Q&A with Iron Mountain
What’s the biggest change you’ve seen in how organisations use cloud and digital infrastructure over the past year, and what’s driving it?
Over the past year, the industry reached a definitive turning point: organisations moved out of the AI honeymoon phase and collided with operational reality.
Eighteen months ago, the mandate was experimentation; teams set up sandboxes, ran proof-of-concepts (PoCs), and built prototypes that looked incredible on paper. But the moment business leaders tried to scale those models into live production, progress ground to a halt. We call this the AI Activation Gap. Enterprises realised their compute and cloud environments weren't actually the primary bottleneck; their data architecture was.
What is driving that shift today is the reality of compute demand. Training models grabbed all the early headlines, but day-to-day operational queries; inference workloads are now projected to consume up to 80% of an enterprise’s total AI compute capacity. When you run models continuously against live workflows, centralising every petabyte in a single public cloud is no longer practical. Moving massive datasets triggers eye-watering egress costs and latency penalties. We are seeing a shift toward hybrid and edge architectures: rather than consolidating all enterprise data within the cloud, organisations are bringing containerised compute directly to where their physical and digital data already are housed.
Sovereignty and resilience are rising up the agenda. What should UK organisations be asking their providers about where their data is held, who can access it, and what happens when something goes wrong?
Sovereignty and resilience have shifted from tick-box compliance exercises into fundamental risk management priorities. If you are sitting down with a cloud or infrastructure provider today, you need to probe their answers on three specific issues:
“Can you prove end-to-end data lineage, or are we operating blind?” In our research for Iron Mountain’s The Information Supercycle report, 70% of organisations admitted they cannot trace the lifecycle or lineage of the data feeding their AI models. If an autonomous agent makes a biased lending recommendation or leaks confidential client data, saying "the algorithm did it" will not satisfy the FCA or the Information Commissioner’s Office. Ask your provider how their platform proves chain of custody from physical archive to digitised vector embedding and whether they can produce a defensible audit trail.
“How are you helping us manage data gravity and egress lock-in?” True resilience means you aren't trapped in one vendor's proprietary walled garden. Ask: If we need to shift an AI workload between our private on-premises setup, sovereign colocation facilities, and public cloud, how much friction and egress penalty will we face?
“What does failure look like beyond simple uptime SLAs?” Traditional resilience looked at 99.999% server availability. Modern resilience is about data integrity. If an ingestion pipeline gets corrupted or an unredacted set of sensitive data bypasses your filters, how quickly can that context be purged from model memory without tearing down the entire infrastructure?
Many organisations want to adopt AI but are held back by the state of their data. What do they need to get right on data before AI can deliver value?
If you connect an enterprise model or Retrieval-Augmented Generation (RAG) pipeline to that landscape, you are not creating intelligence; you are automating chaos. The model ingests Redundant, Obsolete, and Trivial (ROT) data, resulting in hallucinated outputs, bloated token bills, and skewed decisions. Before AI can deliver measurable value, organisations must do the operational heavy lifting:
Aggressive Data Hygiene: You cannot govern what you cannot see. Organisations have to audit their estates, classify legacy files, and defensibly purge dark data.
Build a Governed "Information Layer": Moving past basic PoCs means transforming unstructured records, whether that’s PDFs, contracts, or historical physical records into a structured, searchable knowledge layer enriched with metadata.
Embed Security at Ingestion: Redacting PII and commercially sensitive IP cannot be an afterthought; it must be handled systematically before data enters the model's context window.
Energy and grid access are now real constraints on building new digital infrastructure. What can the industry do to make better use of the power it has?
We talk constantly about grid constraints, planning permissions for sub-stations, and the race for nuclear or renewable power. But we rarely look inside the data centre to ask what that power is actually running. Right now, massive amounts of server capacity, electricity, and cooling are wasted powering petabytes of "dark data" and idle legacy hardware. There are three immediate ways the industry can make better use of available power:
Stop Powering Digital Waste: Purging ROT data from live disks doesn't just reduce cloud bills; it immediately frees up compute cycles and cooling overhead.
Active Hardware Modernisation & ITAD: Only 17% of organisations currently recycle savings from legacy IT into modern AI. Decommissioning aging, power-hungry servers through secure IT asset disposition (ITAD) reclaims floor space, recovers residual cash value, and unlocks the stranded kilowatt capacity needed to deploy modern, energy-dense GPU clusters.
Shift to Sustainable Colocation: Rather than trying to retrofit decades-old internal server rooms that struggle with thermal densities, organisations should partner with ultra-efficient, highly engineered colocation facilities designed around modern cooling techniques and renewable energy matching.
If you could ask the UK Government to do one thing to support the growth of cloud and digital infrastructure, what would it be and why?
Incentivise "data estate remediation" and circular hardware lifecycle management across both public procurement and corporate tax frameworks.
Right now, UK organisations from NHS trusts to FTSE 100 businesses have millions in capital and megawatts of power locked inside legacy systems. They pay recurring operational budgets to store decades of expired physical files and idle, unclassified digital data. If the government wants to accelerate the UK’s ambition to be a global AI superpower without blowing past net-zero targets or exhausting Treasury budgets, it should encourage organisations to clean up their digital foundations. This can be done practically by:
Introducing sustainability incentives or capital allowance reliefs for organisations that actively decommission inefficient legacy IT hardware through circular asset recovery (ITAD).
Updating Crown Commercial Service frameworks to mandate strict data estate hygiene and defensible data minimization before public sector bodies procure new AI solutions.
By helping organisations liquidate their data debt, the UK can free up the trapped power, rack capacity, and operational capital needed to self-fund a sustainable AI future.
