add systems of engagement with systems of record should build on a Power Systems base and there be consistent across these platforms. It probably could be done on a Power E950 midrange box alone with lots of partitions, if Big Blue supported IBM i on it, that is. IBM could help a lot by stop making AIX and IBM i second class citizens on its Power Systems line. Every machine can – and should – run these two operating systems, either bare metal or virtualized. Let customers decide what hardware they want. They just might decide to buy more than IBM is banking on, if Big Blue just loosened up a little and got creative.