On Wednesday, John Healey told the Treasury Committee why data centres had a place in last month’s Budget.
“I do not want to see the US datacentre debate rerun in the UK,” the Chancellor said. “I want to ensure that we do not see a free-for-all in datacentres, as we saw in the US.”[1] He called the policy a “datacentre compact”. The published document is the Data Centre Covenant.[2] Go ahead, he said, “only when there are clear community benefits.” “That means that I am ready to say no to datacentres where they do not bring those benefits, but equally, if there are benefits to local areas, I will back projects, because we have to have this in order to protect and ensure our independence.” He added: “I do not want to see health data used in AI, other than kept in this country, and it is just the same with the security of data.”[1]
That is the right instinct, and it is pro-growth. In the firms I advise, nobody is asking for a blanket no. Britain should host data centres, with power at a price firms can afford, and refuse the ones that do not pay their way. Who owns the compute, who gets a permanent job, and who pays for the power while the grid catches up: a slogan cannot answer those questions. A published, measurable test can.
What the Covenant requires
The Covenant, published on Budget day, sets five tests that all data centres are “encouraged” to meet, and that “strategic” ones must meet to qualify for faster grid connections and planning. “Strategic” will be defined in due course.[2] Developers should cover the energy costs they create. They should agree a Local Benefit Plan that “can be spent on local priorities, like transport, parks and town centres”. New facilities should be designed to a Power Usage Effectiveness of 1.3 or better. Developers should “create good, well-paid jobs that last beyond construction,” with local recruitment and paid apprenticeships, and provide “secure, resilient UK access to compute” for the NHS and other public services.[2]
The Covenant says strategic data centres should report each year on their energy and water use.[2] The Covenant does not ask them to publish how many jobs are permanent, how many are junior, the investment per post, the UK ownership share of the compute, or when any temporary on-site gas plant switches off. Construction jobs and a Local Benefit Plan can meet the letter of “community benefits” while the first-rung roles AI removes go uncounted.
Orbital Data Centres has proposed a scheme near Junction 24 of the M25 at “over £1.5 billion,” with over 1,400 construction jobs and 170 in operations. On those figures, that is about £8.8 million of investment for each operational post (my calculation). The figures are the developer’s, and the scheme is at consultation.[3]
Measure the claim against 0.2 percentage points
The figure still in circulation is that AI could add £400 billion to the economy by 2030. That came from a 2023 report commissioned by Google and produced by Public First. The government has since made its own assessment: AI adoption and data investment could add 0.4 to 1.3 percentage points to UK productivity growth, equivalent to £55 billion to £140 billion of GVA in 2030, with a central estimate of £120 billion. Those estimates, it said, are “hugely uncertain.”[4]
The Office for Budget Responsibility is more cautious. Its October central case is that AI boosts productivity growth by 0.2 percentage points a year by the medium term, and “the scale and timing of any effect is highly uncertain.”[5] That is the number to measure the policy against.
Who holds the keys
Healey is right that independence requires data centres here. Location is not ownership. An FOI request by Aventail, reported by Computer Weekly, found that 10 of the 11 AI tools deployed or bought by the Cabinet Office run on models from Google, OpenAI, Microsoft or Anthropic. The exception is PolyAI, a British firm. The models may be hosted in the UK. The design, access and pricing, the report said, are not.[6] The Cabinet Office says the tools are model-agnostic and can be swapped or self-hosted.[7] That is the right architecture. The switch has yet to be tested in public.
A competition opened this week for a host site for the AIRR supercomputer. Up to about £180 million pays for the building, power, cooling and networking, not the computer, to deliver 10 MW of IT load by the end of 2027 and up to 30 MW by March 2029.[8] The Covenant puts the need at 6 GW of AI data centre capacity by 2030.[2] Thirty megawatts is 0.5 per cent of that (my calculation). It is still the right kind of pound: UK-hosted public compute.
US firms have the capital and, usually, the cheaper power. I want the halls built here. I also want to know who holds the keys.
Power at a price that works
Energy is the other constraint. In April, OpenAI paused Stargate UK until “the right conditions such as regulation and the cost of energy enable long-term infrastructure investment.”[9] A Stage 1 study published by Reabold on 9 November puts the running cost of on-site gas-fired power at West Newton at £44–58 per MWh, against a £217.8 per MWh reference price for extra-large users of the UK grid. Those are concept-level estimates, with first power targeted for the second half of 2030.[10] A faster place in the queue leaves that gap open.
The Covenant allows a “tightly controlled framework” for on-site generation, only for “exceptionally strategic” projects, until a grid connection is secured.[2] BloombergNEF’s 181 GW of proposed on-site gas covers the United States, the UK and Europe, and about 90 per cent of it is in the US. It is mostly an American number.[11] In Britain, Future Energy Networks recorded 113 data-centre gas-connection enquiries in 2024 and 2025, some of them speculative, with 46 applications and seven connections. Wapseys Wood in Buckinghamshire and Quest Park in Bedfordshire are already seeking on-site gas.[11][12] That framework is still to be written, and the Covenant does not yet require a published end date, megawatt figure or residual system cost.
Publish the numbers
One ask: disclosure. It needs no new spending programme. Any site given Covenant support, or priority as a Nationally Significant Infrastructure Project, should publish, as a condition of that support: its permanent jobs and junior hires or paid apprenticeships; its investment per permanent job; the UK ownership share of its compute; and the end date, megawatts and residual system cost of any temporary on-site gas. The gas terms belong in the on-site generation framework before it is finished. These numbers will show whether “community benefits” means what the public hears when the Chancellor says he is ready to say no.
What firms should do now
Firms need not wait. In the businesses I advise, the lesson is the same: if you rent the intelligence, plan for the lease to be called in.
First, have a fallback you have tested. A second model, exportable data, and a measured time to switch are worth more than a clause that says the architecture is model-agnostic. Second, keep the bottom rung. The Bank of England’s Agents report that AI is making firms less likely to hire or replace staff. Firms in the Bank’s Decision Maker Panel expect about half of AI’s productivity gain to come from slower employment growth.[13] Anthropic says junior hires fell from 27 per cent to 9 per cent of its own technical hiring.[14] Redesign junior roles around checking and supervising AI output, and hire into them. An employer still pays £3,750 a year in National Insurance on a £30,000 salary (my calculation), and nothing on the software that replaces it.[15] Third, if your product depends on cheap, reliable inference, price UK power honestly. OpenAI already did.
I want data centres built in Britain, at a power price that works, with compute the country can use. Ready to say no is a stance. Publish the test and it becomes a policy.
Notes & Sources
- Treasury Committee, Oral evidence: Budget 2026, HC 682, Wednesday 11 November 2026, Q336 (John Healey). Official transcript: https://committees.parliament.uk/oralevidence/18279/html/. Published on 11 November 2026; listed at https://committees.parliament.uk/committee/158/treasury-committee/publications/oral-evidence. The official wording differs from some contemporaneous press reports, which used contractions and paraphrases; this article quotes the transcript. Healey said “datacentre compact”; the published policy is the Data Centre Covenant. Session page: https://committees.parliament.uk/event/28691/formal-meeting-oral-evidence-session/.
- HM Government, The Data Centre Covenant, 28 October 2026: https://www.gov.uk/government/publications/data-centre-covenant/the-data-centre-covenant. The 6 GW-by-2030 figure is in this document: “The Government estimates that to protect our national interest… will require 6 GW of AI data centre capacity by 2030.” Tests are “encouraged” for all data centres and required for “strategic” data centres “to qualify for additional government action”; “further details in due course on the definition of strategic data centres.” Temporary on-site generation: “a tightly controlled framework to permit on site generation until a grid connection can be secured… only… for exceptionally strategic projects,” and only where reserved, reallocated or accelerated grid capacity cannot be secured in time, “subject to conditions such as ensuring energy security.” No published requirement for an end date, megawatt figure or residual system cost. Reporting: strategic data centres “should report annually on their energy and water use”; the document asks for no report on jobs or ownership.
- John Doyle, director, Orbital Data Centres, quoted in Enfield Independent, “£1.5bn data centre plans unveiled near M25 Junction 24,” 3 November 2026: https://www.enfieldindependent.co.uk/news/26695700.1-5bn-data-centre-plans-unveiled-near-m25-junction-24/. Developer estimates: “an investment of over £1.5bn… over 1,400 jobs during construction, 170 during operations and support 85 more in the wider area,” plus £6.8 million a year in business rates. Arithmetic on the stated figures: £1.5 billion ÷ 170 ≈ £8.82 million per operational job. Public consultation was open until the end of November 2026. These are proposals, not approved scheme figures.
- Written question UIN 32092, tabled 9 October 2026, answered 26 October 2026 by Kanishka Narayan: https://questions-statements.parliament.uk/written-questions/detail/2026-10-09/32092. “The £400 billion figure was drawn from a publicly available 2023 report commissioned by Google and produced by Public First, which estimated that generative AI tools could create over £400 billion in value for the UK economy by 2030.” The government’s “more recent assessment”: “AI adoption and data investment could add 0.4 to 1.3 percentage points to the UK’s productivity growth. These estimates are hugely uncertain, but this would be the equivalent of adding £55-140 billion to UK GVA in 2030. The central adoption estimate is £120 billion.” The 0.4–1.3 percentage point range is not stated as an annual rate, and is not directly comparable to the OBR’s 0.2 percentage points a year.
- Office for Budget Responsibility, Economic and fiscal outlook – October 2026, paragraph 1.27 (key risks to the economy forecast): “The development and adoption of artificial intelligence could have a significant positive effect – our central forecast is for it to boost productivity growth by 0.2 percentage points a year by the medium term – but the scale and timing of any effect is highly uncertain.” https://www.gov.uk/government/publications/october-2026-economic-and-fiscal-outlook/economic-and-fiscal-outlook-october-2026. Also at https://obr.uk/publications/economic-and-fiscal-outlook-october-2026/.
- Karl Flinders, “UK government ‘house of cards’ heavily reliant on US AI models,” Computer Weekly, 5 November 2026: https://www.computerweekly.com/news/366653713/UK-government-house-of-cards-heavily-reliant-on-US-AI-models. FOI request by Aventail: “10 of the 11 AI tools deployed or bought by the Cabinet Office use AI models developed by US tech firms.” “While the models are hosted in the UK, the design, access and pricing are controlled by US organisations.” Suppliers named: Google, OpenAI, Microsoft or Anthropic, directly or through Azure and AWS; the exception is PolyAI, used for SSCL. The original FOI response was not independently inspected for this draft; the figures are taken from Computer Weekly’s report. The article uses “Cabinet Office,” not a claim about every central government tool.
- Cabinet Office response, as reported by Computer Weekly, 9 November 2026, and summarised by Resultsense, 10 November 2026: https://www.resultsense.com/news/2026-11-10-cabinet-office-model-agnostic-ai-tools/. In-house tools from GDS and i.AI described as using “decoupled, model-agnostic architectures.” Google Gemini is the main authorised enterprise generative AI tool. A contingency plan exists; operational details were withheld.
- Department for Business, Innovation, Science and Trade, AIRR Heterogeneous Supercomputer Host Site and Data Centre Grant Competition, opened 11 November 2026, closes 12:00 on 25 November 2026: https://find-lb.find-a-grant-support.service.cabinetoffice.gov.uk/grants/airr-heterogeneous-supercomputer-host-site-and-data-centre-grant-competition-1. “Funding of up to approximately £180 million is anticipated, subject to approvals.” Capacity: “an initial 10 MW IT load by 31 December 2027, expanding to a total of 30 MW by 31 March 2029.” “The grant does not fund the supercomputer hardware or its installation, integration and acceptance.” Applicants notified by 8 January 2027. The listing is inconsistent in places: its Dates section runs the window to 26 November, and its Objectives give “20–30MW by FY2029/30”, so this article says up to 30 MW. My calculation: 30 MW ÷ 6,000 MW = 0.5 per cent.
- BBC News, “OpenAI pauses UK data centre deal over energy costs and regulation,” 9 April 2026: https://www.bbc.com/news/articles/clyd032ej70o. OpenAI spokesperson: “we continue to explore Stargate UK and will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment.” The reasons given were energy costs and regulation, not energy costs alone. The project was described as a large data centre in North Tyneside, in a partnership with Nvidia and Nscale.
- Reabold Resources plc, RNS 3179C, “West Newton Integrated Gas to Compute Feasibility Study,” 9 November 2026, 07:00: https://reabold.com/announcements/odL22M-reabold-resources-west-newton-feasibility-study. The body of the RNS: “the running cost of power from West Newton gas, including UK ETS carbon costs and operating and maintenance costs, is estimated at £44–58/MWh… This compares with a reference price of £217.8/MWh for extra-large users of grid electricity.” The highlights of the same RNS give a slightly narrower operating-cost range of £44–54/MWh; this article uses the body figure. All-in cost including recovery of power-plant capital is given as £107–121/MWh. “The basis of the cost estimates are concept level (Class 5) estimates.” “First power targeted for H2 2030.” Scope: approximately 67 MW of IT capacity. This is a Stage 1 feasibility study, not a final investment decision.
- Sandra Laville, “AI datacentres plan on-site power stations on huge scale despite climate risks, research reveals,” The Guardian, 11 November 2026: https://www.theguardian.com/world/2026/nov/11/ai-datacentres-on-site-gas-fired-power-stations. BloombergNEF research: 181 GW of proposed on-site gas-fired capacity at data centres “across the US, the UK and mainland Europe”; “90% of on-site gas-fired power stations are planned” in the US. UK sites named: Wapseys Wood, Buckinghamshire, and Quest Park, Bedfordshire, “among the first to be fast-tracked through the government’s nationally significant infrastructure project regime.”
- Future Energy Networks, “Data centre gas connections enquiries surge amid electricity grid queues,” published via IGEM, 16 April 2026: https://www.igem.org.uk/resource/data-centre-gas-connections-enquiries-surge-amid-electricity-grid-queues.html. “A total of 113 enquiries regarding data centre gas connections” over 2024 and 2025; “around three times as many submitted in 2025 compared to 2024.” “Gas networks are currently processing 46 applications… Seven applications have been secured for connection… equalling a combined capacity of 15.4 terawatt-hours.” FEN notes: “a proportion of these enquiries are at a speculative stage only. In some cases, there will be multiple enquiries related to one connection request.”
- Bank of England, Monetary Policy Report, November 2026 (5 November 2026), Box D, p58. The Bank’s Agents report that AI adoption is reducing the likelihood of firms hiring or replacing workers, and is not materially increasing the likelihood of firing so far. Firms responding to the Decision Maker Panel survey expect AI to boost productivity growth by about 0.9 percentage points a year over the next three years, “split roughly equally between slower employment growth and faster output growth.”
- Anthropic, How employment is changing at Anthropic, 16 October 2026: junior hires fell from 27 per cent to 9 per cent of technical hiring. Self-reported by the company about its own staff.
- HM Revenue & Customs, Rates and thresholds for employers 2026 to 2027, last updated 1 September 2026: https://www.gov.uk/guidance/rates-and-thresholds-for-employers-2026-to-2027. Employer (secondary) Class 1 National Insurance is charged at 15 per cent on earnings above the secondary threshold of £5,000 a year. On a £30,000 salary: (£30,000 − £5,000) × 15% = £3,750 a year, before any Employment Allowance or other reliefs. These rates were unchanged at the 28 October 2026 Budget, and were rechecked for this draft.