AI infrastructure can move from site selection to construction faster than the electricity system can prepare the connection. That mismatch has become a practical constraint for developers planning large computing campuses, particularly where new substations, transmission upgrades or dedicated electrical equipment must support the load. The issue is no longer simply whether enough electricity exists somewhere within a regional grid, but whether the required capacity can reach the site on a firm and commercially usable schedule. The International Energy Agency expects global data center electricity consumption to reach about 945 TWh by 2030, while noting that energy infrastructure generally requires longer planning and construction periods than data centers themselves. For operators, that makes power deliverability a core development variable rather than a supporting utility consideration.

Beyond the Megawatt Headline

A power agreement can look compelling on paper while leaving the most important project question unanswered: when can the facility actually receive firm electricity? A site may sit close to generation yet lack transmission capacity, substation headroom or the equipment needed to complete its connection. The physical path from generation to the data center therefore matters as much as the contracted volume, particularly when several large loads are competing for the same network. Developers need to establish which infrastructure element controls energisation before committing capital against an operational date. This changes site selection from a search for cheap or abundant electricity into a search for electricity that can be delivered at the required voltage, capacity and date.

Why “Power Exists” No Longer Means “Power Is Available”

Electricity generation and electricity availability at a particular site are separate engineering questions. A region can add substantial generation while transmission constraints prevent a new data center from receiving the required capacity without network reinforcement. Interconnection studies determine whether the proposed load or generation can connect without creating unacceptable thermal, voltage or stability impacts, and those studies can trigger additional construction obligations. The resulting schedule can extend beyond the building programme and become the controlling milestone for the entire campus. For a C-level project team, the critical document is therefore not a headline megawatt figure but a defensible energisation pathway backed by the utility and its construction responsibilities.

Generation, Interconnection, or Equipment — What’s Actually Setting Delivery Dates

Different projects face different constraints, but the delivery date usually depends on the slowest unresolved dependency. One site may have adequate generation and an approved connection but no available transformer manufacturing slot, while another may have equipment procurement underway but still require transmission upgrades. Equipment specifications can further narrow the supplier pool because voltage ratings, protection requirements and utility standards limit direct substitution. In practical terms, developers should build the power schedule around verified milestones for generation, interconnection, equipment manufacturing, delivery, installation and commissioning. That approach exposes schedule risk earlier than treating electrical capacity as a single procurement item.

The Interconnection Bottleneck

Interconnection queues provide a useful indication of how difficult it can be to move proposed electricity infrastructure from planning into operation. Lawrence Berkeley National Laboratory’s 2026 analysis found that U.S. projects reaching commercial operation in 2025 had spent more than five years in interconnection processes in regions with available duration data. The same research found roughly 8,200 active projects seeking connection at the end of 2025, representing about 1,312 GW of generation and 749 GW of storage. These figures primarily describe generation and storage rather than data center loads, but they reveal the broader pressure facing transmission systems. AI developers entering constrained territories should therefore investigate the network’s existing queue and planned reinforcement programme before treating available capacity as immediately usable.

Multi-Year Queues and What’s Driving Them

Connection studies can identify upgrades that were never included in the original project schedule, creating additional engineering, permitting and construction requirements. High withdrawal rates can also distort headline queue numbers because proposed projects do not necessarily translate into operational capacity, while multiple projects may compete for the same network improvements. Berkeley Lab found that only 13% of generation capacity entering U.S. interconnection queues from 2000 through 2020 had reached commercial operation by the end of 2025. Developers should consequently distinguish between an application, an executed interconnection agreement and a connection with funded construction work. The further a project has progressed through those stages, the stronger the evidence that its stated power date represents a deliverable milestone rather than an intention.

Regional Variation — Where Interconnection Is Worse, and Why

Queue pressure varies because transmission investment, market design, generation geography and utility planning differ between regions. A location with strong renewable resources may still face congestion if transmission expansion has not kept pace with generation development, while an established industrial corridor may have stronger infrastructure but competing large-load requests. The important comparison for AI operators is therefore not simply regional electricity price but the availability of firm network capacity at the proposed connection point. That said, a favourable queue position does not eliminate equipment, permitting or construction risk. Regional diligence needs to combine utility planning documents, connection studies and equipment availability rather than relying on broad market-level power assessments.

The Equipment Crunch Behind the Timeline

Transformers and high-voltage switchgear can become schedule-critical assets because they connect the planned facility to the electrical network rather than merely supporting internal building systems. Large transformers often require project-specific engineering, manufacturing, testing and transportation, while switchgear must meet detailed electrical and protection specifications. Recent reporting on transformer procurement indicates that some orders can face waits extending toward five years, illustrating the potential mismatch between data center construction schedules and electrical equipment supply chains. A delayed transformer can prevent an otherwise completed substation from being energised, leaving servers and cooling infrastructure unable to operate at their planned capacity. Procurement therefore needs to begin alongside power design rather than after the broader construction programme has already been fixed.

Transformers and High-Voltage Switchgear Lead Times

The equipment problem becomes more difficult when projects require specialised voltage classes, redundancy configurations or utility-approved designs. Developers cannot always replace a delayed component with another product because electrical ratings, protection schemes and interface requirements must remain compatible with the wider network. Early procurement can reduce exposure, but it also commits capital before every project variable has been finalised. Operators should therefore establish approved supplier alternatives, reserve manufacturing capacity where possible and verify testing and delivery schedules as part of the investment case. The objective is not simply to buy equipment early, but to ensure that equipment availability aligns with the exact electrical architecture required for energisation.

How Equipment Scarcity Is Reshaping Project Planning

Equipment constraints are pushing developers toward earlier engineering decisions and greater coordination between utilities, electrical contractors and manufacturers. A project that delays final electrical specifications may lose manufacturing capacity even when the site itself remains on schedule. Procurement teams increasingly need visibility into factory capacity, testing windows, transportation requirements and installation sequencing before financial close. This also changes how investors should evaluate construction risk because a building that appears ready for fit-out can remain commercially inactive if critical electrical equipment has not arrived. The most useful project schedule therefore links procurement milestones directly to the date on which the first powered computing load can operate.

Utilities Under Pressure

AI workloads create a different planning challenge because large computing facilities can concentrate substantial demand within a relatively small geographic area. Utilities must consider peak demand, ramping behaviour, redundancy requirements and the effect of multiple campuses connecting within the same network. The IEA projects accelerated-server electricity consumption, driven largely by AI, to grow about 30% annually through 2030 in its base case. That growth places greater importance on understanding not only how much energy a facility consumes annually but also when and where its highest demand occurs. Utilities therefore need more detailed information from large customers before determining how network investments should be sequenced.

The Challenge of Volatile, Spiky AI Training Loads

AI training can create concentrated computing demand that differs from the smoother consumption profiles associated with many traditional commercial facilities. The utility must plan for the electrical characteristics of the campus rather than simply multiply an average load by operating hours. Large clusters also introduce questions around staged commissioning because the facility may begin with a fraction of its ultimate computing capacity and increase demand as additional halls become operational. As a result, accurate load forecasts and commissioning schedules can become important inputs to utility planning. Operators that provide better load visibility may have greater scope to negotiate phased connections or operating arrangements that match network availability.

Grid Planning for Loads That Don’t Behave Like Traditional Demand

Traditional grid planning often relies on long-term forecasts of relatively predictable customer categories, while AI campuses can introduce rapid demand growth concentrated around individual sites. Utilities must account for the possibility that several large projects may seek capacity within the same planning horizon, creating uncertainty around reinforcement requirements. This makes early communication between developers and utilities more valuable because a project can affect network planning well before construction begins. Operators should provide realistic load curves, commissioning stages and ultimate capacity assumptions instead of presenting only a maximum megawatt requirement. Better planning visibility can help utilities distinguish committed demand from speculative pipeline capacity.

Utility Responses — Tariffs, Curtailment, and Load Management

Utilities have several tools available when network capacity cannot immediately support every requested load. These can include differentiated tariffs, phased connections, demand-management arrangements and curtailment provisions that allow certain customers to reduce consumption during defined system conditions. Such arrangements can accelerate access in some circumstances, but they also change the operational assumptions behind redundancy and workload planning. A data center operator needs to know whether contracted capacity remains firm during system stress or depends on conditions attached to the connection agreement. Investors should treat those provisions as part of the project’s technical operating model rather than as minor commercial clauses.

Bridging the Power Gap

Renewable PPAs can provide long-term energy procurement and support emissions objectives, but a PPA does not automatically create physical capacity at the data center. The renewable project still needs a grid connection, and its generation profile may not match the data center’s continuous requirement for firm power. On-site generation can address some of the physical-access problem by placing generation closer to the load, but fuel supply, emissions, permitting and operating economics remain material considerations. The correct solution therefore depends on whether the immediate objective involves energisation, reliability, carbon reduction or long-term energy cost management. In other words, a power source should be judged by when it can provide usable capacity, not merely by how much generation it represents on paper.

Renewable PPAs — Scale, Speed, and Limitations

Renewable procurement can move relatively quickly compared with some large grid projects, particularly when developers can contract existing or advanced projects with available transmission access. Yet the physical electricity delivered to a data center remains subject to network constraints, generation availability and the structure of the procurement arrangement. Operators should therefore separate contractual energy procurement from firm capacity procurement when evaluating a PPA. That distinction becomes especially important for AI campuses that require high availability and cannot simply reduce computing activity whenever renewable output declines. A PPA can strengthen the energy strategy, but it cannot by itself eliminate a local interconnection bottleneck.

On-Site and Behind-the-Meter Generation

Behind-the-meter generation can provide a practical bridge where grid capacity will arrive later than the desired computing schedule. Gas generation, renewable systems, batteries and hybrid configurations can potentially support staged operations, although each technology introduces different capital, fuel, emissions and reliability requirements. The key question is whether the system can provide the required firm capacity at the required duty cycle rather than simply its nameplate output. Operators also need to understand how islanding, synchronization, maintenance and grid reconnection would work before treating on-site generation as a substitute for grid service. This makes engineering integration as important as the generation technology itself.

Nuclear and SMRs — Realistic This Decade, or Not Yet?

Nuclear has a credible role in the longer-term electricity supply mix for AI, but the timing varies sharply between existing reactors, new large reactors and small modular reactors. The IEA expects nuclear generation to contribute increasingly toward data center demand later this decade and states that the first SMRs are expected to come online around 2030. That timeline makes SMRs relevant to projects planning beyond the immediate build cycle, but it does not make them a universal solution for campuses seeking near-term energisation. Developers should therefore avoid using future nuclear capacity as a substitute for a connection plan that needs to work within the next two or three years.

Weighing Bridging Options Against Delivery Timelines

Every bridging option should be assessed against four dates: when it can be contracted, when construction can begin, when electricity can first flow and when full capacity becomes available. A technology with excellent long-term economics may have limited value if it cannot support the first operational phase of a campus. Conversely, a more expensive temporary supply arrangement may protect a project schedule if it enables early deployment while permanent grid infrastructure is completed. Capital allocation should therefore compare solutions on delivered capacity, reliability, permitting risk and schedule certainty rather than generation cost alone. This approach keeps the power strategy aligned with the commercial objective of bringing computing capacity online.

Announced vs. Deliverable

The difference between announced capacity and powered capacity is becoming increasingly important as developers publish larger AI infrastructure pipelines. An announcement can represent land secured, a proposed campus, a planned expansion or a future phase that still depends on financing, permits, grid access and equipment procurement. Berkeley Lab’s queue data demonstrate why proposed capacity should not automatically enter investment models as operational capacity, because only a small share of historical interconnection requests ultimately reached commercial operation. Operators should therefore communicate the status of each project through measurable milestones rather than a single headline capacity number. This creates a more useful picture for customers, lenders and investors evaluating when compute capacity will actually become available.

The Widening Gap Between Project Announcements and Powered Capacity

The gap becomes visible when a project has a site and development plan but lacks a completed electrical pathway. A credible pipeline should identify whether the connection application has advanced, whether network upgrades have been assigned, whether critical equipment has been ordered and whether the utility has established a firm energisation schedule. Each milestone reduces a different category of execution risk, making them more useful than cumulative announced megawatts. Investors should also distinguish between first-phase power and ultimate campus capacity because the two may have very different delivery dates. A project can therefore be commercially credible at an initial capacity while still carrying substantial uncertainty around later expansion phases.

How Operators and Investors Should Read Announced Pipelines

Pipeline analysis should classify projects according to development maturity rather than aggregate every proposed megawatt into one number. Useful categories include announced, site-secured, permitted, power-contracted, construction-ready and energisation-ready capacity, with each stage carrying a different probability of completion. The distinction matters for investors because revenue timing depends on powered and commissioned capacity rather than announced capacity. Operators should apply the same discipline internally when deciding where to allocate GPUs, cooling equipment and construction capital. Ultimately, the strongest pipeline is not the largest one but the one with the clearest evidence connecting development commitments to physical power delivery.

Who Pays for the Grid Build-Out

Large AI loads can trigger network investment that benefits the wider electricity system while initially serving a specific customer. That creates a difficult allocation question for utilities and regulators because charging the entire cost to ratepayers may create affordability concerns, while assigning every upgrade to the new customer can discourage economically valuable development. The appropriate structure depends on the regulatory framework, the degree of shared network benefit and the likelihood that other customers will use the upgraded infrastructure. Developers therefore need to understand connection charges and network contribution requirements before finalising site economics. Grid investment should be treated as part of the total infrastructure cost rather than an external assumption.

Cost Allocation Between Utilities, Operators, and Ratepayers

Cost allocation becomes particularly important when an AI campus requires a dedicated substation or transmission reinforcement that exceeds existing local capacity. The operator may fund some assets directly, while the utility may recover other investments through regulated tariffs or connection charges. The commercial outcome can materially change project economics because a low-cost power market may still require significant upfront infrastructure expenditure. Investors should model these obligations explicitly and identify which costs remain with the developer if the project is delayed, downsized or cancelled. Clear allocation also reduces disputes when the original load forecast changes during development.

Policy and Regulatory Levers on the Table

Regulators can influence the pace of grid expansion through transmission planning, connection reform, tariff design and rules governing large-load participation. In the United States, reforms under FERC Order 2023 seek to improve interconnection processes, although Berkeley Lab says it remains too early to assess their full impact. The broader objective is to make connection processes more predictable while maintaining system reliability and appropriate cost allocation. For developers, regulatory reform matters because a faster queue process has limited value if physical network construction and equipment procurement remain slow. Policy therefore needs to address both procedural delay and physical infrastructure constraints.

The India Lens

India offers a different combination of grid regulation, renewable procurement options and captive generation structures that can shape how large AI facilities secure electricity. The Green Energy Open Access Rules reduced the eligibility threshold for green-energy open access from 1 MW to 100 kW, while captive consumers have no minimum limit under the framework described by the Ministry of Power. That creates procurement flexibility for large technology customers, but physical transmission capacity and state-level implementation still determine whether contracted electricity can reach a particular facility reliably. Developers planning large campuses therefore need to evaluate regulatory access and physical grid capability together.

Grid Availability and Open Access Rules

Open access can expand the number of procurement routes available to an AI operator, particularly when renewable electricity can be sourced outside the immediate distribution utility’s generation portfolio. Yet open access does not remove the need for transmission capacity, scheduling arrangements, charges and appropriate network infrastructure. The Ministry of Power has positioned green open access as a mechanism for improving renewable-energy procurement, but project-level feasibility still depends on the applicable state and network conditions. For large computing campuses, the relevant question is therefore whether open access can deliver firm and commercially predictable electricity at the required site and scale.

Captive Power as a Workaround

Captive generation can provide greater control over supply where grid expansion does not match the desired project schedule. India’s framework includes ownership and consumption requirements for qualifying captive arrangements, which means developers must structure projects carefully rather than treating captive generation as a simple backup purchase. The model can also introduce additional responsibilities around fuel, operations, maintenance, transmission and environmental compliance depending on the technology selected. For AI operators, captive generation should therefore be evaluated as an integrated power project rather than a standalone workaround. Its value depends on whether it can provide dependable capacity at a cost and schedule that support the computing business case.

What India’s Constraints Mean for Gigawatt-Scale AI Ambitions

India’s ability to support large AI campuses will depend on coordination between generation expansion, transmission investment, open access mechanisms and site-level infrastructure. Renewable availability alone does not guarantee that a gigawatt-scale computing campus can receive firm power when required, particularly if network reinforcement must precede connection. Developers should consequently evaluate state-level grid capacity and transmission access before selecting locations based solely on renewable potential or land availability. The country’s regulatory flexibility can create useful options, but those options still require physical infrastructure capable of supporting sustained high-density computing loads.

What This Means for Operators, Utilities, and Investors

AI infrastructure site selection now needs to begin with a power-delivery map rather than ending with an electricity procurement strategy. Operators should identify the generation source, transmission path, connection point, substation requirements, equipment status and commissioning sequence before committing to a major capacity target. Utilities need realistic information about load growth, phasing and operating behaviour so that network investments reflect credible demand rather than speculative announcements. Investors should test whether every major power milestone has documentary evidence, an accountable party and a realistic construction window. This creates a more defensible basis for capital deployment than relying on available megawatts without understanding how those megawatts reach the facility.

Reframing Site Selection and Capital Deployment Around Power Realism

The strongest sites will increasingly be those where power can be delivered predictably, not simply those offering attractive electricity prices or large theoretical generation potential. A credible development assessment should examine connection status, transmission capacity, utility commitments, equipment availability, permitting and the timing of each required upgrade. Developers can then sequence capital around verified milestones instead of committing the full project budget against uncertain future capacity. This approach also allows operators to consider phased deployments where the first computing halls become operational before the ultimate campus reaches full capacity.

Key Questions for Due Diligence Going Forward

The essential questions are now straightforward: what capacity exists today, what capacity has been contractually secured, what network upgrades remain, which equipment has been ordered and when can the utility energise the site? The answers should distinguish firm capacity from conditional capacity and operational power from announced future capacity. Developers should also identify the consequences of delays, including whether temporary generation, storage or workload flexibility can protect the first operating phase. Investors should require the same level of evidence for power infrastructure that they already expect for land, financing and construction contracts. AI compute can scale only when the electrical system supporting it reaches the same level of readiness, making power deliverability one of the central determinants of infrastructure value.