Summary

  • Great Britain's contracted demand queue rose from 41GW in November 2024 to 125GW in June 2025. Ofgem's consultation identifies around 73GW across roughly 315 data-centre projects, compared with 45GW of peak national demand in 2025.
  • Ofgem is consulting, not announcing a final rule. For relevant projects at or above 40MW, it proposes security of £237,500 to £712,500 per requested MW from connection-offer acceptance to energisation.
  • A 100MW project would secure £23.8m to £71.3m. Cash is only the most onerous form: the proposal also allows a qualified-bank letter of credit or performance bond, or a qualified-company bond.
  • The parallel milestones would test pathway and non-binding compute demand at M0.5.Dc, long-lead electrical procurement at M2.Dc, and finance, design capability and—on a lease or sale path—binding compute demand at M6.Dc.
  • The fee prices the right to remain in the queue; it does not prove land, power, customers or capital will become an operating data centre. Investment committees should value it as a wasting option with explicit stop and reprice triggers.

“Contracted demand” is Ofgem’s label for demand sitting behind connection offers or agreements. It does not mean the project already has binding compute offtake or consumes that electricity.

One hundred megawatts of British grid capacity may soon require a developer to support as much as £71.3m of security. That is before the servers, cooling plant, building and revenue arrive. The cheque is not necessarily cash, and the policy is not yet law, but it changes the economics of saying, “we have a grid connection.”

The phrase currently covers too many states. It can mean an application, an accepted offer, a queue position, an agreed design, a project with ordered switchgear, a site with a compute customer, a building under construction or an energised facility. These states are not interchangeable. PowerPoint has been known to survive the distinction. Cash flow is less forgiving.

Ofgem opened its Curate consultation on 29 July 2026 and will accept responses until 16 September. The proposal addresses a remarkable queue. Contracted demand increased from 41GW in November 2024 to 125GW in June 2025. Ofgem's underlying dataset attributes around 73GW to approximately 315 data-centre projects, ranging from 1MW to 1,500MW. For scale, Great Britain's peak demand in 2025 was 45GW.

The numbers do not prove that 73GW is fraudulent, duplicated or impossible. Public evidence does not disclose enough project-level ownership, customers, finance or readiness to make that accusation. They do show that queue capacity is not a load forecast. Even if every application is submitted lawfully, the system cannot treat every requested megawatt as equally likely, equally timed or equally useful for network investment.

The proposed fee is an option price, not a grid auction

Ofgem proposes applying the commitment fee to data centres at or above 40MW that connect to the transmission system or have a transmission impact. It says that threshold captures 99% of the capacity in its 73GW data-centre dataset. The cliff is therefore administratively powerful. It also creates an obvious incentive to split or describe projects below the line, a risk Ofgem itself acknowledges.

The proposed formula is linear: requested MW multiplied by assumed average project capital expenditure per MW, multiplied by a selected percentage. Ofgem uses £9.5m/MW as the average capital-cost assumption and consults on 2.5% to 7.5%. The resulting range is £237,500 to £712,500/MW.

This is not a market price for scarce network capacity. Ofgem explicitly says it cannot administratively value the external harm created by speculative queue holdings or the market value of the capacity in this phase. Nor is it trying to reduce the queue to a predetermined number. The fee is calibrated as a deterrent: high enough to make a weak project reconsider, but low enough, in the regulator's view, not to derail a viable one.

Economically, that makes the fee the carrying cost and downside of an option. A compliant project gets the security back at energisation. A project that terminates, reduces capacity or fails the relevant rules can lose it.

The existing securities framework would permit a bank letter of credit, a performance bond from a qualified bank or company, or a cash deposit. A letter of credit reduces the cash drag but consumes bank line and credit capacity. A parent guarantee is not air either; it moves contingent risk onto a balance sheet.

Ofgem's hypothetical model makes the time value visible. It assumes a 100MW project, £9.5m/MW of capex, a 10.95% weighted average cost of capital, two years of construction and an unlevered pre-tax IRR of about 14.7% before the fee. With the least onerous 2.5% cash case and a 2029 connection, the modelled IRR falls by 20 basis points. With 7.5% held until a 2035 connection, it falls by 170 basis points, to 13.0%.

Those are scenario outputs, not observed returns. The regulator does not know each project's hurdle rate, capital structure or instrument cost. The useful lesson is narrower: a distant connection date makes refundable security expensive because capital remains tied to a promise rather than an operating asset.

Three milestones try to make evidence arrive before concrete

The commitment fee is only one half of Curate. Ofgem also proposes data-centre-specific milestones in addition to the existing connection milestones.

At M0.5.Dc, a project would choose a self-operation or lease/sale pathway and provide non-binding compute-offtaker evidence, such as heads of terms or a memorandum of understanding. At M2.Dc, it would show a material commitment to long-lead electrical equipment. The consultation mentions transformers and switchgear but avoids prescribing one universal configuration. At M6.Dc, the developer would show financial capability and project-specific technical capability. Lease or sale projects would also provide binding compute-offtaker evidence.

Ofgem is minded to require a binding customer contract covering at least 20% of compute capacity for at least one year. That is a proposal within the proposal. It may be enough to distinguish a real commercial route from a land option. It may also be a slender anchor for financing the other 80%.

There is a circularity problem. A customer may refuse to bind capacity until the grid date is credible; the network process may demand customer evidence before it treats the project as credible. Incumbent hyperscalers with captive demand and established developers with repeat customers can cross that circle more easily than a new entrant. Stronger evidence can improve queue quality while also raising the cost of entry. Both can be true.

The proposed technical evidence has the same boundary. Project-specific EN 50600, TIA-942 or Uptime Institute design certification can demonstrate that independent eyes have reviewed a mature design. It cannot guarantee energisation, construction, customer acceptance or profit. A certificate is evidence of a state. It is not permission to skip the remaining states.

Curate, Plan and Connect distribute different powers

Ofgem divides demand reform into three pillars. Curate filters queue membership. Connect considers new contractual and operating arrangements to accelerate physical connection while maintaining a secure system. Plan is led by government and can prioritise strategically important demand.

That separation matters because each pillar answers a different question. Is the applicant progressing? Can the system physically connect it? Should a project receive policy priority?

Government's AI Growth Zone policy proposes removing speculative demand, reallocating released capacity and reserving future capacity for strategically important projects. Strategic reservation may speed projects government regards as nationally useful. It also changes the allocation rule. A queue filtered by evidence is different from a queue reordered by policy.

The government has authority to set strategy through law and accountable policy. The word “strategic” is not itself authorisation. A legitimate reservation process needs published criteria, identifiable cost bearers, review and evidence that the selected project can deliver the promised public result. Otherwise optionality has not disappeared; it has merely acquired a ministerial adjective.

The capital decision begins with six evidence states

A board should not ask whether a project “has power”. It should ask which evidence state has been reached:

  1. A connection application has been made.
  2. A Gate 2 offer has been accepted and the required security can be supported.
  3. The connection design and network works have dated, costed responsibilities.
  4. Long-lead equipment has been ordered on terms aligned to those dates.
  5. Compute demand and finance survive delay and cost sensitivities.
  6. The site is energised, commissioned, accepted and billable.

Each transition consumes capital and reduces flexibility. Land is location-specific. Transformers and switchgear are not infinitely portable. A customer contract may contain deadlines, remedies and escape clauses. Political announcements create local expectations. Debt terms can outlive the schedule that justified them.

The relevant comparator is therefore not a free queue place. It is the next credible route to powered capacity: an energised brownfield facility, a less constrained grid zone, a modular first phase, a lawful flexible connection, or no project. Brownfield power will look expensive. So will six years of security carry attached to equipment and land that cannot earn revenue.

The editorial inference for number-resource holders and network operators is familiar. An ASN, prefix, registry record or announced capacity can describe an intended system; it does not create a running one. Logical coordination remains essential, but it cannot energise a rack. Global Internet-governance institutions should report this dependency plainly rather than turn applications, memberships or official plans into evidence of operational continuity. This is an analogy about evidence, not a claim that number-resource governance and electricity law are one control surface.

Sources