Grid connection queues in Great Britain reached 125 GW by mid-2025, up from 41 GW in late 2024, against a national peak demand of around 45 GW.¹ Modern data centres can be deployed in 18 to 24 months, while grid connection upgrades typically take three to five years, a mismatch the International Data Center Authority calls a "delay tax" costing operators more than $10,000 per megawatt per day in lost revenue.² For operators who need capacity now, capacity already sitting inside their existing estate is worth more attention than it usually gets.
The scale of idle capacity
Research by Jonathan Koomey of Stanford University and Jon Taylor of Anthesis Group found that 30 percent of physical servers in their 2015 sample were comatose, delivering no useful computing over six months or more, representing more than $30 billion in idle capital worldwide.³ A larger follow-up sample put the physical server figure at 25 percent.⁴ The Uptime Institute has independently reported similar levels of unused capacity, and in one study of 20,000 servers found that shutting down comatose machines cut IT load by 5 MW and cooling and infrastructure load by a further 4 MW.⁵
IDCA's 2026 global energy report brings the same finding into the current capacity crunch: unused workloads that were never properly decommissioned account for more than 3 GW of US data centre capacity, up to 13 percent of the national total, and reclaiming it is a faster and cheaper route to usable capacity than building new.²
Idle equipment also carries a security cost. Koomey and Taylor note that a server doing no useful work is unlikely to be receiving current security patches, making unmanaged equipment an open door into the environment.⁶
Finding it and acting on it
Identifying which equipment is genuinely idle is a monitoring exercise. The Anthesis research relied on discovery tooling measuring processor, memory and network activity over extended periods, because utilisation figures alone don't confirm a server is doing nothing useful.³ That work sits with an operator's own monitoring, DCIM or FinOps function.
Acting on the findings is typically a physical activity. Equipment identified as idle has to come out of the rack under a documented chain of custody, be sanitised and certified to the required standard, and be routed to redeployment, resale or recycling on its merits rather than by default. The space it leaves behind then has to be reconfigured, cabling reworked, rack layouts adjusted, documentation brought back in line with what's actually on the floor, before the freed power and floor capacity is available for reallocation.
Estates that skip the second half end up with idle equipment removed on paper but capacity that nobody can confidently use, because the documentation no longer matches the room and the cabling still reflects a layout that changed two refreshes ago.
The commercial case
Recovered capacity has value on both sides. AOL's decommissioning programme reportedly saved the company $4.3 million in operating costs, reduced its carbon footprint by 35 percent, and generated a further $6 million from reselling the retired hardware.⁵ Grading retired equipment for redeployment or resale, rather than defaulting to recycling, can raise recovery value by around 40 percent.⁷
For operators facing connection timelines they don't control, capacity recovered from an existing estate arrives on their own schedule. Sunspeed supports that work at the physical layer: decommissioning and IT asset disposal with documented chain of custody, structured cabling reconfigured as the estate changes shape, and the on-site delivery to carry it out inside a live environment.
References
Data Center Frontier, "DCF Trends Summit: AI Compresses the Data Center Hardware Lifecycle and Raises the Stakes for ITAD," September 2026. https://www.datacenterfrontier.com/sponsored/podcast/55402193/dcf-trends-summit-ai-compresses-the-data-center-hardware-lifecycle-and-raises-the-stakes-for-itad
Ofgem, "Ofgem acts to free up grid capacity by tackling speculative data centre projects," 29 July 2026. https://www.ofgem.gov.uk/press-release/ofgem-acts-free-grid-capacity-tackling-speculative-data-centre-projects
Computer Weekly, "IDCA warns of 'delay tax' as AI datacentre power surges 50%," 15 July 2026. https://www.computerweekly.com/news/366645704/IDCA-warns-of-delay-tax-as-AI-datacentre-power-surges-50
Anthesis Group, "Zombie Servers: Hunting Down the Lost Capital," Koomey and Taylor. https://www.anthesisgroup.com/insights/zombie-servers-hunting-down-the-lost-capital/
Computerworld, "A third of virtual servers are zombies." https://www.computerworld.com/article/1680190/a-third-of-virtual-servers-are-zombies-2.html
Horizon Technology, "How To Attack Zombie Servers in the Data Center." https://horizontechnology.com/news/attacking-zombie-servers-in-the-data-center/
Koomey, J. and Taylor, J., "Zombie/Comatose Servers Redux." https://info.anthesisgroup.com/hubfs/Website%20PDFs/Comatose-Servers-Redux.pdf