The Value of Connection

Grid Connection Auctions

Grid connections in Great Britain are allocated on a first-come-first-served basis at administrative cost. This has led to a tragedy of the commons, with no barrier to entry, gigawatts of speculative applications have smothered this critical system. The alternative being attempted of ‘first ready and needed’ first-served has led to increased complexity without fixing the driver of this crisis: handing out scarce resources at far below their true value. Timelines continue to slip with projects delayed at great private cost and at the expense of economic growth.

The energy department has released a consultation on accelerating connections for “strategic demand” (closes 15th April 2026). It proposes two things. First, tighter entry requirements for the demand queue: developers will need to demonstrate exclusive land access, development history, commercial offtake, and planning progress. Projects that cannot meet these tests will exit the queue, freeing capacity at their grid nodes. Second, ministerial designation of “strategically important” projects using powers under the Planning and Infrastructure Act 2025. Designated projects get reserved capacity ahead of need, priority access to freed slots, and priority in future queue rounds.

The consultation’s own diagnosis is correct: “the scarcity value associated with a place in the connections queue was not reflected in the very low private cost associated with obtaining a connection agreement.” The government already plans to clear the queue, which would create a pool of freed grid capacity at specific nodes. The question is what happens to it, the consultation proposes giving it to ministerially designated projects for free. But includes a question 13 that asks whether auctions should play a role.

This analysis argues the freed capacity is worth billions and should be auctioned. Without an auction, the government gives away capacity that developers would willingly pay £1.3m per MW per year accelerated to secure. With an auction, that revenue funds the grid reinforcement that creates more capacity, which funds further auctions, and keeps these costs off household bills.

The analysis finds a year of grid connection (either accelerated or not delayed) has an absolute value of £1.5-1.8m per MW. With 4-5 competing bidders, auction theory (Vickrey 1961) predicts a clearing price of 75% or £1.1-1.5m per MW per year, giving a central estimate of £1.3m per MW per year accelerated (or not delayed).

The government already plans to tighten entry requirements for the demand queue. Of the ~50 GW of data centre capacity in the transmission queue, only ~20 GW has reached Final Investment Decision (Ofgem, February 2026). The remaining ~30 GW has not committed capital and would likely exit when required to demonstrate financial commitment or bid real money for a connection slot. Not every freed slot produces a contestable auction. In the near term, a realistic estimate is that 1-3 GW of freed capacity at contested nodes would be auctioned, with real projects behind them gaining an average of 2-3 years of earlier connection. Each MW-year of grid connection, one megawatt for one year of operation, is worth £1.3m at auction. At 1 GW gaining 2 years, that is 2,000 MW-years: £2.6bn. At 3 GW gaining 3 years, that is 9,000 MW-years: £12bn.

This would clear the current backlog by removing speculative projects, allocate scarce resources to those who can create the most economic growth from it, and would generate billions in revenue that could fund household bills, further grow demand through network reinforcement, and invest in electricity supply chain acceleration.

Annex

1. Methodology

The goal is to calculate earnings per year per megawatt of connection, then estimate how much of that total value would be bid in an action, in terms of both price and volume.

1. Calculate earnings: EBITDA is taken from Companies House filings.

2. Estimate power: Since 2019 every large UK company must report total energy consumption in kilowatt-hours in its Directors’ Report under the Streamlined Energy and Carbon Reporting (SECR) regulations. To convert this into grid connection capacity:

   • SECR kWh to average facility MW. Divide annual kWh by 8,760 hours. This gives the average total facility power draw (IT load plus cooling, lighting, and all other systems) during the accounting period. What matters is total facility demand, not the fraction that reaches the servers. PUE (Power Usage Effectiveness) is irrelevant to this question.

   • Average facility MW to grid connection MW. A data center does not use 100% of its grid connection at all times. NESO’s Future Energy Scenarios 2025 applies a 70% load factor as the official GB planning assumption for data centers (cited in National Grid DSO / Regen, Data center Impact Study, November 2025, p.28). Dividing average facility power by 0.70 gives the implied grid connection capacity.

3. Divide Step 1 by Step 2. EBITDA per MW of grid connection is the annual profit foregone for each MW of connection delayed by one year.

4. Add carry cost: the financing cost of capital deployed ahead of grid energisation.

   • Capital deployed per MW of connection is taken from PP&E on the balance sheet divided by grid connection MW. The cost of capital is derived from interest payable in the filings, adjusted to market rates where parent loans are interest-free. This assumes the developer has built the facility ahead of grid power arriving, which is the observed market behaviour given 3-7 year connection timelines. Where construction has not yet started, only the EBITDA component (Step 3) applies.

5. Auction clearing price is the total delay cost (Step 4 + Step 5) multiplied by the Vickrey bid-shading factor.

   • In a first-price sealed-bid auction with n competing bidders, each bidder’s equilibrium strategy is to bid (n-1)/n of their true willingness to pay (Vickrey 1961). With 4 bidders: 75%. With 5: 80%.

6. Volume bid is the most debatable input, this paper does not consider optimal design and is only estimating the initial revenue from the first round of auctions, over a rolling programme, cumulative revenue would be substantially higher. NESO’s grid queue process identified approximately 13 GW of demand connectable before 2030 and 86 GW between 2030 and 2035. The auction would release connection capacity at specific grid nodes or reinforcement areas in defined time tranches. Developers bid for the right to connect at a given location by a given date. A developer who does not bid loses their position. The structure creates competition at many nodes, because multiple projects in the same grid area are bidding against each other for the same physical capacity.

2. Calculations

This analysis is anchored on two operators, Vantage DC UK and Equinix (UK) Limited, which are the only UK data center companies that meet all four evidence criteria: (1) full audited accounts filed at Companies House, (2) mature and fully utilised facilities, (3) SECR energy data allowing period-average MW to be derived, and (4) profitable operations with clean, readable economics. Eight additional operators were investigated and discounted for specific reasons documented in this report.

Exhibit 1a: Vantage DC UK

Cross-check on the load factor. Vantage’s SECR table also states IT load of 35.28 MW at year-end, with a prior year figure of 27 MW. Average IT load during the year was approximately 31 MW. Total average facility power was 46.6 MW. The ratio (46.6 / 31 = 1.50) is the implied PUE, within the normal range for a hyperscale facility. The 46.6 MW average draw against an implied 66.6 MW connection gives a 70% load factor, confirming consistency with the NESO assumption.

Exhibit 1b: Equinix (UK) Limited

Note on revenue accounting. Equinix reports revenue on a gross basis including power billed to customers. Cost of sales is 89% of revenue. The EBITDA includes margin on that power pass-through. This is the correct basis for valuing grid connection capacity, because the connection enables both the colocation service and the power delivery.

Exhibit 2: EBITDA/MW Comparison

Best estimate: £1.0-1.2m EBITDA per MW of grid connection per year. The two anchors are independent operators with different business models (Equinix: retail colocation and interconnection; Vantage: hyperscale wholesale) that converge in the same range.

Exhibit 3: Capital Intensity

Capital intensity converges at ~£11m per MW of grid connection across both operators. This aligns with industry benchmarks for data center construction cost (CBRE: £8-11m/MW).

Exhibit 4: Cost of Capital

Exhibit 5: Carry Cost per MW of Grid Connection per Year

Best estimate: ~£0.6m carry cost per MW of grid connection per year.

A note on timing and this assumption. The carry cost assumes that capital is deployed during the grid connection delay period. In a textbook development sequence, the developer would wait for the grid connection before building, and no carry cost would arise. In practice, current GB grid connection timelines of 3-7 years are forcing developers to build ahead of energisation. Facilities are being constructed on spec, with shells and fit-out completed while waiting for grid power to arrive. In this scenario, the developer genuinely has capital at risk earning nothing. This assumption is slightly bullish, but it reflects the actual behaviour observed in the market. Where a developer has not yet committed construction capital, only the EBITDA foregone component of the delay cost applies.

Exhibit 6: Total Cost of One Year’s Delay (per MW of grid connection)

Exhibit 7: From Delay Cost to Auction Clearing Price

In a standard independent private values auction, each bidder shades their bid below their true willingness to pay. In a first-price sealed-bid format, the equilibrium bid is (n−1)/n of the bidder’s true value, where n is the number of competing bidders (Vickrey 1961, revenue equivalence theorem). The seller’s expected revenue is (n−1)/(n+1) of the maximum possible value (Milgrom and Weber 1982).

The UK’s own spectrum auction history shows these dynamics at work. The UK 3G auction (2000, designed by Klemperer and Binmore, 13 bidders for 5 licences) raised £22.5bn, 6.4× the government’s £3.5bn forecast. Switzerland’s equivalent 3G auction (4 bidders for 4 licences) raised the equivalent of €20 per capita against the UK’s €650. Same spectrum, same underlying value, 30× difference in revenue. The variable was competition (Klemperer, “How (Not) to Run Auctions”, European Economic Review, 2002).

The Bulow-Klemperer (1996) result reinforces this: attracting one additional serious bidder generates more revenue than any optimisation of reserve prices or auction format.

Exhibit 8: Implied Auction Clearing Price by Competition Level

With 4-5 serious bidders per connection slot, the auction clears at £1.1-1.5m per MW per year. For context, the DESNZ consultation notes approximately 140 data center projects in the transmission queue. Even if only a fraction bid for any given connection slot, competition of 4-6 bidders per slot is plausible.

Exhibit 9: Sensitivity Matrix (Auction Clearing Price, £k/MW/year, at 4 bidders / 75%)

Carry rate held at 5.5%. Clearing price = 75% × (EBITDA + capital × 5.5%). 75% corresponds to 4 competing bidders.

Exhibit 10: Illustrative Auction Revenue

At the conservative end, 1 GW of data center capacity accelerated by two years at the central clearing price generates £2.6bn. This is capital that could fund network reinforcement, offsetting the network charge increases projected under the coming price controls.

For context: the UK 3G spectrum auction raised £22.5bn. The principle is identical. A scarce public resource, allocated without a price, will be claimed without value being shared with the public.

3. Cross-Checks

Exhibit 10: External Validation

Exhibit 11: Revealed Preference (Private Wire)

Developers spending £2-5m/MW in capital to bypass the grid queue have revealed their willingness to pay for timely power access.

4. Risk Factors

1. Average facility MW is the denominator, not IT load or design capacity. The MW figures are derived from SECR total energy consumption divided by 8,760 hours. This gives the average power drawn by the entire facility (IT, cooling, lighting, everything) during the accounting period. It does not measure peak demand, contracted connection size, or design capacity. The 70% load factor used to estimate grid connection capacity is NESO’s published planning assumption (FES 2025); actual load factors vary by site. Measured utilisation on distribution-connected sites averages 28-42% (NGED/Regen 2025), though this sample excludes hyperscale operators.

2. Two data points. The analysis rests on two operators. This reflects the state of UK public filings, not a choice to ignore data. Section 10 documents the ten other operators investigated and the specific reasons each was discounted. Both anchors independently converge at similar EBITDA per MW (£1.0-1.2m) and capital intensity (~£11m/MW), which strengthens confidence despite the small sample.

3. The carry cost assumes build-ahead-of-connection. If a developer waits for grid energisation before constructing, no carry cost arises during the delay period and only the EBITDA foregone component applies. The carry cost is included because current GB grid timelines (3-7 years) are forcing developers to build on spec, deploying capital before grid power arrives. This is a slightly bullish assumption. Where build-ahead has not occurred, the total delay cost falls to £1.0-1.2m/MW/year (EBITDA only) and the auction clearing price to £0.7-1.0m/MW/year at 4-5 bidders. Even at this lower figure, the auction clears 70-100× above the current Ofgem fee.

4. EBITDA includes power margin. Both operators’ EBITDA includes margin on electricity billed to customers. A grid connection enables both the colocation service and the power delivery, so this is the correct basis. But a developer evaluating a grid connection bid would need to consider their specific power margin, which varies by contract structure and wholesale electricity prices.

5. Auction design and competition determine outcome. The clearing price depends on the number of competing bidders per slot. With 4 bidders, theory predicts 75% of the winner’s delay cost; with 6, 83%. The UK’s own spectrum auction history proves the point: the 3G auction (13 bidders, 5 licences) raised €650 per capita; Switzerland’s equivalent (4 bidders, 4 licences) raised €20. Same spectrum, 30× difference. Bulow and Klemperer (1996) show that attracting one additional bidder generates more revenue than any optimisation of reserve prices. Auction format matters less than competitive tension. See also Binmore and Klemperer, “The Biggest Auction Ever” (Economic Journal, 2002) and Pollitt et al., “Managing the GB connection queue” (Utilities Policy, 2025).

6. Regional variation. Both anchor data points are from the South East and South Wales. Equinix operates in London; Vantage in Newport. Regional facilities elsewhere in GB would likely face lower lease rates, lower occupancy, and lower EBITDA per MW. A locational auction would produce different clearing prices by region.

7. The analysis values operational facilities, not development-stage projects. A developer bidding for a new connection faces construction risk, uncertain occupancy, and no existing revenue stream. The delay cost model assumes a facility ready to operate. Development-stage economics would produce a lower willingness to pay, but the margin is still large: even at half the central estimate, the auction price would be 50-75× the current Ofgem fee.

8. Tax treatment favours the developer. The delay cost in this analysis is expressed as EBITDA, which is pre-tax. In principle, corporation tax at 25% would reduce a developer’s after-tax willingness to pay. In practice, UK data center operators pay very little tax as the sector is heavily debt-financed.

5. Operators Investigated and Discounted

Eight additional operators were investigated as potential data points for this analysis. The financial data from these operators remains in the evidence base and is available for cross-reference, but none met all four criteria required for a primary anchor: (1) full audited Companies House accounts, (2) mature and fully utilised, (3) SECR energy data to derive period-average MW, and (4) profitable with clean economics.

1. Kao Data (CH 11756346, FY Mar 2025, audited by BDO)

Revenue £63.8m. EBITDA £4.9m. SECR energy 122 GWh at PUE 1.52, giving ~9.2 MW average IT load.

Reason discounted: Early-stage facility with very low utilisation. The 9.2 MW average draw is against a design capacity of 96 MW, under 10% utilised. EBITDA per MW of £533k reflects a half-empty campus, not steady-state economics. A developer bidding for a grid connection would bid on expected returns at full utilisation, not ramp-up performance.

2. CyrusOne UK5 (CH 11475681, FY Dec 2024, audited by Deloitte)

Revenue £38.1m. EBITDA ~£11.1m. Commissioned capacity 31 MW (22 MW phase 1 in 2022, 9 MW phase 2 in 2023).

Reason discounted: The 31 MW is commissioned design capacity stated in the strategic report, not measured actual IT load. No SECR energy data is disclosed in this filing. There is no way to verify how much of the 31 MW is actually drawing power. Also an early-stage operation with low margins (29% EBITDA margin vs 41% for Vantage), suggesting the facility is still filling up.

3. Colt DCS UK (CH 07306352, FY Dec 2024, audited by PwC)

Revenue £71.2m. EBITDA £29.8m. Two sites near London (Welwyn Garden City and West London).

Reason discounted: No MW figure of any kind appears in the filing. The word ‘MW’ does not appear anywhere in the accounts. The SECR section on page 8 of the Directors’ Report states that energy data is reported in the consolidated Group annual report (Colt Group Holdings Limited), not in this entity filing. This is a permitted subsidiary exemption.

4. Telehouse Europe (CH 02138407, FY Mar 2025, audited by PwC)

Revenue £257.0m. Operating profit £81.5m. London Docklands campus plus Paris.

Reason discounted: Three problems. First, the accounts are consolidated UK and France, so no UK-only revenue or energy figure is available. Second, the SECR methodology changed in this reporting period: customer electricity pass-through was reclassified from Scope 2 to Scope 3, causing reported energy to drop from 182 GWh to 71.5 GWh. The post-reclassification figure covers only Telehouse’s own consumption, not total facility power, making MW derivation unreliable. Third, the SECR covers calendar year 2024 while the revenue period ends March 2025, a 9-month mismatch. Only one site-level figure is available: Telehouse South at 7.7 MW current / 18 MW full build. Total London Docklands campus MW is not disclosed.

5. Global Switch (CH 06238341, FY Dec 2024, audited by Deloitte)

Revenue £436.6m. EBITDA £213.2m. 488 MVA utility power supply across all sites globally.

Reason discounted: The financial statements are non-statutory consolidated accounts filed with Euronext Dublin, not Companies House statutory accounts. Revenue and EBITDA cover the UK, Europe (Amsterdam, Frankfurt, Madrid, Paris), and Asia-Pacific (Hong Kong, Singapore). Asia-Pacific contributed 42% of revenue (£182.1m) and 47% of EBITDA (£101.1m) in 2024. No UK-only financial breakdown is published. The 488 MVA figure is utility supply power across all global sites, not IT load, and cannot be split by geography. Per-MW calculations are impossible without a matched UK revenue and UK MW figure.

6. Interxion Carrier Hotel Limited (CH 03753969, FY Dec 2024, audited by PwC)

Revenue £62.5m. Operating loss £4.1m. SECR energy 59.2 GWh.

Reason discounted: The operating loss is driven by £22.9m of intercompany management fees charged by the parent (Digital Realty Trust Inc.), a transfer pricing charge, not an operational deterioration (gross profit actually improved year-on-year). The economics are distorted by corporate structure. The SECR energy of 59.2 GWh implies only ~6.8 MW of total facility power, very small scale. No MW capacity or PUE is disclosed in the filing. The company flags going concern risk and relies on a parental support letter.

7. NTT Global Data Centers EMEA UK Ltd (CH 04239332, FY Mar 2024, audited by KPMG)

Revenue £86.1m. Operating loss £5.3m. SECR electricity 247.2 GWh. PP&E £227.3m.

Reason discounted: The company is in active restructuring. One of its two UK sites (Hemel Hempstead HH1) was fully decommissioned in September 2024, triggering a £4.9m impairment charge in the accounts. Revenue fell 19% year-on-year, driven by lower power pass-through rather than core colocation. The company is loss-making at the operating level and relies on NTT Ltd letters of support and intercompany refinancing (£105m of loans, some at 6.021%). The SECR energy of 247 GWh would imply ~28 MW of facility power, but one site was closing during the period. Not steady-state economics.

8. VIRTUS Data centers Ltd (CH 06762600, FY Dec 2024)

Major London hyperscale operator with 7+ facilities, owned by ST Telemedia / Macquarie.

Reason discounted: The operating entity files audit-exempt subsidiary accounts (16 pages), relying on a parent company guarantee. This abbreviated filing does not contain sufficient financial detail: no full P&L, no SECR energy data, no depreciation note. The parent holding company (VIRTUS UK Holdings Limited, CH 12148633) turned out to be an unrelated ticketing company (Vivaticket S.p.A.), not the data center business. The actual parent entity in the Macquarie ownership chain could not be identified from publicly available Companies House filings.

6. Sources