In partnership with
In Australia, a megawatt of data centre now costs up to £9.1m ($12.1m, €10.6m). In Tokyo, the most expensive market on earth, it reaches £11.4m ($15.2m, €13.4m). The processors are not what moved these numbers.
Globally, the average build cost rose from £5.8m ($7.7m, €6.8m) per megawatt in 2020 to £8m ($10.7m, €9.4m) in 2025. JLL forecasts £8.5m ($11.3m, €9.9m) for 2026, a seven percent compound annual rise. That is a 39% increase in five years.
Look inside the megawatt. The silicon is a separate line, installed later by the tenant. The build itself is electrical and mechanical work. Its price is set by trades and by the equipment they connect, not by chips.
The opening problem is not silicon. Electrical and mechanical work is 76% of a build
Turner and Townsend's 2025 index splits an air-cooled build into four parts. Electrical systems are 54% of the cost. Mechanical systems add another 22%. Core, shell and general works make up the rest.
The graphics processors are not in that figure. They arrive as tenant fit-out, once the hall can take power. Even a high-density AI hall costs only 7 to 10% more to build than a standard one. The megawatt price is a labour and equipment price. It is not a chip price.
The pressure is broad, not niche. Almost half of the firms surveyed reported build-cost rises of 6 to 15% in the past year. A fifth saw rises above 15%.
Traditional cost models fail when a transformer takes five years to arrive
Traditional budgeting assumed the gear turned up on time. That assumption has broken. Sightline Climate reports high-power transformers once took 24 to 30 months. Today the wait can stretch to five years.
Switchgear and batteries sit in similar queues. The scarce input is not compute. It is the electrical plant that delivers power to the campus. A model that prices the chips but ignores the switchgear now prices the wrong risk.
The engineering reality: power equipment, not processors, sets the schedule
Of 16 gigawatts planned globally for 2026, only about 5 gigawatts is under construction. The rest waits on power and parts. A large build takes more than a year to energise once it has permits. A junior estimate treats the building as the long pole. The real long pole is the substation.
This is where expertise earns its place. Sequencing the electrical order ahead of the silicon protects the schedule. Ordering a transformer late can idle a finished hall for years. The engineering decision precedes the procurement decision.
The strategic disconnect: £22.6bn campuses rise as a third of 2026 capacity slips
The ambition and the delivery curve are pulling apart. The UAE's Stargate campus in Abu Dhabi is now costed at more than £22.6bn ($30bn, €26.4bn) for 5 gigawatts. The campus spans 19 square kilometres, with a first phase due this year. Capital at that scale is not the binding constraint.
Delivery is. Sightline Climate estimates 30 to 50% of large data centres planned for 2026 will slip. The causes are power access, equipment shortages and local opposition. Backers span the United States and South Korea. The money is ready long before the megawatt is.
When a hall is built but cannot yet be powered, how does your organisation account for it?
Regional prices diverge: Tokyo builds at £11.4m, Dublin at £7.5m a megawatt
The same megawatt costs wildly different sums by place. Turner and Townsend ranks Tokyo, Singapore and Zurich as the most expensive, at £11.4m ($15.2m, €13.4m), £10.9m ($14.5m, €12.8m) and £10.7m ($14.2m, €12.5m) per megawatt. Dublin and Madrid sit near £7.5m ($10.0m, €8.8m). London, the UK's largest market, builds at about £9.0m ($12.0m, €10.6m).
Australia runs from £5.9m ($7.9m, €7.0m) to £9.1m ($12.1m, €10.6m), up 3.8% in a year. What separates these markets is not chip price. It is land, contractor pools, workforce depth and access to power. In Australia, Cushman and Wakefield notes legacy halls are increasingly hard to retrofit for dense AI loads. Silicon costs the same everywhere. People and permits do not.
Policy can permit a site. It cannot manufacture a transformer
Governments have moved on paperwork. Yet in the United States, moratoriums have been proposed in at least ten states, including Louisiana, Michigan, New York, Ohio and Virginia. One £752m ($1bn, €880m) Michigan project was withdrawn after local opposition.
A permit does not conjure an engineer or a substation. Policy can speed a planning decision and a grid application. It cannot compress a factory queue for high-voltage equipment. The regulatory lever and the supply-chain lever are not the same lever.
The path forward prices power readiness first, as Australia already shows
The fix is to price power readiness before land and before chips. In Australia, projects that secure reliable power early move ahead. Cushman and Wakefield calls power readiness the defining factor of development.
The discipline is simple to state. Sequence the electrical order ahead of the silicon. Underwrite equipment lead times honestly. Prefabricated and modular construction can shorten the critical path, where the electrical order is placed early. Intelligence can model that sequence and flag the long-lead item. A competent engineer still places the order and signs the plant off. Intelligence plans the build. Expertise still delivers it.
The investment implication: at £8m a megawatt, a stranded hall is dead capital
The cost of getting this wrong is measurable. A hall built at £8m ($10.7m, €9.4m) per megawatt earns nothing until it can be energised. Grid connection lead times now exceed four years in primary markets.
Every idle month is capital sitting still. The pattern is not new. A quarter of 2025's planned projects already slipped into this year. The visible cost is the build. The hidden cost is the wait between finished and working. Underwriting that ignores equipment lead times and commissioning capacity understates the real exposure.
The bottom line
The industry prices itself in chips and gigawatts. The build is priced in electrical labour and the equipment that labour installs. Australia, Japan, Ireland, the UAE and the United States all meet the same wall.
Intelligence can plan the sequence and flag the gap. Expertise still orders the transformer and commissions the plant. At £8m a megawatt, the cost that matters is not the silicon. It is the wait for everything around it.
Next week: Grid connection now takes four years. Behind-the-meter power is the workaround, and its economics.
Own Search With Podcasts
Your competitors are fighting over the same keywords. The smartest brands are building the authority that search engines, AI platforms, and customers trust everywhere.
Every relevant podcast appearance can produce branded mentions, backlinks, transcripts, citations, clips, expert content, and third-party proof that keeps compounding across search and AI discovery.
PodPitch searches millions of podcasts, finds the shows that matter to your market, develops the angle, sends personalized pitches, and follows up automatically until your experts are booked.
Growth teams are already using podcast appearances to build distributed authority that cannot be manufactured by publishing another generic SEO article.
Only 20 SEO, AEO, and GEO demo spots are available this month. Once they’re claimed, the offer disappears.
Start building searchable authority now, before your competitors own the conversations shaping your market.