The trade everyone is waiting for
The AI build-out has run into a limit that has nothing to do with software, and construction has not held this much leverage in half a century.
Somewhere in a factory order book there is a high voltage transformer with a delivery date late in this decade, and a project team already building its programme backwards from it. Before 2020 a unit of that size took six to twelve months to arrive, long enough to be irritating and short enough to plan around. Reuters has reported lead times on the largest units stretching to four years. The queue is not long because somebody in procurement forgot to press send. It is long because every hyperscaler, grid operator and industrial developer on the planet now wants the same steel, the same copper and the same specialist winding capacity inside the same narrow window, and there is no software patch for a factory that can only build so many transformers a year. The most consequential bottleneck in artificial intelligence is a lump of iron on a lorry.
Why is the real bottleneck physical rather than digital?
That fact has been slow to reach the commentary. Read most coverage of the AI boom and you would think the constraint was chips, or capital, or the regulatory mood in Brussels. The numbers suggest otherwise. Of the twelve gigawatts or so of American data centre capacity targeted for 2026, industry trackers reckon only about a third is under active construction; the rest has been announced, funded and photographed, but nothing has happened on the ground, despite build programmes that are supposed to run to under eighteen months. The Wall Street Journal has reported the American build-out falling well behind schedule. The gap between an announcement and a running facility is precisely the territory that construction has occupied for two hundred years, and for the first time in living memory the whole global economy is standing in it, waiting.
The scale of what is being attempted is genuinely difficult to hold in your head. At the Stargate campus outside Abilene in Texas, a crew of several thousand has flattened hills across an eight hundred acre site and laid enough fibre to wrap the earth more than a dozen times. Sam Altman has set out an ambition to build a factory capable of producing a gigawatt of new AI infrastructure every week, describing it in a blog post as the most important infrastructure project ever attempted, and he has told reporters at Abilene that what is there now is a fraction of what the site will become. The detail that ought to interest anyone in this industry is quieter and more revealing: OpenAI has been negotiating with the North American building trades unions about pathways for skilled construction workers. A company whose entire product is intelligence has worked out that its binding constraint is pipefitters.
How is Britain competing for this work?
Britain has decided to compete for this work, and the terms on which it is competing are notably physical. To qualify as an AI Growth Zone, a site has to demonstrate access to at least 500 megawatts of power by 2030 and at least a hundred acres available for construction by 2028. That is not a technology policy. That is a groundworks specification with a ministerial foreword. Five zones have been designated so far, from Culham in Oxfordshire to Anglesey and Gwynedd, with the North East securing government backing for a zone worth up to thirty billion pounds around Cobalt Park and the QTS campus at Cambois near Blyth. In March, planners approved the largest AI data centre campus in the country, 176 hectares on the former RAF airfield at Elsham Wolds in North Lincolnshire, which means a site that once dispatched Halifax bombers will spend the 2030s cooling processors. Glenigan expects data centre demand alone to lift office sector project starts under a hundred million pounds by thirteen per cent this year, and around twenty nine consented schemes are queued up in London.
Is there anybody left to build it?
For the people who actually deliver this work, the interesting question is not whether the money is real. It is whether there is anybody left to build it. The CITB's Industry Picture 2026 puts the position bluntly, describing the skills gap as structural rather than temporary and warning that leaving it unaddressed will mean projects delayed or cancelled outright and tender prices pushed up as firms compete for the same people. The board's own arithmetic calls for something in the order of forty one thousand additional workers every year to 2030. Roughly two fifths of the current workforce is over forty five, which is a demographic fact rather than a recruitment problem, and no plausible increase in apprenticeship starts closes that gap inside a decade. Turner and Townsend found every single one of its UK respondents reporting shortages of qualified mechanical, electrical and plumbing labour, the precise trades on which a data centre lives or dies, and forecasts construction cost inflation climbing to 3.7 per cent this year and 4.2 per cent next. AECOM has warned that demand for specialist MEP subcontractors in London has already outrun supply, leaving developers to wait their turn or pay to jump it.
What does a shortage look like from the other side of the table?
Turn that round and look at it from the other side of the table, which is where most of this industry has spent its working life. A shortage of electricians is a crisis if you are a client trying to hit a commissioning date. If you are a twenty six year old with a gold card, three years of panel work behind you and a phone that rings on a Sunday, it is the strongest negotiating position your trade has held since the seventies. The same shift is working its way through the professions. Assistant quantity surveyors are being handed intermediate packages earlier than the traditional five to seven year progression allowed, not because anyone has rethought career development, but because the alternative is watching a decent QS walk across the road for four thousand pounds more. Commissioning managers, temporary works coordinators and anyone who can genuinely run an M and E package to a fixed energisation date have become the scarcest commodity in British construction, and scarcity, in the end, is just pricing power with a hard hat on.
This is the part that deserves more attention than it gets, because it inverts an arrangement that has held since the middle of the 1970s. For fifty years the industry has been the squeezed party. Lowest price wins, retentions held for two years and then argued over, sixty day payment terms enforced by people with better lawyers, margins of two per cent on turnover that would embarrass a supermarket, and the whole attritional cycle punctuated every decade or so by a collapse that takes a supply chain down with it. That model worked, in its miserable way, because capacity was abundant and clients could always find somebody hungrier. When capacity becomes the constraint, the party holding it starts writing the terms. Two stage tenders and paid preconstruction stop being a favour extended to a favoured contractor and become the only way a client can get a credible date. Buildability advice offered in month two is suddenly worth more to a developer than a keen number offered in month nine.
Will the industry actually take the opportunity?
None of which means the industry will take the opportunity. The obdurate truth about construction is that it has been handed leverage before and has reliably bid it away within two years, because the tender that keeps the office busy always looks more urgent than the principle that keeps the business alive. There is also a serious argument, made by people who are not fools, that the data centre pipeline is a capital cycle rather than a permanent shift, and that firms which gear up their overheads to chase it will discover in 2031 what the PFI generation discovered in 2010. Anyone who lived through Carillion is entitled to a degree of scepticism about announcements measured in billions, and the honest answer is that some of these schemes will never leave the planning file. What has changed is not the certainty of the work. It is the balance of who needs whom, and that balance has moved further in construction's favour than at any point in most careers.
Where do the new tools fit?
The tools have arrived at a convenient moment, which is the genuinely propitious part of all this. The version that matters on site is duller and closer than the futures being sketched in California, and it is already here. It is being able to search four years of drawings, RFIs, inspection records and WhatsApp threads in the time it takes to walk from the cabin to the gate. It is spotting that the same edge protection observation has appeared on three projects run by the same subcontractor. It is a commercial team seeing cost drift in week six rather than in the month end CVR. None of that replaces judgement, and anyone selling it as such has never had to defend a programme to a client. What it does is give experienced people back the hours they currently spend hunting for information that somebody else already recorded, at exactly the moment when there are not enough experienced people to go round.
The transformer, meanwhile, is still in the queue. It will arrive on a Tuesday, on a low loader, escorted through a Lincolnshire village that spent eighty years watching aircraft instead, and it will be craned into a substation by people who learned the trade on schools and sheds and distribution warehouses and who now find themselves building the plant that runs the century. The industry has spent a long time being told it is slow, fragmented and resistant to change, usually by people who have never had to sequence a lift over a live grid connection. It is going to spend the next decade being asked, very politely, whether it could possibly go a bit faster.
What do these terms mean?
| Term | Meaning |
|---|---|
| Programme | The construction schedule, showing what happens when and in what order |
| Preconstruction | The design, pricing and planning phase before work starts on site |
| Two stage tender | Procurement where a contractor is appointed early on preliminaries, then prices the full job |
| Package | A defined parcel of work let to one subcontractor, such as groundworks or M and E |
| Subcontractor | A specialist firm engaged by the main contractor for one package of work |
| MEP / M and E | Mechanical, electrical and plumbing services: the pipes, ducts, cabling and plant inside a building |
| Commissioning | Testing and proving that installed systems work as designed before handover |
| Energisation | The point at which a building or substation is first connected and made live |
| Groundworks | Excavation, drainage, foundations and everything below the finished floor level |
| Temporary works | Structures built to allow permanent work to happen, such as propping or falsework |
| Buildability | How straightforward a design is to actually construct on site |
| Retention | Money the client holds back until the work is proved complete and defect free |
| Tender price | The figure a contractor offers to carry out the work |
| QS | Quantity surveyor, the person who measures, values and prices the work |
| CVR | Cost value reconciliation, the monthly check of what a job has earned against what it has spent |
| RFI | Request for information, a question to the design team that the site cannot proceed without |
| Gold card | JIB grading card confirming a qualified electrician's competence |