Datacenter capital expenditure is forecast to grow 17 percent annually through 2030, reaching $1.6 trillion, with supply chain constraints pushing up the price of components.
The latest Cloud and Datacenter Market Snapshot [PDF] from analyst Omdia states that investment into AI infrastructure continues apace, despite all the talk of it being a bubble ready to burst.
However, adoption of AI remains relatively low, it claims, with many more users and higher usage per user expected. At the same time, the models are getting larger and using more compute for inference.
All of this translates to higher performance infrastructure that consumes more power, leading to rising server, rack, and datacenter power density - and even more spending on AI infrastructure.
So much so that Omdia expects datacenter capex to grow at a compound annual growth rate (CAGR) of 17 percent over the next several years, touching $1.6 trillion by 2030.
Whether this is sustainable is a good question. Management consultant Bain & Company calculated that the industry will have to make $2 trillion in annual sales by 2030 to afford the forecast level of investment, yet returns on all this spending are still somewhat elusive for vendors and users.
Various senior tech industry execs claimed last week that AI isn't a bubble, and dismissed comparisons with the dot.com bust.
Omdia considered four scenarios in coming to its own conclusion. The first takes into account the actual order pipeline and demand, balancing it against constraints such as Nvidia's order backlog for GPUs, and that datacenter construction is lagging behind deal announcements. This is considered the most likely outcome.
The second scenario is based on the possibility that constraints have a lesser impact in the short term, producing an accelerated pace of development that results in failure for some developers.
A third, dubbed "the bubble scenario," assumes that a failure to realize productivity gains from AI use sees investors getting spooked after a few years, leading to a fall-off in capex after 2026 - though still hardly the crash that many fear.
The fourth, or "Nvidia" scenario, is based on the GPU giant's predictions, which eschew any constraints and forecast investment continuing unabated to hit $2 trillion by 2028.
However it isn't just AI servers that are pushing up the datacenter spend. A refresh cycle covering general-purpose servers is also boosting server shipments. Omdia had previously found that buyers (chiefly hyperscalers) were holding off replacing their standard servers to prioritize investment in high-spec systems to handle AI.
Omdia has now raised its investment forecast across all market segments, including the AI neo-cloud operators, Tier 1 and Tier 2 colocation providers, enterprises, and the hyperscalers.
Supply chain constraints are leading to higher costs of some commodity components, notably memory, as The Register reported last week, which channel sources told us is likely to result in server prices increasing by 15 percent.
Omdia says new datacenters will likely be engineered differently from existing facilities because AI demand is rapidly changing the IT infrastructure inside them. This extends from the silicon in use to the servers and racks, the thermal management systems, and the power distribution and backup systems.
Datacenter and IT teams need to avoid complacency and keep an open mind about new developments, it warns. ®
Gotta pay for those datacenter buildouts somehow
Does that kind of time saving actually pay for itself?
Microsoft claims it's a Secure Launch bug
Newer kernel, newer Cinnamon, new tools, and even new icons
Analyst: We'll hit a spot where 'we go from that was a great idea to where's my revenue?'
Check Point observes 40K+ attack attempts in our hours, with government organizations under fire
Opinion For trivial projects, it's fine. For serious work, forget about it