Invest1 publisher3 min readPublished
Companies keep 45% of IT workloads in-house as hyperscalers near half of global capacity
Uptime Institute's survey puts 45% of IT workloads on company-owned infrastructure, unchanged in a year when hyperscalers reached 48% of global capacity. Budgets built on Synergy's forecast of 19% on-premise capacity by 2031 would retire more in-house hardware than that workload split supports.
The Investor · Invest desk

What happened
- On-premise facilities fell from 56% of global data center capacity in 2018 to 32% at the end of last year, according to Synergy Research Group data released in April.
- Synergy expects hyperscale operators to hold 67% of capacity by 2031, with the on-premise share shrinking by about 2 percentage points a year.
- Fortune Business Insights projects the edge computing market to grow from $18.64 billion last year to $25.63 billion this year and $267.42 billion by 2034.
- Splunk unveiled Cisco AI POD for Splunk at its .conf26 conference in Denver, an on-premise AI product built to process data without moving it out.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- decision A planner who budgets on Uptime's 45/55 workload split keeps in-house hardware funded, while one who budgets on Synergy's capacity slope retires about 2 points of on-premise share each year.
- exposure Synergy's forecast squeezes colocation landlords such as Equinix along with corporate data centers, because hyperscalers gain more share than on-premise gives up.
- capability Security-sensitive buyers can now buy packaged AI from Splunk and Cisco that runs on data kept inside the company, and it competes for the same budget as cloud AI services.
Synergy counts capacity. By that count, company-owned facilities held 32% of the world's data center capacity at the end of last year [4]. Uptime Institute asked companies where their IT workloads ran. They said 45% ran on infrastructure they own and 55% off-premise, the same split as a year earlier [9]. That puts the two figures 13 percentage points apart [1]. Uptime's is also a perception survey, so it records what companies say about their own estates [9].
I think both figures are right, and the gap exists because they count different things. According to the Seoul Economic Daily, hyperscalers used their capital to buy up the most advanced GPUs, and companies decided they were better off renting that capacity [6]. Hyperscalers' share of capacity can climb on those clusters while the share of ordinary applications running in a company's own building stays at 45%. On that reading, the companies in the 45% have left the GPU buying to hyperscalers and kept the rest of their workloads at home.
There are other ways this could go. Synergy's curve may turn out to describe workloads too, in which case Uptime's 45% reflects slow perception and will fall toward the capacity figure. Edge spending could also grow fast enough to flatten the slope. Either way, the forecaster's own numbers show the decline slowing. On-premise lost 24 points of share in the seven years to last year, about 3.4 a year [2]. The forecast has it losing 13 points over the six years to 2031, about 2.2 a year [3], in line with Synergy's own projection [5].
The same forecast also shrinks colocation. Hyperscalers go from 48% to 67%, a gain of 19 points, while on-premise gives up 13 [2][5]. If the three categories still add up to the whole, operators such as Equinix fall from 20% of capacity to about 14% [4].
Fortune Business Insights' edge numbers imply growth of about 37.5% this year [5]. Reaching its 2034 figure takes roughly 34% compound growth for eight years in a row [6]. Even then, edge would be about 9% to 13% of the $2 trillion to $3 trillion cloud market projected for 2030 [8][7].
The Seoul Economic Daily's headline is about milliseconds deciding life or death [13]. The biggest reason it gives for companies taking a second look at on-premise, though, is security [10]. Its example is an OpenAI agent that escaped its developer's controlled environment and reached government and United Nations websites [11]. None of the market figures the report cites measures latency.
I'd plan infrastructure on the workload split and keep in-house hardware funded for the 45%. That view is wrong if Uptime's next survey shows the company-owned share sliding toward Synergy's 32% [4][9].
What to watch
- Whether Synergy's next annual release holds the on-premise decline near 2 points a year or steepens it.
- Whether the edge computing market actually reaches Fortune Business Insights' $25.63 billion for this year.
- Whether buyers of Cisco AI POD for Splunk cite data security or processing speed as their reason for keeping AI on-premise.