Spending On Legacy Systems Stalls In Q1, 2023 Forecast Looks Weak
July 12, 2023 Timothy Prickett Morgan
Neither Gartner nor IDC put out their quarterly server and storage reports for the public anymore, but IDC does still release a converged dataset that adds server and storage spending together and carves it up based on if it is spend on the clouds or for traditional – what we would call legacy – systems. The latest results are out for the first quarter of 2023, and there is what looks like a temporary stall in legacy system spending and it looks like it is going to get worse here in 2023 but return to growth in the coming years.
Overall, the market for servers and storage collectively is growing nicely, and among the cloud builders and hyperscalers it looks like it is booming. According to IDC, overall datacenter server and storage revenues were up 8.2 percent to $35.3 billion.
Spending on infrastructure for server and storage capacity to build shared clouds – what many call public clouds, but that is imprecise these days – rose by 22.5 percent to $15.7 billion. Spending for infrastructure for dedicated cloud services – be they outposts sold under a cloud subscription pricing model for installation in your datacenter or in a co-location facility, or hosting by clouds or traditional hosting service providers – was down 1.5 percent to $5.8 billion. Add it up, and server and storage revenues in Q1 aimed at all cloud use cases came to $21.5 billion, up 14.9 percent.
Interestingly, the non-cloud segment of server and storage sales worldwide – and that does not mean that the systems are not virtualized, but rather that they are not sold by capacity in a rental fashion after the servers and storage are installed – was off nine-tenths of a point to $13.8 billion. Frankly, considering the skittishness of enterprise, government, and academia customers because of the uncertainties in the national economies, that is not a particularly bad decline.
What is a bit more disconcerting, however, are the projections that IDC is making for growth between 2022 and 2023, which is shown below, as well as a forecast that runs between 2022 and 2027, inclusive, and shows the compound annual growth rate. This provides some optimism here for non-cloud, legacy platforms – but don’t get too excited.
As usual, estimates that we are making in the table above are shown in bold, red, italics.
IDC is forecasting that sales of non-cloud server and storage machinery will decline by 6.3 percent to $60.4 billion in 2023, and that is against an overall server and storage market that will rise by only 1.1 percent to $156.8 billion in sales this year. Server and storage gear sold into dedicated cloud and shared cloud environments, by contrast, will grow by 7.3 percent to $96.4 billion. And if you look at the bottom of the table above, you will see that server and storage spending by enterprises, government, and academia is forecast to shrink by 5.1 percent this year to $62.3 billion. So there is a definitely correlation between EG&A purchases and non-cloud purchases of compute and storage capacity.
For whatever it is worth, IDC has a rosier forecast out past 2024. By the end of 2027, the market researcher expects for EG&A server and storage spending to reach $78.9 billion, which represents a CAGR of 3.8 percent between 2022 and 2027. Service providers in their many forms will account for $142.8 billion in sales, with a 10.6 percent CAGR over the forecast period.
The interesting thing about Q1 is how AI and inflation are taking turns driving the market growth for servers and storage, and this is happening particularly in the cloud. This is evidenced by that $14.9 percent increase in spending for servers and storage for cloud uses, to $21.5 billion, mentioned above. But unit shipments of servers and storage to clouds fell by 11.4 percent and average selling prices were up 29.7 percent. Machines laden with GPUs, main memory, flash, and network interfaces – all required for massive AI training workloads – are pushing up those average selling prices. A typical machine with eight GPUs and lots of memory, flash, and network interconnect to drive those GPUs can cost between $350,000 to $400,000 without any trouble at all. That is big bucks for a node – even for IBM i and IBM System z shops.