The data center industry has traditionally treated access to the electricity grid as something to secure before development begins. AI is exposing the weakness of that assumption as projects grow faster, loads become larger and power networks struggle to distinguish genuine demand from an increasingly crowded queue of proposed capacity.
Britain offers a particularly stark illustration of what happens when the development cycle of artificial intelligence collides with the development cycle of electricity infrastructure. In July, Ofgem opened a consultation on new rules for data center connections after demand applications surged from 41GW to 125GW in less than a year, with data centers accounting for at least 80GW. The regulator is considering commitment fees and development milestones designed to remove speculative projects from a queue that is becoming increasingly difficult for the grid to manage.
The significance reaches far beyond Britain because the same mismatch is appearing across major data center markets. Global real estate and infrastructure advisory firm JLL now identifies power, rather than location or cost, as the primary site-selection criterion, while Capgemini’s 2026 research finds that 77 percent of utility executives expect data center demand to grow faster than their ability to expand supply. The industry is therefore entering a period in which securing land, planning consent and customers may matter less than answering a more fundamental question: when, and under what conditions, can enough electricity actually reach the site?
AI has changed the scale of the request
Data centers have always consumed substantial amounts of electricity, but AI is changing both the scale and character of what operators ask grids to provide. A conventional development might once have progressed through phases measured in tens of megawatts; hyperscale AI campuses can be planning for several hundred megawatts within relatively short periods, creating demand that electricity networks were never designed to absorb at software-industry speed.
Capgemini found that 78 percent of data center leaders regard grid infrastructure construction timelines as a critical constraint. Among electricity executives, 84 percent cite insufficient reserve margins and the same proportion point to delays securing permits for infrastructure upgrades, while lengthy interconnection studies and aging infrastructure remain widespread concerns. The challenge is not simply adding generation. New capacity can require transmission upgrades, substations, transformers and switchgear before the first server is energized, creating dependencies whose timescales are measured in years.
JLL estimates that the average grid connection wait in primary data center markets now exceeds four years, with some established European locations facing considerably longer delays. Its analysis shows why developers are being forced to secure power earlier, consider phased developments and engage utilities alongside alternative energy providers long before the building itself is ready. In a market expected to add almost 100GW of global capacity between 2026 and 2030, the most valuable attribute of a site may increasingly be the certainty of its power delivery date rather than the quality of the land around it.
That pressure is already changing operator behavior. CyrusOne’s partnership with E.ON was created to explore local generation, power purchase agreements and other energy solutions in Frankfurt, where grid constraints have complicated expansion. The operator subsequently revised the power strategy for its FRA7 development, prioritizing fast-track grid-connected infrastructure while continuing to pursue alternative solutions elsewhere. The episode is revealing because it shows that the emerging model is not simply to abandon the grid, but to create several credible routes to power and retain the flexibility to use whichever can deliver capacity at the right time.
The queue has become part of the constraint
Electricity scarcity is only part of the problem because utilities are also struggling to determine how much of the enormous demand appearing in connection queues will ever become real. Capgemini found that two-thirds of electricity executives believe around one fifth of data center load requests ultimately fail to materialize, while 77 percent say speculative applications make demand forecasting more difficult. Sixty percent are concerned that overestimating future requirements could leave them with stranded infrastructure.
That uncertainty creates a difficult planning dilemma. Utilities cannot ignore credible requests for hundreds of megawatts when major technology companies are committing unprecedented sums to AI infrastructure, but building generation and network capacity around every proposed campus could direct billions of dollars towards projects that never proceed. Ofgem’s proposed reforms are an attempt to distinguish genuine developments from speculative queue positions by requiring applicants to demonstrate progress and retain more financial commitment to the capacity they request.
The issue also changes the relationship between data center growth and the wider economy. A connection request for a large AI campus can influence how scarce network capacity is allocated between digital infrastructure, industrial projects, housing and existing customers. The debate therefore moves beyond whether a data center can pay for its connection towards whether the electricity system can accommodate competing forms of demand without creating new reliability or investment problems.
Uptime Institute’s 2026 survey shows how strongly operators are already feeling this pressure. Power availability ranks among the leading management concerns alongside cost, capacity forecasting and supply-chain disruption, while the organization describes an industry expanding under conditions of declining grid reliability and increasing complexity. For developers, this makes a grid offer less valuable if the date, firmness or long-term conditions attached to that capacity remain uncertain.
From grid customer to power-system participant
The most important consequence may be that data centers are beginning to move from passive electricity customers towards active participants in the power system. Capgemini reports that 29 percent of data center executives already use some form of on-site generation and another 39 percent plan to add on-site or behind-the-meter capacity within the next two years. More than seven in ten expect to reduce their reliance on the grid significantly over the next five years.
The shift does not mean every operator wants to become an electricity producer. It reflects the reality that waiting passively for network reinforcement is increasingly incompatible with the timetable on which AI capacity is expected to arrive. Batteries, local generation and private-wire arrangements can bridge some delays, while more flexible commercial structures may allow operators to secure capacity that would otherwise remain inaccessible.
Demand flexibility could become another part of that relationship. Capgemini notes that only half of utilities currently design data-center-specific demand-response programs, although three-quarters acknowledge the need for greater agility in how they work with large digital loads. Workloads that are not latency-sensitive may eventually be shifted between regions or scheduled around periods of electricity availability, provided customer agreements, data sovereignty and operational requirements allow it.
The attraction for utilities is that a data center capable of adjusting demand, using batteries or supporting the grid during periods of stress is very different from one that insists on maximum uninterrupted consumption at all times. The attraction for operators is potentially faster access to constrained networks. Neither removes the need for major investment in generation and transmission, but both create ways of using existing infrastructure more intelligently while those larger upgrades are being built.
Power strategy moves to the front of development
The traditional sequencing of data center development is therefore being reversed. Energy can no longer be treated as a utility service to be secured once a promising location has been identified, because the credibility and timing of the power solution increasingly determine whether the location is promising in the first place. Developers need to understand grid capacity, connection certainty and alternative generation options before making decisions that once revolved primarily around land, connectivity and customer demand.
This also explains why secondary markets can suddenly become strategically important. Locations previously considered less attractive may gain an advantage if they can provide substantial, reliable electricity sooner than established hubs, although power alone will never make a market viable without fiber, skills and an appropriate regulatory environment. The geography of data center development is consequently being shaped by time-to-power as much as by traditional measures of demand.
The grid connection will remain fundamental, but it is becoming only one component of a broader energy strategy. AI has made power too important, too scarce and too closely tied to project timing for operators to regard it as somebody else’s infrastructure problem. The winners in the next phase of data center expansion will be those able to combine grid capacity, alternative supply and operational flexibility in a way that converts ambitious development plans into usable compute before the opportunity moves elsewhere.
