One of 325 local authorities has an average half-hourly meter as big as it was in 2013: Slough

What these figures are, and are not

• 2024 figure. A large reclassification break appears in the 2024 half-hourly data: half-hourly meters rose 23.6 per cent in one year, or 23.4 per cent with the Unallocated row left out, while total non-domestic meters rose 1.3 per cent, and 344 of the 349 local authorities with at least 100 half-hourly meters in 2023 gained more than 5 per cent. No explanation for it appears in the documents read here. It is not Market-wide Half Hourly Settlement: Ofgem's CR055 decision puts MHHS meter migration between late October 2025 and May 2027, after this data year. Any 2024 figure on this page sits on a meter base that earlier years do not share.
• Settlement migration, not demand. BSC modification P272, with P322 alongside it, moved profile classes 5 to 8 into half-hourly settlement between 5 November 2015 and 1 April 2017. Ofgem states the condition: 'These changes only apply to non-domestic businesses that are in profile classes 5-8 and have an advanced meter installed.' The migration was therefore partial, and P322 keeps moving meters after 2017 as businesses change supplier or contract. Half-hourly meters in Great Britain went from 127,478 in 2015 to 265,975 in 2017 while the average per meter fell from 0.9916 to 0.5093 GWh. That fall is reclassification, not a change in electricity use.
• Measuring per meter reduces that distortion but does not remove it: the meters that moved in are smaller than the ones already there, so the per-meter mean is pushed down wherever migration happened, which is almost everywhere. The 2019 to 2023 test is reported alongside for exactly this reason: it sits between the two documented settlement breaks. Even that window is not still — half-hourly meters rose 8.5 per cent across it — so it is the quietest window in the data rather than a clean one.
• A 'large site' here is a meter, not a building and not a business. DESNZ splits domestic from non-domestic by settlement profile class, with profile 1 and 2 meters reallocated to non-domestic above 100,000 kWh a year, or above 50,000 kWh with supporting address information. A site with several meters counts several times; a single meter serving several businesses counts once.
• Half-hourly data covers January to December. Non-half-hourly data covers February to January; the 2024 non-half-hourly year is February 2024 to January 2025. The two are never added here, and the all-non-domestic series used for the migration-immune cross-check contains both, so it mixes the two periods as well.
• The data-centre share of large-site electricity divides an all-meter numerator by a half-hourly denominator. It is an approximation and is labelled as one wherever it appears.

All 18 caveats are listed in full in the method section, and in the README that ships with the dataset.

What the DESNZ data show

Of the 325 local authorities in Great Britain with more than 100 half-hourly meters in 2013 and still present in 2024, one has an average half-hourly meter drawing as much electricity now as it did then: Slough, 1.8892 GWh against 1.8878 GWh. The median authority is at 29 per cent, and Great Britain's own mean fell to a ratio of 0.301 over the same period, so the median authority roughly tracks the national reclassification. Between 2019 and 2023, 326 of 350 local authorities used less large-site electricity and 24 used more, and the ten largest risers account for 94.6 per cent of the 1,093.9 GWh of increases, with falls not netted off.

What was already known

Two things in this study have been published before, and both should be credited before anything new is claimed. DESNZ itself publishes the finding that data centres accounted for around two-thirds (65 per cent) of the electricity taken from the grid in Slough in 2024, in its June 2026 Energy Trends special feature. Utility Bidder published a local authority league table with a per-meter intensity measure in February 2026, comparing 2020 with 2024 on a per non-domestic meter basis, and put Slough at the top of it. That description of their work is not checked: a request with our identifying user agent returned HTTP 403 and we did not disguise it or retry, so what we know of their page came back through a general-purpose fetch tool instead. No figure of theirs is reproduced here and no data from them is used.

What this study adds

• A different measure on a different base: consumption per half-hourly (large-site) meter, not per non-domestic meter, over 2013 to 2024 rather than 2020 to 2024, on the set of authorities that had a large-site base in 2013.
• The settlement-migration artefact measured rather than assumed away, with a second test over a window that contains neither of the two documented settlement breaks.
• Every riser checked against the all-non-domestic series, which reclassification cannot move.
• The data-centre overlay recomputed against a large-site denominator instead of the all-electricity denominator, homes included, that DESNZ uses.

Not claimed. We do not claim to be first to notice that Slough's electricity consumption is rising, or that almost every local authority's is falling.

Scroll the table sideways to see every column →

The artefact that breaks the naive comparison

DESNZ classifies a meter as half-hourly or not by its settlement profile class, not by what the site does. Two changes moved large numbers of meters across that line without changing a single unit of electricity used. The first is documented: BSC modification P272, with P322 alongside it, moved profile classes 5 to 8 that had an advanced meter installed into half-hourly settlement between November 2015 and April 2017, and it is the larger of the two by a wide margin — half-hourly meter numbers more than doubled, rising 108.6 per cent in those two years. A second break appears in 2024. It is smaller, at 23.41 per cent in one year, but it is the largest single-year jump outside the P272 window, and no explanation for it appears in the documents we read. The printed national series carries a third movement as well, before that window: half-hourly meters rise from 121,327 in 2013 to 145,104 in 2014 and fall back to 127,478 in 2015, across the change of geography basis. It is smaller than either break, no explanation for it appears in the documents we read either, and no finding here is measured on it — but the 2013 to 2024 comparison runs across it, which is one more reason the 2019 to 2023 window is reported alongside.

Because a local authority total moves when meters are reclassified into it and a per-meter mean is far less sensitive to that. Slough's half-hourly meter count went from 383 in 2013 to 918 in 2024. Those are not 535 new sites.

Migration still drags the per-meter mean down wherever it happened, because the meters that moved in were smaller than the ones already there. That is why the median ratio is 0.29 rather than something near 1, and why the interesting question is which authority stayed at 1 despite it. DESNZ makes the same point about its own statistics: percentage changes in total consumption stay reliable, but meter counts and per-meter means can be distorted by meters crossing a classification threshold.

Scroll the table sideways to see every column →

The two breaks marked on the chart are settlement changes, not changes in electricity use. Over the P272 window, 337 local authorities were tested and 0 finished at or above parity; the median per-meter ratio was 0.5272. Between 2023 and 2024, half-hourly meters rose 23.41% on a denominator of half-hourly meters in Great Britain in 2023, excluding the Unallocated row — the caveats quote 23.61%, which is the same rise with the Unallocated row left in — and the median per-meter ratio was 0.7941. No explanation for it appears in the 2024 summary report or on the cover sheets of the three workbooks used here; we did not obtain the current subnational energy statistics methodology note, which the MSOA cover sheet points readers to for discrepancies of this kind, so it may be explained somewhere we did not read. What can be ruled out is Market-wide Half Hourly Settlement: Ofgem's CR055 decision of 29 November 2024 puts MHHS meter migration between late October 2025 and May 2027, entirely after this data year.

The large-site intensity test

Take every local authority that had a meaningful large-site base in 2013, defined as more than 100 half-hourly meters, and compare the average electricity per half-hourly meter in 2024 with 2013. The test is a ratio of published means, so each authority is its own denominator.

Exactly one local authority in the intensity base had an average half-hourly meter drawing at least as much electricity in 2024 as in 2013: Slough. The base is 325 local authorities: 350 had more than 100 half-hourly meters in 2013, and 25 of those have no 2024 counterpart. Local government reorganisation. Authorities that were merged or replaced between the 2011 Census geography used to 2014 and the 2021 and 2022 geographies used from 2015 have no counterpart in 2024.

Scroll the table sideways to see every column →

The other end of the same ranking, for scale: Cheshire West and Chester at 17.63%, Pembrokeshire / Sir Benfro at 17.41%, Redcar and Cleveland at 16.76%, King's Lynn and West Norfolk at 15.40%, Wokingham at 15.11%. 8 of 325 authorities are at half their 2013 intensity or better; 11 are below a fifth.

The same test over a window with neither documented settlement break

The same test over 2019 to 2023, a window that sits after P272 and P322 completed their mandated migration in April 2017 and before the 2024 break, so it needs no argument about what either did. It is not a window with no meters moving at all: half-hourly meters in Great Britain rose 8.5 per cent across it, because P322 continues to move meters as businesses change supplier or contract. It is the quietest window available in this data, not a silent one. It is the same test with both documented reclassification events outside the window, so it does not depend on any judgement about what P272 or the 2024 break did. It is quieter than 2013 to 2024 but it is not perfectly still: half-hourly meters in Great Britain still rose 8.5 per cent across it, because P322 keeps moving meters as businesses change contract. This test gives 6 authorities at or above parity where the 2013 to 2024 test gives one, which is consistent with the longer window carrying far more reclassification rather than less real growth.

Scroll the table sideways to see every column →

Over 2019 to 2023, a window that contains neither documented settlement break, though P322 continued to move meters across it, 6 of 349 local authorities had an average half-hourly meter at or above its 2019 level, and Slough was the highest at 1.414. The median across those 349 authorities is 0.8265. Recompute: Count rows where intensity_ratio_2019_2023_clean_window >= 1.

Does the answer depend on how authorities are matched across years?

The number of authorities in the base depends on how two authorities in different years are judged to be the same authority. All three rules give the same answer to the question asked.

Scroll the table sideways to see every column →

303 is the number produced by matching on the published name with no normalisation. DESNZ writes the twenty-two Welsh authorities monolingually in 2013 and bilingually in 2024 ('Cardiff' against 'Cardiff / Caerdydd'), so literal name matching drops all twenty-two for a spelling reason. Matching on the published code drops Barnsley and Sheffield instead, which were recoded in 2016. The reconciled basis keeps all of them and gives n=325. The finding is the same on all three: exactly one authority at or above parity, and a median of 0.29.

Where the growth is, and whether it is real

Over 2019 to 2023, chosen because it sits after P272 finished and before the 2024 break, the picture is a broad decline with a handful of places going the other way, and almost all of the growth sits in ten of them: 326 of 350 local authorities used less, 24 used more, and the ten largest risers account for 94.6 per cent of the 1,093.9 GWh of increases. Because the half-hourly series can be moved by reclassification, every riser is shown here on the all-non-domestic series as well, which a reclassified meter stays inside.

Scroll the table sideways to see every column →

Falls are not netted off either share: the denominators are the sum of the increases of the 24 local authorities that used more, 1,093.9 GWh; falls are not netted off and the sum of the increases of the 13 local authorities that used more, 941.7 GWh; falls are not netted off. Only 13 local authorities used more on this series. That is above the ten-case floor this study sets for reporting a percentage, but it is a small group and the percentage should be read with that in mind. The two series are not comparable in level and must not be added: half-hourly covers January to December and non-half-hourly February to January, and the all-non-domestic figure contains both.

Scroll the table sideways to see every column →

Read the last column first. Wiltshire gained 57.3 GWh of half-hourly load but only 21.5 GWh on the migration-immune series, a factor of 2.7. That gap is a warning sign, and it is why Wiltshire's rise is not treated here as a like-for-like large-site increase. Slough's two figures, 483.7 and 480.0 GWh, and Hillingdon's, 196.1 and 183.3, track each other closely.

The denominator that changes the answer

DESNZ expresses data-centre consumption as a share of all electricity taken from the grid in an area, homes included. That is stated on its own cover sheet. For a question about industrial load it is the wrong denominator: 39.1 per cent of the Great Britain total is domestic. How much difference that makes depends on the area: on the domestic share, which is not uniform — 21.5 per cent in Hillingdon, 18.4 per cent in Tower Hamlets and only 10.0 per cent in Slough — and on how large the published share already is. Recomputed against large-site electricity, the four shares that can be recomputed here, Great Britain and the three published local authorities, rise by between 1.88 and 10.22 percentage points. That recomputation is an approximation: an all-meter numerator over a half-hourly denominator, covering slightly different twelve-month periods. The national figure roughly doubles; Slough's moves by about a seventh, because where data centres already take most of the electricity the two denominators must converge.

Scroll the table sideways to see every column →

DESNZ publishes eighteen areas and they mix levels: Great Britain, combined regions, regions, residuals such as 'the rest of Outer London', two county-level areas (Hampshire and Greater Manchester) and three local authorities. They must not be summed. Only the three local authorities can be given a large-site denominator from the half-hourly panel, so the overlay can be attached to three of the 325 authorities in the intensity base and no more. Hampshire and Greater Manchester have no single row in the local authority panel, so neither is given a large-site denominator here. Three cases. No percentage is computed across them, and none should be. Each figure below is a ratio for one place. The 3 authorities are Hillingdon, Slough, Tower Hamlets. DESNZ's data-centre statistics cover 239 operational sites in Great Britain in 2024 and count only colocation, managed hosting and hyperscale sites serving external organisations. Enterprise data centres are excluded, so the scope exclusion pushes the true share above what the overlay shows. A second bias runs the other way: an all-meter numerator sits over a half-hourly denominator, which inflates the figure the overlay reports. Neither has been quantified, so the net direction of error is not determinable from published data. In Slough, data centres were 74.6 per cent of large-site electricity in 2024 — an approximation, an all-meter numerator over a half-hourly denominator — against the 65 per cent of all electricity DESNZ publishes.

Great Britain, 2020 to 2024

Scroll the table sideways to see every column →

Approximation: an all-meter numerator over a half-hourly denominator, covering slightly different twelve-month periods. The 2020 to 2024 series below is not a like-for-like trend at its last point. Its denominator is large-site electricity, and the half-hourly meter base jumped 23.41 per cent in 2024 while half-hourly electricity barely moved (121.380 to 121.522 TWh). The 2024 row therefore sits on a different meter basis from 2020 to 2023, and the rise from 2.655 to 3.670 per cent should be read with that in mind rather than as four clean years of growth. Two biases run in opposite directions and neither has been quantified, so the net direction of error is not determinable from published data. Scope pushes the true share up: DESNZ counts only colocation, managed hosting and hyperscale data centres serving external organisations, so enterprise data centres are out of scope, as are some smaller in-scope sites and 17 of the 259 identified sites to which no meter could be matched. The construction of the ratio pushes the published figure up: an all-meter numerator sits over a half-hourly denominator, which inflates the share by whatever part of the data-centre load is not on a half-hourly meter. Both are stated; neither is netted off, because nobody has published the size of either. DESNZ classifies the data-centre statistics as Official Statistics in Development, which in its own words carry a higher degree of uncertainty than Official Statistics. 2024 figure. A large reclassification break appears in the 2024 half-hourly data: half-hourly meters rose 23.6 per cent in one year, or 23.4 per cent with the Unallocated row left out, while total non-domestic meters rose 1.3 per cent, and 344 of the 349 local authorities with at least 100 half-hourly meters in 2023 gained more than 5 per cent. No explanation for it appears in the documents read here. It is not Market-wide Half Hourly Settlement: Ofgem's CR055 decision puts MHHS meter migration between late October 2025 and May 2027, after this data year. Any 2024 figure on this page sits on a meter base that earlier years do not share.

The warning on the front of the file, what it governs, and the one sum we make anyway

The MSOA workbook's cover sheet says, in terms: 'These statistics should not be aggregated to local authority, regional or country level'. It is a fair warning and it is worth answering rather than ignoring. Two sentences on the same cover sheet answer it. 'While these statistics are presented at MSOA level for non-half hourly meters, the data on half hourly meters (which is far more disclosive) is only broken down by local authority.' The half-hourly rows carry the MSOA code 'All MSOAs' and a local authority name, because DESNZ publishes them at local authority level itself. This study reads only those rows. It never builds a local authority total out of MSOA rows, which is what the warning is about. The subnational workbook, the local authority product the warning redirects users to, carries no such warning, and it is the series every riser is ranked on here. The warning does name country level as well, and one country-level sum is made here, so it is set out below with what was and was not checked about it.

In 2024 the Unallocated row holds 4,132 half-hourly meters and 2.6826 TWh. It is excluded from every local authority figure and from the 121.5 TWh national large-site denominator, and the full year-by-year series is in the JSON.

5 disagreements this study reports rather than resolves

Data-centre electricity is not a settled number. Where published estimates conflict, the conflict is reported and the denominator is named, rather than one figure being picked and the rest dropped.

Scroll the table sideways to see every column →

What was measured, when, and how

Three things. First, electricity per half-hourly (large-site) meter by local authority, 2013 against 2024, as a ratio, over authorities with more than 100 half-hourly meters in 2013. Second, the direction and concentration of change in large-site electricity by local authority over 2019 to 2023, computed twice: once on the half-hourly series and once on the all-non-domestic series that reclassification cannot move. Third, data-centre electricity as a share of large-site electricity, where DESNZ publishes it as a share of all electricity including homes.

Source files downloaded 2026-09-20T06:48Z. The half-hourly and subnational workbooks are the 18 December 2025 editions, covering data years to 2024; the data-centre tables are the 30 June 2026 edition, covering 2020 to 2024. Analysis computed 2026-09-20.

The three denominators in use

Scroll the table sideways to see every column →

Why 2013? P272, with P322 alongside it, moved meters into half-hourly settlement between 5 November 2015 and 1 April 2017, so 2013 and 2014 are both before settlement migration. 2013 is used rather than 2014 because the printed national series moves sharply between them: half-hourly meters rise from 121,327 in 2013 to 145,104 in 2014 and fall back to 127,478 in 2015, across the change of geography basis, so 2014 is the noisier of the two base years. 2013 is the last quiet year in which the half-hourly category meant what it had meant for the previous decade. Why more than 100 meters? Below about a hundred half-hourly meters a single large site moves the mean, and a single reclassification moves it further. The threshold is a judgement, stated here so it can be varied: it excludes 21 of the 371 authorities present in 2013.

Every headline figure, with its denominator and how to recompute it

Scroll the table sideways to see every column →

Where the plan was wrong, and what was done about it

Scroll the table sideways to see every column →

All 18 caveats

1. 2024 figure. A large reclassification break appears in the 2024 half-hourly data: half-hourly meters rose 23.6 per cent in one year, or 23.4 per cent with the Unallocated row left out, while total non-domestic meters rose 1.3 per cent, and 344 of the 349 local authorities with at least 100 half-hourly meters in 2023 gained more than 5 per cent. No explanation for it appears in the documents read here. It is not Market-wide Half Hourly Settlement: Ofgem's CR055 decision puts MHHS meter migration between late October 2025 and May 2027, after this data year. Any 2024 figure on this page sits on a meter base that earlier years do not share.
2. Settlement migration, not demand. BSC modification P272, with P322 alongside it, moved profile classes 5 to 8 into half-hourly settlement between 5 November 2015 and 1 April 2017. Ofgem states the condition: 'These changes only apply to non-domestic businesses that are in profile classes 5-8 and have an advanced meter installed.' The migration was therefore partial, and P322 keeps moving meters after 2017 as businesses change supplier or contract. Half-hourly meters in Great Britain went from 127,478 in 2015 to 265,975 in 2017 while the average per meter fell from 0.9916 to 0.5093 GWh. That fall is reclassification, not a change in electricity use.
3. Measuring per meter reduces that distortion but does not remove it: the meters that moved in are smaller than the ones already there, so the per-meter mean is pushed down wherever migration happened, which is almost everywhere. The 2019 to 2023 test is reported alongside for exactly this reason: it sits between the two documented settlement breaks. Even that window is not still — half-hourly meters rose 8.5 per cent across it — so it is the quietest window in the data rather than a clean one.
4. A 'large site' here is a meter, not a building and not a business. DESNZ splits domestic from non-domestic by settlement profile class, with profile 1 and 2 meters reallocated to non-domestic above 100,000 kWh a year, or above 50,000 kWh with supporting address information. A site with several meters counts several times; a single meter serving several businesses counts once.
5. Half-hourly data covers January to December. Non-half-hourly data covers February to January; the 2024 non-half-hourly year is February 2024 to January 2025. The two are never added here, and the all-non-domestic series used for the migration-immune cross-check contains both, so it mixes the two periods as well.
6. The data-centre share of large-site electricity divides an all-meter numerator by a half-hourly denominator. It is an approximation and is labelled as one wherever it appears.
7. DESNZ's data-centre statistics cover 239 operational sites in Great Britain in 2024 and count only colocation, managed hosting and hyperscale sites serving external organisations. Enterprise data centres are excluded, so the scope exclusion pushes the true share above what the overlay shows. A second bias runs the other way: an all-meter numerator sits over a half-hourly denominator, which inflates the figure the overlay reports. Neither has been quantified, so the net direction of error is not determinable from published data.
8. DESNZ classifies the data-centre statistics as Official Statistics in Development, which in its own words carry a higher degree of uncertainty than Official Statistics.
9. Geography break at 2015. The 2010 to 2014 years use 2011 Census geographies and carry 370 to 371 local authorities; 2015 onwards uses 2021 (England and Wales) and 2022 (Scotland) geographies and carries exactly 350. No single local authority set spans 2010 to 2024, which is why no figure here is anchored on 2010.
10. Revisions. DESNZ revises 2015 onwards at each annual publication, as postcode lookups are updated, meter addresses corrected and anomalous year-on-year changes investigated. Years to 2014 are not revised on the same basis. These figures are the 18 December 2025 edition of the half-hourly and subnational workbooks and the 30 June 2026 edition of the data-centre tables. A later edition will not reproduce them exactly. One headline figure is thin enough for that to matter: Slough clears parity by 0.08 per cent, or 0.0014 GWh per meter, so a revision of a tenth of one per cent to its 2024 mean would move the parity count from one to none. Nothing else in the study turns on a margin that fine.
11. The number of local authorities in the intensity base depends on how an authority in one year is matched to an authority in another. Matching on the published name gives 303, because DESNZ writes twenty-two Welsh authorities bilingually in 2024 and monolingually in 2013. Matching on the published code gives 323, because Barnsley and Sheffield were recoded in 2016. The reconciled basis used here gives 325. The finding is the same on all three.
12. MSOAs with fewer than five meters are moved to an 'Unallocated' row for disclosure control. In 2024 that row holds 4,132 half-hourly meters and 2.6826 TWh. It is reported separately and is excluded from every local authority figure and from the national large-site denominator.
13. DESNZ's subnational data excludes Central Volume Allocation users and electricity generated and consumed on site. Some of the largest sites in the country may therefore sit outside the large-site denominator entirely.
14. Only three of the eighteen areas in the data-centre tables are local authorities. The data-centre overlay can be attached to three authorities and no more. There is no published per-authority data-centre figure for the rest.
15. Market-wide Half Hourly Settlement will migrate 33 million MPANs between late October 2025 and May 2027. It will contaminate the 2025, 2026 and 2027 data years far more severely than P272 did. Any comparison drawn through those years will need the same treatment this study gives P272, or worse.
16. The PDF text of the DESNZ special article runs superscript footnote markers into the preceding number, so '2423' is 242 with footnote 3 and '2594' is 259 with footnote 4. 242 matched plus 17 unmatched is 259 identified, and the published tables give 239 operational in 2024. Anyone re-extracting that PDF should not read those as 2,423 or 2,594.
17. We have not read the current edition of DESNZ's subnational energy statistics methodology note. We could not locate a December 2025 edition at a stable address, and nothing here depends on it: the defence of what is aggregated rests on the cover sheet of the workbook itself.
18. Nothing here identifies a site, an operator or a person. The unit throughout is a local authority.

Sources

Scroll the table sideways to see every column →

Computed 2026-09-20 from DESNZ statistical releases downloaded 2026-09-20T06:48Z. Half-hourly and subnational data are the 18 December 2025 editions; the data-centre tables are the 30 June 2026 edition. scripts/research/uk-large-site-electricity-data-centres-2026/analyse.py reads the three tidy panels in data/research/uk-large-site-electricity-data-centres-2026/ and writes all three published files. It performs no network access and is deterministic: running it twice on the same inputs produces byte-identical output. The panels themselves are rebuilt from the raw workbooks by build_panels.py in the same directory, and the raw workbooks carry recorded md5s. None. No email, request or notice of any kind was sent to anyone. Everything here comes from already-published statistics. None. The unit of analysis is a local authority. The source documents name authors, responsible statisticians and press contacts and carry departmental mailboxes; those were stripped from every derived text before anything was written, by scripts/research/uk-large-site-electricity-data-centres-2026/evidence_redact.py, which holds the names it strips as salted hashes so that no name is retained in the repository either. No individual is named in the dataset, the README, the page or any working file. Downloads: CSV, 378 rows and 47 columns · README and attribution · JSON.

What is new here. Two measures that we could not find published anywhere: electricity per half-hourly meter by local authority on a 2013 base with settlement migration handled, and the data-centre share of large-site rather than all electricity. DESNZ publishes the inputs and publishes the data-centre share against an all-electricity denominator, stated on its own cover sheet. Utility Bidder publishes a per non-domestic meter measure on a 2020 base. Neither measure here is either of those. Not that we are first to notice Slough, first to notice the national decline, or first to build a local authority league table. All three have been published.

Questions

Power, density and what a rack actually draws

More Servnet research · UK business e-waste statistics 2026 · Oil prices and the cost of IT