By Barry Lewington | Bushey

For most of my career, Data Centre decisions were made in a relatively stable environment. The variables were familiar – capacity, cost, location, connectivity, and the perennial build-versus-buy debate that has occupied infrastructure teams for decades. Those decisions mattered, but they were largely predictable in their scope and manageable in their complexity. What I am seeing now is fundamentally different, and organisations that are still applying the old decision-making framework to a new set of conditions are finding themselves exposed in ways they did not anticipate.

Three forces have converged to reshape the infrastructure landscape in a short period of time, and none of them is going away. Artificial intelligence is creating demand patterns that existing Data Centre infrastructure was not designed to accommodate. Power availability is emerging as a genuine constraint on where and how organisations can operate. And the resilience expectations placed on infrastructure, by boards, regulators, and customers have risen sharply at exactly the moment when the infrastructure itself is under the greatest stress. Understanding each of these forces individually is necessary. Understanding how they interact is what makes the difference between a Data Centre infrastructure strategy that holds and one that does not.

AI has fundamentally changed the way teams undertake load calculation

The arrival of AI at scale is doing something to the Data Centre infrastructure demand that most organisations did not fully model in their planning assumptions. Traditional enterprise workloads are relatively predictable in their resource consumption. They grow, they have peaks, but the underlying pattern is manageable and the infrastructure can be sized with a reasonable degree of confidence.

AI workloads are different in character. Training large models requires sustained, intensive compute at a scale that most enterprise Data Centre environments were not built to support. Inference workloads running models in production to serve real business applications create a different kind of demand that can spike unpredictably depending on usage patterns, and as organisations move from isolated AI experiments to AI embedded across multiple business processes, the aggregate infrastructure requirement grows rapidly and non-linearly.

The consequence is that organisations which thought they had adequate capacity are discovering they do not. Those that planned for AI in the abstract are finding that the real-world infrastructure implications are more significant than the models suggested, and those that relied on hyperscaler environments to absorb AI demand without fully understanding the cost implications are encountering a very different kind of problem when the bills arrive.

None of this means that existing infrastructure is wrong or that the hyperscaler model is flawed. It means that the infrastructure strategy needs to be built around what AI actually demands of it, not what was convenient to assume.

To add to the challenges, power is no longer a background consideration. Of all the changes reshaping Data Centre decisions, the one that has surprised the most organisations is power. For a long time, power was a cost line and an efficiency metric. It was managed by the facilities team, optimised through PUE improvements, and reported on as part of sustainability commitments. It was rarely a strategic constraint.

That has changed. In many markets, the availability of sufficient power capacity has become a genuine limiting factor on where Data Centres can operate and how quickly they can scale. Grid constraints are real with planning timelines for new power infrastructure now measured in years, and the energy density requirements of modern AI hardware being substantially higher than the environments that current Data Centre designs were built around, meaning that simply adding more servers is not as straightforward when the power infrastructure cannot support the additional load.

For organisations making infrastructure investment decisions today, power availability, both current and projected, needs to be part of the whole site assessment from the outset. This is not a facilities question, it is a strategic question, because a Data Centre that cannot scale its power capacity is a Data Centre that cannot scale its AI capability, which means it is a constraint on the business itself.

The sustainability dimension adds another layer of complexity. Boards are under increasing pressure from investors, regulators, and customers to demonstrate that energy consumption is being managed responsibly. AI workloads are energy-intensive, and that tension between AI ambition and sustainability commitment is one that infrastructure leaders need to navigate carefully and transparently. The organisations doing this well are those that have made energy sourcing, including renewable procurement, part of their infrastructure strategy rather than a separate sustainability workstream.

The third shift is in what resilience actually means, and how much of it is required. Organisations have always understood the importance of uptime. What is changing is the baseline expectation, the regulatory environment surrounding it, and the business consequences of falling short.

As more critical business processes run on digital infrastructure and as AI becomes embedded in those processes the tolerance for unplanned downtime shrinks. A system that was previously a productivity tool is now a decision-making engine, and a platform that was previously useful is now essential. The blast radius of an infrastructure failure has grown, and in many cases organisations have not updated their resilience posture to reflect the increased dependency.

This has not come unnoticed to regulators either. Across financial services, healthcare, critical national infrastructure, and increasingly other sectors, the requirements around operational resilience are becoming more prescriptive and more demanding. Demonstrating that your infrastructure can withstand disruption, recover within defined tolerances, and maintain continuity of critical services is no longer a best-practice aspiration. In many regulated environments it is a condition of operating.

The practical implication is that resilience needs to be designed in from the start of any infrastructure decision, not retrofitted afterwards. This means looking honestly at single points of failure, at geographic risk concentration, at the supply chain dependencies that sit behind the infrastructure itself, and at the testing regime that gives the organisation genuine confidence rather than assumed confidence in its ability to recover.

When we look at sourcing now in a more complicated world, we need to make some very concrete decisions on what to own, what to colocate, what to run in a hyperscaler environment, and how to move between those options as conditions change. The three forces described above have made it more so, and they have also made the cost of getting it wrong higher.

Organisations that locked into long-term Data Centre commitments before the full implications of AI infrastructure demand were clear are finding those commitments constraining. Those that moved aggressively to public cloud for everything are finding that the economics of AI workloads at scale do not always favour that model, and those that have retained significant on-premise infrastructure without updating the resilience architecture are sitting on risk that has not yet materialised but is considerably larger than it was a few years ago.

The right answer is not the same for every organisation, and anyone who tells you otherwise is selling a single solution to a nuanced problem. What is consistent across the organisations navigating this well is that they are making infrastructure decisions from a clear strategic position, they know what they are trying to achieve, they understand the constraints they are operating within, and they are making deliberate choices rather than drifting into outcomes by default.

At Bushey, our Data Centre infrastructure advisory work starts from that strategic position. The technology decisions follow from the business requirements, the risk appetite, and the realistic assessment of what the environment, power, planning, regulation, and capability will actually support. That sequencing matters enormously, and it is where the difference between a sound Data Centre infrastructure strategy and an expensive set of regrets is made.

The Data Centre decisions that organisations make in the next two to three years will shape their infrastructure capability for a decade or more. The capital commitments are significant, the lead times are long, and the pace of change in the underlying technology is rapid enough that flexibility needs to be built into whatever is decided.

This is not a moment for infrastructure decisions made by habit or by analogy with what worked before. The variables have changed materially enough that the old heuristics may not hold. What it requires is clear thinking, honest assessment, and the willingness to challenge assumptions that may have been valid five years ago but are not valid today.

The organisations that approach it that way will build infrastructure that serves them well through the changes ahead. Those that do not will find themselves revisiting these decisions sooner than they planned, and at considerably greater cost.

Barry Lewington is a technology strategist and Managing Director at Bushey, working with organisations across the UK to align their technology investments with business outcomes. He has been writing and speaking about enterprise technology for over 25 years.

Bushey provides independent governance and assurance for technology transformation. Through structured oversight and disciplined programme control, we ensure outcomes are achieved with clarity, accountability, and confidence, supported by specialist capability across change, project leadership, AI, Cyber, Data Centre, and M&A services. Our focus is on aligning transformation to business objectives, applying proven frameworks, and enabling secure, resilient, and future-ready environments.

#ArtificialIntelligence #AIGovernance #ResponsibleAI #DigitalTransformation #AILeadership #EnterpriseAI #TechnologyStrategy #GovernanceRiskCompliance #BusinessTransformation #InnovationLeadership