The share shift is bigger than the spending headlines
Every quarter brings a new capex number from a hyperscaler earnings call, and those numbers get attention because they are large and easy to compare year over year. The more consequential figure from Synergy Research Group's latest report is a share number rather than a dollar figure: hyperscalers are on track to control 67 percent of all global data center capacity by 2031. That figure describes where the industry's physical infrastructure actually sits and how the market is structured, which is a more durable signal than any single quarter's spending total.
The other side of that figure is just as stark. On premises enterprise data center capacity, which held 56 percent of the global total as recently as 2018, is projected to fall to just 19 percent by 2031. That trajectory describes a structural handoff of the majority of the world's computing capacity from enterprise owned facilities to a small number of hyperscale operators in roughly one decade, a pace of change that most enterprise infrastructure planning cycles were not built to anticipate.
The scale numbers behind the share shift
Synergy's data shows hyperscalers operated 1,297 large data centers globally in the third quarter of 2025, triple the number they ran in 2018, and that count reached 1,360 by the end of the year. Roughly 60 percent of that hyperscale capacity sits in facilities the operators own outright, with the remainder leased, a mix that gives the largest cloud providers direct control over both the pace of expansion and the underlying cost structure in a way most enterprise IT buyers cannot replicate.
By 2031, Synergy projects hyperscalers will hold 14 times the data center capacity they controlled in 2018. Satya Nadella has put a number on Microsoft's own piece of that trajectory, stating the company will increase total AI capacity by more than 80 percent this year and roughly double its overall data center footprint within two years. Numbers at that scale are only possible for an operator with Microsoft's capital access and construction pipeline, which is precisely the point.
Why this validates cloud first, cautiously
For any CIO who has spent the last several years defending a cloud first roadmap against skeptics who wanted to keep workloads on premises, this data is useful ammunition. The structural trend lines up with the roadmap: capacity, innovation pace, and increasingly even raw compute availability for AI workloads are concentrating with the hyperscalers rather than dispersing back toward enterprise owned infrastructure. Betting against that trend at this point means betting against the physical direction of the entire industry, a bet that gets harder to justify to a board every year the share numbers move further in the hyperscalers' favor.
The caution is what comes with concentration. A market where 67 percent of global capacity sits with a small number of operators by 2031 is a market where those operators have correspondingly more pricing power, more influence over which workloads get prioritized during capacity constraints, and less competitive pressure to negotiate on terms that currently still favor large enterprise customers, a dynamic that tends to show up gradually in renewal cycles rather than all at once in a single contract.
The negotiating window is shrinking
Today, with on premises capacity still at roughly a third of global share and multiple hyperscalers competing hard for enterprise workloads, buyers still have real leverage: multi cloud threats are credible, repatriation is a believable negotiating tactic, and vendors compete on price and terms to win large accounts. That leverage depends entirely on there being a credible alternative to any single vendor, and that alternative gets less credible every year on premises share keeps shrinking toward the 19 percent Synergy projects for 2031, a number low enough that repatriation stops being a believable threat for most workload categories.
Enterprises that want to lock in favorable terms, whether on pricing, data portability, or exit rights, have a better negotiating position doing it now than they will in five years, when fewer buyers will have a realistic on premises fallback to threaten vendors with. This is the kind of structural trend that rewards acting before it fully plays out, since the terms available to a buyer with a credible alternative are consistently better than the terms available once that alternative has quietly stopped being credible.
Concentration risk belongs on the governance agenda
A 67 percent hyperscaler share by 2031 is also a vendor concentration risk that boards and risk committees should be tracking explicitly, not treating as an IT department detail buried in a quarterly technology update. When the majority of the industry's capacity sits with three or four operators, a capacity constraint, a price change, or a service disruption at any one of them has systemic implications that ripple well beyond any single customer's contract, closer to a supply chain concentration risk than a routine vendor management issue.
This argues for enterprises to treat exit planning, data portability clauses, and genuine multi cloud architecture as risk management investments rather than discretionary engineering projects. The cost of maintaining that optionality is real and should be budgeted honestly, but it is a cost that buys protection against a market structure that is becoming more concentrated every year this trend continues, with nothing in Synergy's data suggesting the trajectory is close to leveling off before 2031.
What belongs on the roadmap now
The structural case for cloud first was already strong before this report. This data makes it stronger, and it should accelerate any migration plan that was moving cautiously while leadership waited for more certainty about where the market was heading. The market has given CIOs that certainty now: it is heading toward a small number of hyperscale operators controlling two thirds of global capacity within five years, and planning around any other outcome is planning around a scenario the data no longer supports.
The parallel task is protecting negotiating leverage while it still exists. That means pushing for favorable multi year terms now, insisting on data portability commitments in every new contract, and keeping enough workload diversity across providers to make an exit threat credible. Waiting until on premises capacity has shrunk to a fraction of what it is today is waiting until the leverage that would have made these terms easy to get has already disappeared.
