Britain did not triple its digital-infrastructure investment in a year. The Office for National Statistics has instead tested a broader definition and found that market-sector investment was £11.2 billion in 2025, compared with £3.6 billion under its current, narrower measure. The £7.6 billion difference is economically interesting, but it is a measurement gap, not a growth rate.
The ONS research article says the expanded estimate incorporates assets that support digital services beyond traditional telecommunications structures. It also labels the work statistics in development and says it should not be used for policy or decision-making. For investors, the right use is diagnostic: the experiment shows where conventional sector and asset boundaries fail to capture digital capacity. It does not yet provide a clean signal of a new capital-expenditure cycle.
Two boundaries produce a £7.6 billion gap
The current headline approach focuses mainly on physical communications infrastructure. The expanded method looks at assets held by telecommunications and data-processing, hosting and web-portal industries, then includes selected intellectual property and rights alongside physical structures. Because the perimeter changes, comparing £11.2 billion with £3.6 billion answers "how much more does the broad definition count?" It does not answer "how much did investment grow?"
This distinction prevents two common errors. First, the larger figure cannot be added to the smaller one; they are alternative measures with overlapping content. Second, the gap cannot be read as previously invisible data-centre construction alone. The expansion includes several asset classes and industry allocations.
The broader map may be more useful. Networks cannot provide digital services without software, databases, spectrum access and the structures holding equipment. A telecoms-heavy definition can understate the productive system. But a wider aggregate also becomes harder to interpret because assets with different depreciation, pricing and economic functions sit under one label.
Code, spectrum and buildings share one label
Across 2020 to 2025, the ONS says 77.6% of market-sector digital-infrastructure investment under the expanded method lay in two broad groups: software and databases, and other buildings, structures and transfer costs. That composition is a warning against treating "digital infrastructure" as a synonym for fibre trenches or server halls.
Software and databases are reproducible intangible assets whose value can scale without another physical site. Spectrum permits are rights to use a scarce public resource. Buildings and structures are long-lived physical capital. Combining them can describe the inputs behind a digital service, but the total does not reveal which constraint is binding. A cloud operator short of electrical capacity has a different problem from a telecom operator buying spectrum or a business capitalising software.
Ownership adds another layer. Government capital grants funded £0.7 billion, or 5.7%, of all market-sector digital-infrastructure investment in 2024 under the expanded framework. Public funding can enable private assets without making government their economic owner. Analysts therefore need to separate who finances an asset, who holds it and who earns from it.
The wider series fell in 2025
The same-method comparison gives a very different momentum signal. The £11.2 billion expanded current-price estimate was £0.5 billion lower than in 2024, a 4% decline. A larger level than the old definition coexists with a year-on-year fall. That is why the word "hit" can mislead when the underlying change is a revised boundary.
A separate ONS infrastructure release estimated total market-sector infrastructure investment at £31.3 billion in 2025, up 12.1%. That figure is in 2023 chained-volume measures, while the expanded digital estimate is reported in current prices and uses a specialised perimeter. The two cannot be compared as if they were shares of the same nominal total.
The decline also does not prove digital capacity contracted. Investment is a flow; the installed capital stock can still grow when new investment exceeds depreciation. Falling equipment prices can raise real capacity without an equivalent increase in nominal spending. Conversely, expensive construction can lift expenditure before a facility produces services. The experimental series needs price and stock measures before it can answer more of those questions.
A data centre disappears into its owner
The ONS's data-centre methodology explains why even a physical facility is difficult to isolate. National-accounts surveys collect data from reporting units classified by their dominant activity. A specialist host may sit in data processing, a real-estate owner in property and an enterprise-owned facility in its parent company's industry. The same kind of server hall can therefore appear in different sectors.
The ONS says it cannot currently identify data-centre investment or output separately. Construction, equipment and software enter investment during development; hosting, cloud and storage services enter output once the facility operates. This lifecycle is economically sound, but it prevents a single headline series from cleanly showing the sector's build-out.
That limitation matters after large project announcements. A pipeline is not expenditure, expenditure is not completed capacity and capacity is not utilisation. The statistical boundary improves coverage of assets without resolving those timing gaps. Company filings, planning approvals, grid connections and operator occupancy remain necessary for project-level analysis.
Coverage and use must confirm the capital stock
Independent outcome data can test whether investment is producing accessible infrastructure. Ofcom's 2025 Connected Nations report put UK full-fibre coverage at 77% of premises and gigabit-capable availability at 86% in the first quarter of 2025. Yet connections per 100 people were lower, showing that availability and adoption are separate.
The same principle applies across the expanded ONS perimeter. More software capital does not guarantee productivity, and a completed data centre does not guarantee high utilisation. The £11.2 billion estimate becomes more decision-useful if future releases provide stable asset weights, price-adjusted changes, capital stocks and clearer links to service output. It would be less useful if revisions are dominated by changing classifications or if data centres remain inseparable from unrelated parent activities.
The experiment has found capital that the narrow measure did not name. Its value lies in making the map more complete. Until the series matures, it should not be mistaken for evidence that Britain suddenly built three times as much digital infrastructure.