How British Software Exports Show Up in Economic Growth
When A British Data Centre Buys An Imported Chip, Whose Economy Grows?
A British company orders servers, hires electricians and pays software engineers to get a new computing facility running. Every invoice belongs to the project. Only some of that spending measures production in Britain. A chip built overseas, a cable fitted in Yorkshire and a programme written by a team in London enter the national accounts in different ways. Work out where each contribution was produced and a headline about artificial intelligence lifting UK growth becomes much easier to understand.
The first estimate published on 12 November put UK real gross domestic product growth at 0.7% in the third quarter of 2026, following 0.5% in the second. That is growth across the economy between consecutive three-month periods. The Office for National Statistics also identified computer programming, consultancy and related activities as a major contributor, with several fast-growing businesses involved in artificial intelligence and cloud computing. It did not attribute the whole 0.7% to AI.
Where The Technology Signal Actually Appeared
Services output increased by 0.9% in the quarter. Within services, information and communication grew by 2.1%. Within that industry, programming, consultancy and related activities grew by 4.4%. These are nested measures. Adding the percentages would count parts of the same activity several times. Britain's programmers also represent a fraction of national output, so their industry can grow much faster than GDP without generating an equivalent national growth rate.
Imagine two industries, one large and nearly unchanged, the other small and growing fast. To calculate their combined rate you need the size of each at the start. The smaller industry’s growth can matter disproportionately at the margin without transforming the whole economy. This is what makes a contribution to growth different from a share of GDP. An industry can account for much of a quarter’s increase and still account for a modest portion of the total level.
The statistical classification adds a further complication. A consultancy might supply cloud migration to one customer, conventional accounting software to another and a machine-learning system to a third. All three contracts can appear under a broad business activity. Evidence that firms reporting strong turnover are involved in AI supports an explanation for some of the growth. Isolating AI precisely would require a breakdown of the products and services actually sold, their prices and the associated domestic production.
Follow One Server From The Port To The Rack
Suppose a UK operator buys a server manufactured abroad and retains it as productive equipment. In the expenditure account its acquisition can appear as investment. Imports are deducted in the same calculation because the overseas manufacturing output is already included in what the purchaser paid. Without the import deduction, Britain would appear to have produced a machine merely because a company here bought it.
Now add domestic activity around that purchase. A distributor may earn a margin. Transport workers may deliver the machine from the port. An electrical contractor may install the connection. A British engineering firm may design the cooling system. These services can add UK output. What counts depends on who performed the work, where the producing business is resident and how much value it added after purchased inputs. The server’s country of manufacture need not determine the origin of every service supplied alongside it.
A simple illustration uses round numbers. Suppose a business spends £100 on an imported item at the border and a further £20 on a wholly domestic installation service. Assume away taxes, imported inputs in the installation and other complications. The combined installed asset may be worth £120 to the purchaser. Deduct the £100 import and the domestic amount left in this example is £20. The example explains the accounting mechanism. It cannot tell us the actual British content of any particular data centre.
Why Turnover And Value Added Give Different Answers
A contractor can invoice a customer for a large sum and immediately owe a supplier much of it. Counting both invoices in full as new domestic output would double count intermediate purchases. The production approach to GDP instead focuses on value added, broadly the value of output less the goods and services consumed in producing it. That is why a dramatic increase in the prices of imported components need not produce a corresponding increase in real UK growth.
Software makes the distinction more interesting. A UK business developing software for customers abroad can generate British output and exports without installing a single server on British soil for the buyer. A company operating here might buy cloud services from a non-resident provider, contributing to imports of services. A multinational may operate UK facilities through a resident unit. Looking at the nationality of the brand on an invoice is a poor shortcut to the statistical answer.
The same underlying production can also be viewed from the income side. Engineers earn wages, businesses may earn operating surpluses, and taxes and subsidies enter the reconciliation between different measures. These are ways of looking at activity already measured. Adding the engineers’ wages on top of the full value of the software they produced would treat a second perspective on the work as a second piece of production.
Construction Today, Computing Tomorrow
A data centre affects the economy at different times. Surveying, design and building work generate activity while construction is under way. Equipment purchases arrive on their own schedule. Once operational, the facility can supply computing services over years. An announced investment budget covering several years should never be counted as if it were all output in the quarter of the announcement. Planning consent, a signed order and completed work represent different stages.
Nor does the opening of a large facility guarantee an equally large permanent workforce. Building and fitting it out require particular trades. Running it requires different skills and may use fewer people on site. Output can still be substantial because much of the service comes from expensive equipment supported by power, cooling, maintenance and software. For the local economy, the interesting questions concern jobs by phase, local procurement and what other activity the electricity connection enables or displaces.
Productivity comes later still. A retailer might buy access to an AI service because it expects fewer stockouts. Its spending can support output at a supplier before the retailer knows whether the system improves its own performance. A worthwhile productivity claim would need evidence about output relative to inputs, allowing for installation costs, training and any mistakes the new system introduces. The fact that businesses are buying a technology is easier to establish than the return they will ultimately earn.
What The Next GDP Release Can Tell Us
The November figures are a first quarterly estimate. Early estimates use the data available at publication and can change as more businesses report and figures are reconciled. Revisions need not imply that a particular company changed its story. They can arise because the statistical office received a fuller picture of a complicated economy. A small change in the overall rate may coexist with a persistent trend in one industry.
The word real deserves attention. Real GDP measures changes in the volume of production after allowing for price changes. If a consultancy bills the same number of hours at a higher hourly rate, its cash turnover rises. A statistician needs additional information before deciding that more service has been produced. Technology complicates this because quality can change rapidly. An improved service may complete a task faster even when the number of billed hours falls. Measuring a whole industry therefore involves more than summing invoices and applying the retail inflation rate. The published estimates are the result of that work, and they remain open to revision as better evidence arrives. A quarterly headline captures one stage in a much longer process of measurement.
One figure worth following is programming and consultancy output over several quarters. Another is business investment, which rose by 0.4% in the third quarter on the initial estimate. That rate covers investment across businesses and assets. Its difference from the 4.4% programming figure is unsurprising. One measures a category of spending, the other output in a particular industry. Neither can serve on its own as a national AI growth rate.
For the next announcement, ask three questions. Where was the added value produced? Over what period did the work take place? How specifically does the evidence connect the activity to AI? In the British data centre example, the answers lead from an overseas chip factory to local installers and on to software sold to customers. Britain can earn from several stages of that chain. The November figures show why those stages matter while leaving the exact contribution of artificial intelligence open.
Sources
GDP first quarterly estimate, UK — Initial quarterly output estimates and the programming industry figures.
AI spending helps lift UK growth to fastest since early 2024 — The contemporary reporting on AI activity and investment.
Annual National Accounts: Frequently Asked Questions (FAQs) — The accounting principles behind national output measures.
Next Reads
OpenAI’s A$7 Billion Sydney AI Campus: Why the NEXTDC Deal Matters Far Beyond Australia — An example of a large AI infrastructure commitment.
Starcloud Hits $2.3bn Valuation As Nvidia Backs AI Data Centres In Space — The engineering and economics of proposals to build data centres in orbit.
Anthropic’s $64.35 Billion IPO: Who Gets the Money? — Follow the difference between money raised and productive investment.
