Britain’s Digital Investment Story Has Been Hiding in Plain Sight

The digital economy is often discussed as if it floats above the physical world: an ecosystem of platforms, software licences and AI models. The latest Office for National Statistics analysis offers a useful corrective.

Using a broader definition, the ONS estimates that UK market-sector investment in digital infrastructure reached £11.2 billion in 2025. Its existing, narrower approach would have put the figure at £3.6 billion. The difference is not a sudden discovery of billions of pounds in new spending. It is a classification problem with strategic consequences.

Digital infrastructure has traditionally been associated with masts, ducts, fibre and exchange equipment. Yet the systems on which modern businesses depend also require data-centre buildings, servers, network software, databases and rights to use radio spectrum. Treating these elements as separate, peripheral categories makes the economic foundation of cloud services and AI look smaller than it is.

That matters well beyond the national accounts. Boards making decisions about automation, customer data, cyber resilience or AI deployment need a more complete view of the assets that make those ambitions workable.

Digital transformation has a balance sheet, a supply chain and a power requirement.

---

The overlooked cost of digital capability

The ONS’s expanded measure includes selected investment in buildings, hardware, telecommunications equipment, software and databases across telecommunications and data-processing activities. Between 2020 and 2025, nearly four-fifths of the investment it identifies sat in just two categories: buildings and structures, and software and databases.

This is a more realistic description of how digital services are delivered. A retailer’s recommendation engine, a bank’s fraud controls or a manufacturer’s predictive-maintenance system may appear to users as a screen-level service. Their performance relies on a chain of physical and intangible capital: connectivity, compute, storage, cooling, software integration, security controls and skilled operations.

For many organisations, these costs arrive through operating contracts rather than a single capital project. Cloud capacity is purchased on demand. Software is subscribed to. Data is licensed, cleaned and governed continuously. This can make the underlying infrastructure easy to overlook in strategic planning, particularly where responsibility is divided among technology, finance, procurement, operations and risk teams.

The result can be a familiar mismatch. Leaders approve a customer-facing AI pilot while assuming the necessary data architecture, security reviews, vendor management and capacity planning are simply technical implementation details. They are not. They determine whether the pilot becomes a reliable business capability or remains an expensive demonstration.

Data centres make the issue harder to ignore

The ONS has also published a companion explanation of how data centres appear across the national accounts. It notes that data-centre investment and output cannot currently be identified separately in official statistics because facilities consist of multiple components and do not have a unique industrial classification.

That fragmentation mirrors the commercial challenge. A data centre is not merely a building: it contains construction work, power and cooling systems, servers, storage, networking equipment, software and data. Some assets will sit with a cloud provider, some with a specialist operator, and some with the customer organisation. Each may be governed, procured and measured differently.

Britain’s decision to classify data centres as critical national infrastructure has already acknowledged their operational importance. The new ONS work extends that argument into measurement. If a disruption to compute capacity can affect payments, logistics, health services, communications and business continuity, it is inadequate to view data-centre capacity as an obscure property or IT-sector concern.

There is also an energy implication. The ONS notes that data centres consumed an estimated 4.5 terawatt-hours of grid electricity in Great Britain during 2024, following substantial growth since 2020. For organisations buying cloud and AI services, energy exposure may be indirect, but it is still economically relevant: it can shape provider pricing, location decisions, resilience requirements and the credibility of environmental commitments.

---

A better board question than ‘what is our AI strategy?’

The more useful question is: which infrastructure dependencies must be true for our digital strategy to perform at scale?

That should prompt a practical inventory. Which services are dependent on a small number of cloud regions or connectivity providers? Where is business-critical data held, and how portable is it? Which software subscriptions have become essential infrastructure despite being managed as departmental spend? What is the recovery plan if a supplier, region or identity service fails?

The exercise should also distinguish between digital tools that improve a local workflow and systems that become a shared production dependency. The latter deserve higher standards of resilience, commercial scrutiny and operational ownership. A tool used by one marketing team can be assessed for utility. A platform used to make pricing, credit, workforce or supply-chain decisions must be assessed for continuity, controls and reversibility.

Ofcom’s latest network data provides a further reminder that availability and adoption are different measures. Full-fibre coverage has expanded rapidly, but take-up still develops over time. The same pattern applies inside firms. Purchasing technical capacity does not create value by itself. Processes must change, employees need confidence and leaders need evidence that a new capability improves outcomes rather than simply adding cost.

Measure the foundation as carefully as the feature

The ONS work should not be read as an instruction for every business to build its own data centre or capitalise every software cost. Its importance lies elsewhere. It shows that the economy’s digital base is wider than its conventional labels suggest.

For business leaders, the lesson is to stop treating infrastructure as the background to innovation. AI, cloud and data products depend on a layered system of physical assets, software, suppliers, permissions and energy. Those layers create both strategic advantage and concentrated risk.

The companies best placed to benefit will be the ones that can connect their technology roadmap to a clear investment map: what they own, what they rent, what they depend on, what can fail, and what commercial outcome each layer is expected to support. That is a more demanding discipline than announcing an AI strategy. It is also the one that makes a digital strategy credible.