Jobs, tax and the promises made

From the paper “Beyond Planning Permission”

The full text of the relevant sections of the paper, with its own numbered notes. Note numbers match the paper.

3. The Economic Case

The economic case for data-centre investment is substantial.

Modern economies depend upon digital infrastructure. Data centres support financial services, telecommunications, scientific research, cloud computing, public administration and AI-enabled technologies.²⁷

There is also a strategic argument. Domestic digital infrastructure may improve resilience, reduce dependence upon overseas capacity and support national economic development.^28,29,30

Nevertheless, economic claims should be examined carefully.

Although construction phases can generate substantial temporary employment, operational facilities are often highly automated and may support relatively small permanent workforces compared with the scale of the infrastructure involved.^31,32

Planning decisions should therefore distinguish clearly between construction employment, long-term jobs and wider economic benefits.

Employment and the Long-Term Jobs Question

This distinction deserves closer attention than it typically receives at the application stage.

Hyperscale data-centre campuses can occupy sites of considerable scale - in some cases land otherwise allocated in local development plans for general employment use - while supporting comparatively modest permanent workforces once construction is complete: one recent Welsh example is put at around 120 permanent jobs.³³

One independent analysis of the wider UK sector has gone further, questioning whether headline job-creation figures quoted by developers at the application stage are consistently matched by actual long-term operational employment once facilities are operational.³⁴

At the same time, UK-wide projections point toward substantial sector growth in employment terms. techUK has estimated that data centres currently support around 43,500 jobs across the UK, with the potential for this to rise significantly - some analysis suggests towards 100,000 or more by 2035 as capacity expands, and higher still once workforce turnover and retirement are taken into account.³⁵ These figures are not necessarily inconsistent with modest site-level headcounts: they are aggregated across a growing number of facilities nationally, and a meaningful share reflects construction, supply-chain and specialist technical roles rather than steady-state operational staffing at any single campus.

Planning authorities and communities should therefore be cautious about treating national sector-level projections as a guide to the number of long-term jobs any particular development will support locally.

A further, more difficult question concerns the effect of increasing artificial intelligence adoption on data-centre employment itself. On one view, greater automation of monitoring, predictive maintenance and remote operations management is likely further to reduce the already modest steady-state headcount required to run a large facility over its operational life, which may extend to twenty years or more.

On another view, the same trend may generate new categories of specialist employment - in AI infrastructure engineering, data-centre design and grid integration - that at least partially offset losses elsewhere, and commentators have separately identified a persistent skills shortage in the sector, with close to half of the existing technical workforce expected to reach retirement age by 2030.³⁶

These two observations are not easily reconciled, and the honest position is that the net long-term employment effect of AI-driven automation on data-centre operations remains genuinely uncertain - the jury is still out.

This uncertainty is not, in itself, a reason to resist development, but it is a reason for government - in both Cardiff and London - to be considerably more explicit than it has been to date. Neither the UK Government's National Planning Policy Framework reforms nor Welsh planning guidance currently offers planning authorities clear, evidence-based guidance on realistic long-term employment expectations for data-centre developments, as distinct from construction-phase activity.

Given the scale of land these developments can occupy - often on sites otherwise capable of accommodating higher-employment-density uses - government should set out a considered, updated view of how AI-driven automation is expected to affect long-term staffing levels at such facilities, so that the employment case for individual proposals can be weighed accurately, rather than assumed, against the alternative uses of the land involved.

International Profits and the Question of Tax

A separate question, distinct from the employment case addressed above, is whether the United Kingdom and Wales secure a fair fiscal return from data centre development once these facilities are built and operating.

Many of the international groups building data centres at scale in England and Wales are large multinationals capable of arranging their corporate affairs so that profits attributable to UK operations are recognised, for tax purposes, in other and more lightly taxed jurisdictions. The Digital Services Tax, often assumed to reach large technology companies generally, does not address this: it applies only to revenue from social media services, internet search engines and online marketplaces, and has no application to cloud computing, data storage or the underlying infrastructure services that a data centre supplies.^37,38

The OECD's Pillar Two reforms, transposed into UK law as the Multinational Top-up Tax, were intended to guarantee a minimum 15% effective tax rate on the profits of large multinational groups wherever they are booked, closing off much of the incentive to shift profits to low-tax jurisdictions.

In practice, a ‘side-by-side’ agreement reached in January 2026 exempted US-headquartered groups from Pillar Two calculations, and HM Revenue and Customs has told the House of Commons Committee of Public Accounts that this exemption will reduce the UK's expected annual benefit from Pillar Two by around £600 million, down to about £1.6 billion. Given that most of the large hyperscale operators active in the Welsh and English data centre market are US-headquartered, this carve-out is directly relevant to the sector under discussion in this appraisal.

The Committee's report, published on 10 July 2026, went further, recording that international tax risks - including businesses artificially shifting profits to lower-tax jurisdictions - account for some £21 billion of the tax currently under investigation by HMRC in relation to large businesses generally, and concluding that the scale of risk posed by large multinationals diverting profits across borders remains significant even with the new minimum tax rate in place.³⁹

A related, more local, fiscal question concerns business rates. Data centres are valued for rating purposes on a basis that depends heavily on technical assumptions about plant and machinery, and on whether unequipped ‘white space’ awaiting tenant fit-out is rateable at all - a point of genuine legal uncertainty until the Upper Tribunal's 2021 decision in Ricketts (VO) v Cyxtera Technology UK Ltd confirmed that such space could, on the facts of that case, be treated as non-domestic property in its own right.

The valuation of large, technically complex data centre premises accordingly remains more contestable, and less transparent to an outside observer, than the valuation of most other commercial premises, which in turn affects the local business rates revenue such sites generate for Welsh and English authorities.^40,41

Notably, when the Senedd's own research service examined the growth of the data centre sector in Wales, its analysis addressed investment, employment claims, energy, water and grid capacity at length, but did not extend to any assessment of the tax or business rates return to Wales from this investment.⁴²

This appraisal does not attempt to resolve these questions, but it suggests that government should be pressed for clarity in the same way as on the employment and AI questions raised elsewhere in this paper. A written question that might be put to the Senedd or a Westminster MP is:

‘What assessment has [the Treasury / the Welsh Government] made of the corporation tax and business rates revenue forecast to be collected from data centre operators in [the UK / Wales] over the next five years, and what steps are being taken to ensure that international operators account for profits attributable to their UK data centre operations in the UK rather than in lower-tax jurisdictions?’

No answer to a question in these terms currently exists, and no Freedom of Information request on this specific point appears to have been made via WhatDoTheyKnow.

The Scale and Deliverability of Proposed Investment

The scale of capital committed to AI-related data-centre infrastructure by the largest technology companies has increased sharply.

Combined 2026 capital-expenditure guidance from Microsoft, Alphabet (Google's parent company), Meta and Amazon reached approximately $725 billion, up from around $410 billion in 2025, with Microsoft and Alphabet each guiding to approximately $190 billion and Amazon raising its guidance to approximately $220 billion.

Separately, the Stargate joint venture involving OpenAI, SoftBank and Oracle, first announced in January 2025 at a headline $500 billion over four years, had by September 2025 committed sites totalling almost 7 gigawatts of capacity and over $400 billion of confirmed investment.^{43,44,45,46,47}

These sums are not spread evenly. They tend to show the speed with which decisions are made which must be unique and possibly dangerous.

Announcements to date concentrate heavily on the United States, the Gulf states and a small number of other jurisdictions, including Wales.

Microsoft and the Emirati company G42 announced a further $15.2 billion of UAE data-centre investment in November 2025, part of a wider pattern of Gulf-region commitments that analysts have described as a 'trillion-dollar bet', though regional conflict has since been reported as testing the pace of that expansion.

Closer to home, the South Wales AI Growth Zone has been the subject of an estimated £10 billion of investment, with figures of up to £100 billion suggested across all UK AI Growth Zones.

Where reasons are given for choosing a particular location, they typically cite access to power (including renewable generation), available land, planning support and connectivity. This appraisal has not identified independent verification that these advantages outweigh the water and grid constraints discussed in Sections 4 and 5 below, and whether they make a location genuinely suitable, rather than merely available, should be treated as an open question rather than a settled one.^48,49

Whether investment of this scale is realistically deliverable is contested among analysts, and the material in this paragraph is commentary and opinion rather than established fact.

Sequoia Capital's David Cahn has argued, in an analysis first published in June 2024 and periodically updated, that a substantial and growing gap exists between AI infrastructure spending and the revenue AI companies currently generate, a concern echoed in mid-2026 reporting of rising anxiety in bond markets over debt-financed capital expenditure and increasingly circular investment arrangements between AI developers, cloud providers and chip manufacturers.

Other analysts have dismissed these concerns, pointing to continued revenue growth.

This appraisal takes no position on which view is correct; it notes only that credible, informed opinion is divided, and that a decision-maker relying on headline investment figures as evidence of deliverability would be relying on a contested assumption.⁵⁰

A related and, so far as this appraisal has been able to establish, largely unexamined question is whether the headline investment figures already include the cost of the grid, transmission and water infrastructure needed to support the facilities they fund.

The evidence available suggests they generally do not.

National Grid's own material on connection reform notes sharply rising costs for transformers and overhead lines, and states an intention to protect existing and future customers from bearing the cost of building more network capacity than is needed, which implies that grid-reinforcement costs are substantially recovered through customer bills rather than from developers' own capital budgets.

If that is correct, the public and other electricity and water users, rather than the companies proposing the investment, may bear a material share of the infrastructure cost of accommodating it, potentially competing for funding with other infrastructure needs such as general grid decarbonisation or water-network renewal. This appraisal has not identified a published assessment that quantifies this effect, and the point is offered as a reasoned inference from the evidence available rather than an established fact.

Arising out of this, a potential question, which could be tabled by a Member of the Senedd to the Cabinet Minister for Enterprise, Connectivity and Energy, or by a Member of Parliament to the Secretary of State for Energy Security and Net Zero, reads:

'What assessment has the [Welsh/UK] Government made of whether the capital expenditure announced by data-centre operators for facilities in [Wales/the UK] includes the cost of the grid, transmission and water infrastructure upgrades needed to support those facilities; and, if that cost is not included, what estimate has been made of who will bear it and whether it will compete with funding for other infrastructure priorities?'

No answer to a question in these terms has been identified in Hansard, the Senedd Record or via WhatDoTheyKnow to date.

Data Retention and the Question of Necessity

A further, more basic question about the scale of data-centre demand deserves attention: how much of it is driven by an apparent assumption, shared across much of the industry and among individual users, that data passing through the internet should be retained indefinitely rather than deleted once it is no longer needed?

UK data protection law already addresses this in principle.

The storage limitation principle in Article 5(1)(e) UK GDPR requires personal data to be kept for no longer than necessary, though it leaves organisations to set and review their own retention periods rather than fixing them in law, and enforcement action specifically targeting excessive retention appears rare; a Finnish data protection authority fine of €856,000 in 2024 for failing to define a retention period at all illustrates the kind of enforcement this appraisal has not identified a UK equivalent for.^51,52

Industry surveys have long suggested that a large proportion of stored data is never needed again. Veritas's widely cited 2016 'Global Databerg' report found that 85% of enterprise data surveyed was 'dark' (unclassified) or 'ROT' (redundant, obsolete or trivial); a similar 2018 survey found comparable results. This appraisal has not identified a large-scale study from 2023 to 2026 that refreshes those figures with current primary data, and they should be treated as an aging, vendor-commissioned claim rather than a current, independently verified fact.⁵³

However, the evidence available suggests that the recent surge in data-centre capacity is driven predominantly by something other than (useful) long-term data retention.

The International Energy Agency's 2025 analysis of energy demand from AI found that data storage accounts for only around 5% of data-centre electricity consumption, while AI-optimised computing hardware, used for training and running AI models rather than for storing data, is projected to grow electricity use by roughly 30% a year and now drives close to half of global data-centre electricity growth.

On this evidence, characterising current data-centre expansion as primarily a consequence of 'saving everything' would not be accurate; the more significant driver, at least on current trends, is AI training and inference compute. This is an inference drawn by this appraisal from the IEA's figures, not a conclusion the IEA itself states, and I flag it as such.⁵⁴

None of this makes data retention practice irrelevant to the wider question of necessity. A rule requiring organisations, and in particular the largest cloud and social-media operators, to demonstrate active use or a continuing lawful basis for data before it is retained beyond a defined period, rather than retaining it indefinitely by default, could plausibly reduce the storage component of data-centre demand, and would sit within the existing framework of UK data protection law rather than requiring wholly new legislation.

On the evidence available, however, such a rule should be understood as a modest and useful measure addressing one, currently secondary, driver of demand, rather than as an answer to the larger and faster-growing demand associated with AI compute addressed elsewhere in this appraisal.

Notes from the paper

1. 27. House of Commons Library, Data Centres: Planning Policy, Sustainability and Resilience, Research Briefing CBP-10315 (27 May 2026). https://commonslibrary.parliament.uk/research-briefings/cbp-10315/
2. 28. Department for Science, Innovation and Technology, Designation of UK Data Infrastructure as Critical National Infrastructure, Written Ministerial Statement HCWS89 (12 September 2024). https://questions-statements.parliament.uk/written-statements/detail/2024-09-12/HCWS89
3. 29. Department for Science, Innovation and Technology, Data Centres to be Given Massive Boost and Protections from Cyber Criminals and IT Blackouts (12 September 2024). https://www.gov.uk/government/news/data-centres-to-be-given-massive-boost-and-protections-from-cyber-criminals-and-it-blackouts
4. 30. House of Commons Library, Data Centres: Planning Policy, Sustainability and Resilience, Research Briefing CBP-10315 (27 May 2026). https://commonslibrary.parliament.uk/research-briefings/cbp-10315/
5. 31. House of Commons Library, Data Centres: Planning Policy, Sustainability and Resilience, Research Briefing CBP-10315 (27 May 2026). https://commonslibrary.parliament.uk/research-briefings/cbp-10315/
6. 32. Senedd Research, Data Centres in Wales: How Will Wales Manage a Rapidly Growing Industry? (July 2026). https://research.senedd.wales/research-articles/data-centres-in-wales-how-will-wales-manage-a-rapidly-growing-industry/
7. 33. DataCenterDynamics, Silicon Valleys: How Wales Aims to Become a Data Center Hub (analysis, 2026). https://www.datacenterdynamics.com/en/analysis/silicon-valleys-how-wales-became-a-data-center-hub/
8. 34. European Powell (Substack), Consent Not Required: How Big Tech Is Rewriting British Law to Build Its Data Empire Beyond Democratic Reach (2026) — an independent critical commentary; cited here for its characterisation of disputed job-creation claims, not as a neutral primary source. https://europeanpowell.substack.com/p/consent-not-required-how-big-tech
9. 35. Data Centre Review, The Data Centre Boom Will Stall Without a Skills Plan (April 2026). https://datacentrereview.com/2026/04/the-data-centre-boom-will-stall-without-a-skills-plan/
10. 36. Ibid.
11. 37. Finance Act 2020, Part 2 (ss.39-72), especially s.43(2), defining "digital services activity" for Digital Services Tax purposes as limited to social media services, internet search engines and online marketplaces. https://www.legislation.gov.uk/ukpga/2020/14/part/2
12. 38. House of Commons Library, Digital Services Tax, CBP-8719 (2026). https://commonslibrary.parliament.uk/research-briefings/cbp-8719/
13. 39. House of Commons Committee of Public Accounts, Large Business Tax Compliance, HC 86 (10 July 2026). https://publications.parliament.uk/pa/cm5902/cmselect/cmpubacc/86/report.html
14. 40. Ricketts (Valuation Officer) v Cyxtera Technology UK Ltd [2021] UKUT 0265 (LC). https://landschamber.decisions.tribunals.gov.uk/judgmentfiles/j1755/LC-2019-611.pdf
15. 41. Browne Jacobson, Data Centres and Business Rates: What Investors Miss (2026). https://www.brownejacobson.com/insights/data-centres-and-business-rates-tax-on-megawatts
16. 42. Senedd Research, Data Centres in Wales: How Will Wales Manage a Rapidly Growing Industry? (2026). https://research.senedd.wales/research-articles/data-centres-in-wales-how-will-wales-manage-a-rapidly-growing-industry/
17. 43. Tom's Hardware, 'Big Tech's AI Spending Plans Reach $725 Billion in 2026' (30 April 2026), reporting FT-compiled combined 2026 capital-expenditure guidance from Microsoft, Alphabet, Meta and Amazon, up from around $410 billion in 2025. https://www.tomshardware.com/tech-industry/big-tech/big-techs-ai-spending-plans-reach-725-billion
18. 44. CNBC, 'Alphabet (GOOGL) Q1 2026 Earnings Report' (29 April 2026), reporting 2026 capital expenditure guidance of up to approximately $190 billion. https://www.cnbc.com/2026/04/29/alphabet-googl-q1-2026-earnings.html
19. 45. CNBC, 'Microsoft (MSFT) Q3 FY2026 Earnings Report 2026' (29 April 2026), reporting 2026 capital expenditure guidance of approximately $190 billion. https://www.cnbc.com/2026/04/29/microsoft-msft-q3-earnings-report-2026.html
20. 46. MLQ News, 'Amazon Raises 2026 AI Capex to $220B, Says Capacity Won't Meet Demand Through 2027' (30 July 2026). https://mlq.ai/news/amazon-raises-2026-ai-capex-to-220b-says-capacity-wont-meet-demand-through-2027/
21. 47. Data Center Dynamics, 'OpenAI Announces Five More US Stargate Data Centers with Oracle and SoftBank' (24 September 2025), reporting committed Stargate sites of almost 7GW and over $400 billion of investment, against the original $500 billion, four-year headline announced in January 2025. https://www.datacenterdynamics.com/en/news/openai-announces-five-more-us-stargate-data-centers-with-oracle-and-softbank/
22. 48. Microsoft, 'Microsoft's $15.2 Billion USD Investment in the UAE' (3 November 2025); see also CNBC, 'Middle East War Testing Gulf's Ambitions to Become AI Hub' (24 May 2026), on the wider scale and fragility of Gulf-region AI infrastructure investment. https://blogs.microsoft.com/on-the-issues/2025/11/03/microsofts-15-2-billion-usd-investment-in-the-uae/
23. 49. Senedd Research (Welsh Parliament), 'Data Centres in Wales: How Will Wales Manage a Rapidly Growing Industry?' (2026), citing an estimated £10 billion in South Wales AI Growth Zone investment (with up to £100 billion suggested across all UK AI Growth Zones) and a UK grid connection queue of around 173 requests totalling over 100GW as of May 2026. https://research.senedd.wales/research-articles/data-centres-in-wales-how-will-wales-manage-a-rapidly-growing-industry/
24. 50. Sequoia Capital, David Cahn, 'AI's $600B Question' (20 June 2024, subsequently updated); see also CNBC, 'Bond Market Anxiety Is Growing Over AI Capex Budgets' (24 July 2026). These are analyst commentary and opinion, not independently verified fact, and are presented as such. https://www.sequoiacap.com/article/ais-600b-question/
25. 51. Information Commissioner's Office, 'Principle (e): Storage Limitation', UK GDPR guidance, requiring personal data to be kept 'for no longer than is necessary' but leaving retention periods to be set and reviewed by each organisation rather than fixing them in law. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-protection-principles/a-guide-to-the-data-protection-principles/storage-limitation/
26. 52. European Data Protection Board, 'Finnish SA: Administrative Fine of EUR 856,000 for Failing to Define a Storage Period' (2024), an example of enforcement action against a failure to observe the storage-limitation principle; no directly comparable UK enforcement action penalising over-retention alone has been identified for the purposes of this appraisal. https://www.edpb.europa.eu/news/national-news/2024/finnish-sa-administrative-fine-eu-856000-failing-define-storage-period_pl
27. 53. Veritas, 'Global Databerg Report' (15 March 2016), a widely cited industry survey finding that 85% of data stored by organisations surveyed was 'dark' (unclassified) or 'ROT' (redundant, obsolete or trivial); a comparable 2018 Veritas survey of UAE organisations found 81% unclassified. No large-scale study from 2023-2026 superseding these figures with fresh primary data has been identified, and the figures should be treated as an aging, vendor-commissioned survey claim rather than independently verified current fact. https://www.veritas.com/news-releases/2016-03-15-veritas-global-databerg-report-finds-85-percent-of-stored-data
28. 54. International Energy Agency, 'Energy Demand from AI', in Energy and AI (2025), finding that storage systems account for only around 5% of data-centre electricity consumption, while AI-optimised 'accelerated servers' are projected to grow electricity use by roughly 30% a year, against approximately 9% a year for conventional servers, and now drive close to half of global data-centre electricity growth. This suggests that current data-centre capacity growth is driven predominantly by AI training and inference compute rather than by the long-term retention of data already collected, an inference drawn by this appraisal from the IEA's figures rather than a conclusion stated by the IEA itself. https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai