In Britain, the energy regulator Ofgem has proposed that large data center projects pay a substantial refundable fee to secure their place in the electricity connection queue.
The proposal is a response to a remarkable surge in speculative applications. Contracted demand connection offers increased from 41GW to 125GW in less than a year, with data centers accounting for at least 80GW of the total.
The numbers are striking. Britain’s peak electricity demand in 2025 was around 46GW, meaning the capacity requested by projects in the queue is almost three times that figure.
Of course, that does not mean Britain is about to build 125GW of data centers. Most of those projects will not happen. That, in fact, is the problem.
The details of how this one country manages the rapid expansion of data centers will differ from those of the USA, Ireland, Australia, or Germany. Still, the underlying challenge is increasingly universal: how do you connect a new generation of power-hungry digital infrastructure to grids that were not designed for it?
The queue has become part of the infrastructure
For a data center developer, securing a grid connection is potentially one of the most valuable assets in the entire project. A site can have land, fiber connectivity, cooling potential and a willing local authority, but without enough electricity at the right time, there is no data center.
That creates an obvious incentive to get into the queue early. It also creates a less obvious problem: the queue itself can become congested with projects that may never be built.
Ofgem says its proposals are intended to tackle “speculative projects” that secure scarce network capacity without a firm intention to connect.
Its solution is a new Data Centre Commitment Fee of between £237,500 and £712,500 per megawatt. The fee would be refunded when a project reaches energization, but could be forfeited if the project exits the queue prematurely.
For a hypothetical 1GW project, that represents hundreds of millions of pounds temporarily tied up.
Project milestones would accompany the financial commitment. Developers would have to demonstrate—mong other things—that they have a credible end user, have begun procuring long-lead electrical equipment, and possess the financial and technical capability to develop the facility.
In other words, wanting to build a data center will no longer necessarily be enough to reserve the electricity needed to build one.
That distinction could become increasingly important as AI infrastructure projects get larger and competition for grid capacity intensifies.
Read the full proposal and provide a response.
Britain is not an outlier
It would be easy to view Ofgem’s proposal as a particularly British response to a particularly British grid problem. But look elsewhere, and a similar pattern emerges.
In the US, the scale is considerably larger. Texas Governor Greg Abbott recently ordered a pause on approvals for new data center projects requiring grid connections while proposed developments undergo an audit.
Reuters reports that the state is reviewing roughly 474GW of proposed new electricity demand, with around 90% of those requests reportedly coming from data centers. That is more than five times Texas’s peak electricity load.
The issue is not confined to Texas. New York introduced a one-year moratorium on construction of large new data centers in July, citing concerns about electricity costs, water use and impacts on local communities.
Ireland provides another illustration of what happens further down the road. Data centers accounted for 23% of metered electricity consumption in 2025, compared with just 5% in 2015.
Australia is seeing rapid growth too. The Australian Energy Market Operator describes data centers as one of the fastest-growing sources of electricity demand in the country’s major electricity markets. At the end of March 2026, 11 large-scale projects representing 5.4GW of maximum demand were progressing through the transmission connection process.
The policy responses differ, but the underlying question is remarkably consistent: how much grid capacity should be allocated to digital infrastructure, when should it be allocated, and who should bear the cost and risk?
The new data center site-selection equation
For decades, the conventional data center site-selection checklist was relatively straightforward: power, land, connectivity, cooling, tax incentives, and proximity to customers.
Power was always important. But the AI boom is changing what “power availability” actually means. Electricity connection has become a key part of the development timetable, rather than something that can be dealt with somewhere in the middle of a project.
A location with a theoretical 500MW connection five or 10 years from now is a very different proposition from one that can provide 500MW on the timetable required by the developer. And a connection offer secured today may be considerably more valuable than the site’s physical characteristics.
This dynamic is beginning to reshape where infrastructure gets built. It also creates a new incentive to think more carefully about where compute is located. If one region has abundant generation but limited transmission capacity, while another has spare grid capacity but less renewable generation, the optimal location for a data center may no longer be obvious.
Australia’s AEMO, for example, says current experience indicates that large data center connections are targeting approximately two years from application to energization. However, the timeframe varies according to project readiness and system conditions. (AEMO)
Compute capacity is becoming power capacity
This matters beyond the companies actually building hyperscale facilities.
Cloud providers and other infrastructure businesses have spent years thinking about capacity in terms of servers, storage, bandwidth and availability zones. Increasingly, they may need to think about megawatts in much the same way they think about compute capacity.
That could influence decisions around:
- Workload location: where compute is physically performed may matter more as power constraints emerge.
- Workload flexibility: some AI and batch workloads may be capable of moving between locations or running when power is more readily available.
- Infrastructure resilience: geographic diversity could become important not only for disaster recovery but for access to electricity.
- Energy procurement: long-term power agreements, onsite generation and storage could become strategic technology considerations rather than facilities-management concerns.
- Project planning: access to a credible grid connection may become a prerequisite for scaling compute capacity.
None of this requires the Cloud industry to stop growing. It means that growth is becoming more physical. And guess what… we haven’t even mentioned data sovereignty and other legal and compliance restrictions.
The Cloud may look virtual from the customer’s perspective, but every AI inference, database transaction, and application request ultimately runs on a physical machine consuming electricity somewhere.
The queue could become a competitive advantage
That leads to an interesting possibility. If electricity connections become scarce enough, having secured a viable connection could become one of the defining competitive advantages in the data center market.
The next phase of the Cloud and AI buildout will be shaped not only by how much compute the industry wants, but also by how quickly the underlying physical infrastructure can be expanded.
The next bottleneck may not be chips, fiber, or even data center space. It may be the queue for the power to run them.
Sign up for the CloudFest newsletter (below) for more news, views, and ways to jump queues.
