Imagine renting four floors of office space and leaving one of them permanently empty, lights on, heating running. Nobody would sign off on that. Yet that is roughly what cloud waste looks like in many organisations.
A new report from Insight, The Digital Sovereignty Trilemma, finds that organisations waste an average of 24% of their annual cloud capacity. At average annual cloud spend of €3.75 million, that is around €901,000 a year spent on nothing (IT Brief UK).
At the same time, AI is pushing infrastructure bills the other way. The same research finds AI driving a 12% year-on-year rise in hosting costs, with 85% of organisations evaluating or deploying dedicated infrastructure for AI.
Put those together and you get a very practical conclusion. For many organisations, the quickest way to fund AI is not a new budget line. It is the capacity they are already paying for and not using.
Where the waste comes from
The causes in the report will be familiar to anyone who has looked at a cloud bill in detail: limited visibility into cloud estates (47%), over-provisioning (47%) and inactive resources that were never switched off (46%).
None of these are exotic. They are the natural result of cloud being easy to start and nobody's job to stop. A team spins up a large environment for a load test and forgets it. A database is sized for peak and runs at 10% the rest of the year. A proof of concept from 2023 is still quietly running because nobody is sure whether anything depends on it.
The more telling figure is that 56% of organisations skip total cost of ownership assessments before major workload decisions. Decisions about where and how to run things are being made without anyone working out what they will cost over their life.
AI changes the cost curve
Traditional cloud waste was annoying but predictable. AI makes it volatile. Training runs, inference endpoints and GPU instances are expensive, and their cost depends on usage patterns that are hard to forecast. An AI feature that becomes popular can double its running cost in a month. One that nobody uses can sit on expensive reserved hardware indefinitely.
This is why the discipline of FinOps, bringing finance, engineering and product together around cloud cost, is moving from a nice-to-have to a core capability. If you cannot see what each workload costs and who owns it, you cannot make sensible decisions about where to invest in AI.
A practical plan to fund AI from waste
You do not need a transformation programme to start. A focused 90-day effort usually recovers a meaningful slice of spend:
- See it. Tag every resource with an owner, product and environment. Untagged resources get flagged, then switched off if nobody claims them.
- Right-size it. Compare provisioned capacity with actual use and resize the worst offenders. Schedule non-production environments to shut down out of hours.
- Commit wisely. Once usage is understood, use reservations and savings plans for the steady base load, not for everything.
Then ring-fence a share of the savings for AI work. It is a far easier conversation with a finance director than a fresh request, and it creates a healthy habit: new AI workloads arrive with cost visibility built in from day one.
Where sovereignty fits
The report's other headline is about digital sovereignty. Two-thirds of organisations see it as a critical strategic priority, and in the UK 78% say it is important today, rising to 90% within one to two years (IT Brief UK).
Sovereignty is a real consideration, especially for health data, public sector work and regulated industries, and it is becoming more pressing as AI models process ever more sensitive information. But it is also a decision with costs. Sovereign or UK-only hosting can mean fewer service options and higher prices.
The sensible approach is to classify your workloads by the sensitivity and regulatory status of their data, then decide where each one needs to live. Not everything needs the same treatment. A marketing site and a patient record system should not be making the same hosting decision, and treating them the same way is its own kind of waste.
The bottom line
Cloud waste is not a technology problem. It is an ownership problem. The organisations that get it under control will find they have room to invest in AI and resilience without asking for more money. Those that do not will find AI simply makes an expensive habit more expensive.
Our Managed Services practice helps organisations get visibility of cloud cost, cut waste and run AI workloads with cost controls built in. If your cloud bill has become a mystery, let's talk.
Frequently asked questions
How much cloud spend is wasted?
Insight's 2026 research found organisations waste an average of 24% of their annual cloud capacity, around €901,000 a year at average spend levels.
Why is AI increasing cloud costs?
AI workloads need expensive compute such as GPUs and their usage is hard to predict. Insight's research found AI driving a 12% year-on-year rise in hosting costs.
What is digital sovereignty?
Digital sovereignty is about controlling where data is stored and processed and which jurisdictions and providers can access it. In the UK, 78% of organisations say it is important today.