It’s Harder To Hear The Pulse In The Server Market
March 22, 2021 Timothy Prickett Morgan
More than any other piece of equipment that does into the datacenter, the server is an indicator of health and wealth. Over the more than three decades that The Four Hundred has been published, we have spent a lot of effort and time to understand how the world is investing in what kinds of servers, including Big Blue’s midrange systems running OS/400 and IBM i, and how the trends change over time. And we are committed to doing that going forward, even though it has just gotten a little bit more difficult.
For the past several decades – I honestly can’t remember how long – the box counters at both Gartner and IDC have been in competition with each other delivering market data, and this competition compelled them to provide a good bit of data about server, storage, and networking spending as a kind of loss leader for the very detailed market models they have. I have been personally grateful for the insight that this publicly available data has provided, and in recent years have concentrated more on the information that IDC put out in these core datacenter markets because of its richness and thoroughness.
So it came as something of a shock when the usual detailed shipment and revenue data, by the top original equipment manufacturers (OEMs) and the collective of original design manufacturers (or ODMs), was not made available when IDC put out its figures for the fourth quarter of 2020. Rather than the detailed tables we have come to expect, IDC has put out some commentary and two graphics showing market shares of vendors by revenues and shipments. Not to be deterred, we have worked from these graphics to do some estimating – you can count pixels in the images to get relatively precise percentages and apply these to the revenue and shipment figures. But the level of precision from estimates is by necessity a lot lower than the actual figures that IDC used to put out, which we presume are derived from market checks and data from vendors themselves before they are published. (This is why there is nearly a three-month lag between the end of a financial quarter and when the IDC and Gartner sales and shipment figures are divulged.)
While this is disappointing, we get it. IDC’s founder, Pat McGovern, died in 2014 and the company, which included the IDG publishing giant as well as the IDC consulting business, was sold to China Oceanwide Holdings Group in 2017, which itself is part of a wide web of investment vehicles centered in Beijing that, oddly enough, is cross coupled with the Chinese Academy of Sciences and its Legend Holdings, which of course is the owner of Sino-American server maker Lenovo. It would take a week to sort out all of the links between the owners and controllers of IDG and IDC, and that is not the point. What is the point is that the writing was on the wall once McGovern died and the company inevitably was sold by his heirs. Both Gartner and IDC have been tightening up their release of information to the public, and no doubt because building and maintaining these market models is difficult and expensive and, thanks to fewer suppliers and fewer buyers of systems due to the dominance of the hyperscalers and cloud builders, that cost is necessarily spread across fewer potential paying customers who need much richer datasets than this publicly available data to make their decisions.
Going forward, we will do our best with the information that IDC does provide to give you what insight that we can. We will clearly delineate data that is actually provided by IDC and that which we estimate based on “data” such as the relatively rough but still useful data embodied in charts.
According to the statement that IDC put out this month, sales of servers rose by 1.5 percent to $25.8 billion, but shipments declined by 3 percent to just under 3.3 million units. (We reckon it is around 3.293 million machines shipped in Q4 2020, against 3.395 million units shipped in Q4 2019, but IDC is not giving that kind of precision anymore.)
Here is the revenue share chart IDC put out:
As you can see, Inspur (which includes sales of Power-based gear in China as part of the Inspur Power Systems joint venture that this Chinese server maker has with Big Blue) grew by a little less than IBM itself shrunk, and Hewlett Packard Enterprise and Dell both shrank a tiny but as Lenovo stayed flat and Huawei Technology rose and the ODM collective was flat. The rest of the market, which we calculate was worth 16.1 percent of the market based on pixel counts from this chart, rose by a bit, accounting for around $4.15 billion of the $25.8 billion. HPE came in at $4.09 billion by our math, and Dell was $3.97 billion. Inspur, we think, rose by 24.8 percent to $2.17 billion, and IBM fell by 20.5 percent to $1.88 billion. Huawei was up 17 percent to $1.54 billion and Lenovo grew along with the market at 1.5 percent to $1.45 billion, as did the ODMs, who hit $6.55 billion. We think, based on our estimates from what IDC said in its companion server shipment char, that the ODMs shipped 29 percent of servers to drive those revenues, or around 955,000 machines. That’s actually a 9.5 percent decline in machines, which tells us the hyperscalers and cloud builders are buying beefier boxes, which the math shows they are. But they buy so many basic infrastructure servers that they have to buy a lot of pretty expensive, GPU and flash laden machines to move that needle.
Here is a chart that lays out the OEM and ODM landscape since the Great Recession:
Our precision in our estimates of IDC’s data us only as good as a pixel size, of course. We make do.
IBM, as you can see, has pretty much leveled off at somewhere around an average of $1.3 billion a quarter in the past two years. Because Cisco Systems was knocked out of the top five vendors five quarters ago by Inspur and Huawei – Oracle/Sun Microsystems has long since been banished to the Others category – we have been estimating its revenues based on its own financial reports and prior IDC data. Again, we make do.
IDC said in its statement that revenues from non-X86 servers, such as IBM’s Power Systems and System z mainframes but also including a growing Arm server category (driven now mostly by Fujitsu’s supercomputing business. Amazon Web Services and its Graviton2 processor for its own cloud, and Ampere Computing’s emerging Altra line), and a few legacy Itanium and RISC platforms, hit around $2.8 billion, a decline of around 9 percent and X86 servers hit around $23.1 billion, an increase of 2.9 percent. (Based on what we think happened a year ago in the IDC data, we think the numbers are closer to $2.75 billion in non-X86 sales and $23.05 billion in X86 sales, for whatever that is worth.) Here is the trend of X86 versus non-X86 since they separately from each other (that is not an accident) during the Great Recession:
There is a kind of détente there, and as Microsoft and Amazon embrace more and more Arm computing, it will fill in the gap in declining RISC/Unix and mainframe sales, we think, and quite possibly bend that curve up.
Here is another interesting chart we pulled out of the historical IDC data we have kept track of for a long time, plotting sales of all types of IBM servers against sales of all types of non-X86 servers:
IBM, as you can see, has been the signal driver outside of the X86 market for a long, long time. (This data only goes back to the middle of 2002, when IDC first started talking about the market this way.) But, as we point out above, as Arm servers rise in the datacenter – and we think they will – this could change. As we have pointed out before, anything that weakens the case for X86 strengthens the case for Power, but if Arm servers get beefier and cheaper, it can also weaken the case for Power. We shall see how this all plays out, and count on us to try to figure that out for you.