The Electricity Race

Data centers support cloud computing, streaming, financial transactions, e-commerce, cybersecurity, scientific research, enterprise software, and the ordinary storage and processing of digital information. They can be planned and built in just a few years. The transmission lines, substations, transformers, and generating capacity needed to serve them often take much longer. This mismatch has turned electricity infrastructure into one of the principal constraints on the expansion of the digital economy.

The technologies needed to expand electricity supply already exist, and large technology companies are committing billions of dollars to secure long-term electricity supplies. The binding constraint lies instead in the institutional architecture of the electricity sector. Permitting, transmission planning, interconnection studies, and geographically fragmented regulation have not kept pace with rapid growth in large-load electricity demand. Physical supply chains have also become strained, particularly for substations, transformers, switchgear, and transmission equipment. These bottlenecks slow the coordination of investment decisions across utilities, regulators, generators, data center developers, and large electricity consumers.

Recent experience illustrates both the challenge and the opportunity. Georgia’s Vogtle Units 3 and 4, completed years behind schedule and after substantial cost overruns, demonstrate the difficulty of delivering large-scale electricity infrastructure under existing institutional arrangements. Emerging procurement models, including long-term power purchase agreements (PPAs) and co-located electricity arrangements, as illustrated by Amazon’s agreement with Talen Energy and Microsoft’s agreement with Constellation Energy, offer a promising approach to improving investment coordination and supporting future investment in reliable, carbon-free generation.

Meeting the electricity needs of the data center economy requires a new electricity policy framework built for speed, coordination, and investment, while maintaining high standards of safety and reliability. Leadership in the next generation of digital infrastructure will depend on more than advanced AI models. It will also depend on the ability to build, finance, and coordinate the electricity infrastructure needed to power data centers and the broader set of services that depend on them.

Key Takeaways:

• Data centers are creating a structural shift in electricity demand. AI is an important accelerator, but the broader expansion of cloud computing, digital services, enterprise computing, and data storage is also increasing the need for reliable electricity.
• Existing institutions and supply chains have not kept pace. Permitting, transmission planning, interconnection queues, and equipment procurement still operate on timelines that predate the rapid growth of large, geographically concentrated electricity loads.
• Electricity infrastructure has become a coordination problem. Scaling digital infrastructure depends on aligning investment decisions across utilities, regulators, grid operators, power producers, technology firms, and large electricity consumers.
• PPAs and co-located electricity arrangements facilitate investment coordination. By aligning incentives between generators and large electricity consumers, they support timely investment in new generation while transforming large technology companies from passive electricity purchasers into active participants in financing, coordinating, and
  expanding electricity infrastructure.
• Electricity procurement is evolving. Long-term PPAs and co-located electricity arrangements remain in their early stages but point toward more flexible bilateral contracting between electricity producers and large electricity consumers. As this evolution continues, data centers are likely to play a more active role in developing, financing, and in some cases directly securing electricity generation

After years of modest electricity demand growth, the sudden acceleration has jolted the power sector awake. In retrospect, what seemed like a period of stability was merely the calm before a broader computational storm. The race to scale AI has made the issue more visible, but the underlying pressure is larger: cloud computing, streaming, e-commerce, financial services, cybersecurity, scientific computing, and ordinary data storage all require physical facilities that consume electricity around the clock. The central problem is simple: adding servers is much easier than adding megawatts.

Data center investments rest on the expectation that sufficient electricity will be available when facilities come online, either through timely grid connections or through dedicated generation built specifically for the project, including self-generation and other behind-the-meter resources. Utilities and independent power producers face the mirror-image problem: they are reluctant to commit billions of dollars to new generation and network expansion unless they are confident that the projected demand will actually materialize.

This creates a coordination problem. A data center developer may be ready to lease land, order servers, and sign customer contracts long before electricity suppliers have completed the studies, permitting, and infrastructure investments needed to serve the new demand. Conversely, utilities and independent power producers may hesitate to invest in new generation and grid infrastructure for a project that could be delayed, downsized, or relocated. Long-term PPAs help reduce these risks by creating durable commitments between electricity producers and large consumers, facilitating coordinated investment in electricity infrastructure alongside data center deployment.

The challenge is compounded by the institutional complexity of the electricity sector. Its regulatory framework, lengthy permitting processes, and regional fragmentation evolved in an era of gradual, predictable demand growth and have struggled to adapt to the speed, scale, and geographic concentration of new large loads. The bottlenecks are cumulative: a project can have a customer, a site, and a financing plan and still wait for interconnection studies, transmission upgrades, siting approvals, transformer deliveries, or local substation work.

Recent experience offers little reassurance. Georgia’s Vogtle Units 3 and 4 illustrate the difficulty of delivering major electricity infrastructure under existing institutional arrangements. Although the project ultimately added more than 2.2 GW of reliable, carbon-free generating capacity, it suffered years of delays and substantial cost overruns, demonstrating how regulatory complexity, permitting, supply chain constraints, and project execution can impede strategically important infrastructure. Similar bottlenecks now appear in transmission development and interconnection queues. If these bottlenecks persist, they may become one of the principal constraints on scaling digital infrastructure itself.

The broader lesson is that expanding generation capacity is increasingly an institutional as well as an engineering challenge. The countries that succeed in the next phase of the digital economy will be those that can build electricity infrastructure as rapidly as they deploy computing capacity. PPAs can help coordinate investment, though broader reforms remain necessary. Building electricity infrastructure at the pace required by data centers and other large loads will require a new electricity policy framework built for speed, coordination, and investment, as well as a rethinking of how electricity infrastructure is planned, financed, permitted, and delivered.

Supplying electricity to a new data center requires coordinating multiple infrastructure investments that extend well beyond the facility itself. Planning begins with a practical question: how much electricity will the data center need, and where will that demand appear on the grid? That demand can be met either through the existing electricity system or through dedicated generation developed specifically for the project. Dedicated generation may be owned and operated by the data center itself, often described as self-generation, or provided through a behind-the-meter arrangement.

Depending on the supply model, reliable electricity service may require securing adequate generation, obtaining grid interconnection approval where applicable, expanding substation capacity, upgrading transmission and distribution infrastructure, installing backup power systems, and satisfying a range of operational and regulatory requirements. Interconnection is the process by which a new generator or large customer is studied, approved, and physically connected to the grid. Because these steps are controlled by different institutions and move on different timelines, coordination is essential.

As a result, the pace of data center development is increasingly constrained by the availability of supporting electricity infrastructure. According to the International Energy Agency’s Energy and AI (2025) report, large data centers can typically be planned and built within two to three years, whereas new transmission lines, substations, and generation projects often require a decade or more to deliver. The same report estimates that data centers used about 415 terawatt-hours of electricity globally in 2024 and projects that consumption could more than double by 2030. The IEA attributes this growth to AI-focused accelerated servers as well as conventional servers, storage, networking, cooling, and other infrastructure. The infrastructure challenge therefore turns on both the scale of electricity consumption and the speed with which the electricity system can deliver capacity in the right places.

The scale of the interconnection backlog illustrates the problem. An interconnection queue tracks proposed projects as grid operators conduct studies, estimate costs, and negotiate connection agreements. Lawrence Berkeley National Laboratory’s Queued Up: 2026 Edition reports that, as of the end of 2025, more than 2,060 GW of generation and storage capacity were actively seeking transmission interconnection in the United States. The median time from interconnection request to commercial operation exceeded five years for projects built in 2025, and only 13 percent of capacity that entered queues from 2000 through 2020 had reached commercial operation by the end of 2025. Queue data show that connecting new supply has become a major institutional bottleneck.

Permitting and supply chains reinforce the queue problem. DOE’s Office of Electricity reports that distribution-transformer lead times increased from three to six months in 2019 to 12 to 30 months in 2023. Similarly, in its 2023 report Electricity Grid: DOE Could Better Support Industry Efforts to Ensure Adequate Transformer Reserves, the Government Accountability Office (GAO) found that utilities identified supply-chain constraints—including long and increasing manufacturing lead times, limited manufacturing capacity, and labor and material shortages—as the most pressing challenge to maintaining adequate reserves of large power transformers.

In practical terms, a data center or generator may be commercially ready before the wires, substations, or transformers needed to connect it are available. These constraints shape how quickly generation and load can actually be connected, even as the need for new generation continues to grow.

Closing this gap will require more flexible electricity markets and regulatory frameworks that support timely investment in generation and supporting infrastructure. Long-term PPAs are one such mechanism. By creating long-term contractual commitments, they can help electricity producers and large consumers plan around each other rather than making separate investments based on uncertain expectations. Hyperscalers, large technology companies that operate massive data center networks, are also increasingly investing in dedicated electricity infrastructure, including self-generation and behind-the-meter arrangements, to reduce dependence on grid expansion and better align electricity supply with the pace of data center deployment.

The completion of Georgia’s Vogtle Units 3 and 4 offers an important lesson in delivering large-scale electricity infrastructure. As the first new U.S. nuclear reactors to enter commercial operation in more than three decades, the AP1000 units demonstrate that advanced nuclear technology remains a viable source of reliable, carbon-free baseload electricity. Yet their development also illustrates how hard it can be to turn an approved project into an operating asset. Construction began in 2009 with an expected completion schedule of 2016 and 2017 and an estimated cost of $14 billion. Instead, first-of-a-kind engineering, post-Fukushima design revisions, evolving regulatory requirements, contractor turnover, supply chain disruptions, and project management challenges delayed commercial operation until 2023 and 2024, while total construction costs climbed to more than $36 billion.

Rather than reflecting an inherent limitation of nuclear technology, the Vogtle experience underscores the importance of predictable permitting, stable regulatory frameworks, resilient supply chains, standardized designs, and disciplined project execution. Despite its prolonged construction, the project ultimately added more than 2.2 GW of new carbon-free generating capacity, making Plant Vogtle the largest nuclear power station in the United States.

As electricity demand accelerates with the rise of data centers and other large loads, the ability to deliver large, capital-intensive infrastructure on time and on budget may prove just as important as the technologies themselves. Vogtle therefore serves less as a cautionary tale about nuclear power than as a reminder that the success of the digital economy will depend as much on institutional capacity and effective project delivery as on technological innovation.

The broader lesson is that institutional capacity must be complemented by market mechanisms that better coordinate investment decisions across electricity producers and large consumers, while accounting for interconnection, permitting, transmission, and supply-chain constraints. Among these mechanisms, long-term PPAs have emerged as one of the most important tools.

Power purchase agreements (PPAs) are long-term contracts under which an electricity buyer, often called the offtaker, agrees to purchase electricity from a specific generating facility under agreed commercial terms, typically for a period of 10 to 25 years. The basic idea is straightforward: the buyer receives a more predictable electricity arrangement, while the generator receives a revenue commitment that can help support financing for new generation. There are two main types of PPAs: physical and financial (or virtual). The distinction can be understood by asking a simple question: Does the buyer actually receive the electricity generated by the project?

In a physical PPA, the buyer receives the electricity generated by the project. Electricity from a specific generating facility is delivered to the buyer through the electricity grid, while the buyer pays the agreed contract price. The generator continues to own and operate the facility, and the buyer secures a long-term electricity supply at a predictable cost. Because electricity is physically delivered, the buyer and the generating facility are generally located within the same regional electricity market.

In a financial PPA, the buyer does not receive the project’s electricity directly. The generator sells electricity into the wholesale market, while the buyer continues purchasing electricity from its local utility. The parties instead exchange payments based on the difference between the agreed contract price and the wholesale market price. The buyer receives price certainty, while the generator secures predictable revenues that support investment in new generation.

PPAs can also differ in the physical relationship between the electricity buyer and the generating facility. In some cases, the generating facility is co-located with the customer’s operations, allowing electricity to be supplied directly through a behind-the-meter arrangement in which power is consumed before it reaches the utility meter. In other cases, the generating facility is located elsewhere and connected directly to the transmission network through an in-front-of-the-meter arrangement, with electricity delivered through the existing transmission and distribution system. Co-location describes the physical configuration of the project rather than a separate type of PPA.

From the perspective of the data center economy, the importance of PPAs extends well beyond electricity procurement. By creating long-term contractual commitments between electricity producers and large consumers, PPAs help coordinate investment decisions that might otherwise be delayed. A PPA cannot clear an interconnection queue, obtain a permit, or manufacture a transformer. It can, however, make a project’s future revenues more credible and give both sides a stronger basis for planning. For that reason, PPAs play an increasingly important role in supporting new generating capacity, easing one of the principal bottlenecks to expanding digital infrastructure, and helping distinguish serious projects from more speculative proposals.

Although still relatively uncommon outside specific market segments, long-term PPAs and co-located arrangements are becoming increasingly important as data center electricity demand grows. AI increases the urgency of aligning electricity infrastructure with rapidly growing demand, while the broader expansion of digital infrastructure gives that challenge wider significance. Their evolution is likely to encourage a wider range of bilateral contracting models tailored to large electricity consumers and infrastructure developers, especially where conventional utility planning, interconnection processes, and supply chains cannot move at the same pace as private computing investment.

The growing importance of PPAs is best illustrated by the recent agreement between Amazon and Talen Energy

The long-term agreement between Amazon Web Services (AWS) and Talen Energy provides one of the clearest examples of this emerging model. Rather than relying exclusively on electricity purchased from the local utility, Amazon secured long-term access to electricity from Talen’s Susquehanna Nuclear Power Plant in Pennsylvania through a PPA.

Announced in June 2025, the agreement provides up to 1,920 MW of carbon-free nuclear electricity through 2042, with deliveries ramping up over time and reaching the full contracted quantity by 2032, subject to possible acceleration.
The agreement illustrates that meeting the electricity needs of large data centers involves far more than purchasing electricity. It requires coordinating generation assets, transmission infrastructure, electricity markets, interconnection requirements, equipment procurement, and regulatory approvals. Initially structured as a co-located, behind-the-meter arrangement, the partnership later transitioned to an in-front-of-the-meter model, under which electricity from the Susquehanna plant is delivered through the PJM transmission network. In practical terms, this shifted the arrangement from power supplied directly next to the facility to power delivered through the regional grid. This structure allows Amazon to support a broader portfolio of operations while preserving the integrity of the regional wholesale electricity market.

The agreement also demonstrates how modern PPAs align the incentives of multiple stakeholders. For Amazon, the contract secures a long-term supply of reliable, carbon-free electricity while reducing exposure to future price volatility. For Talen, it provides stable contracted revenues, reducing dependence on wholesale electricity markets. The arrangement also supports continued investment in transmission infrastructure and the long-term operation of the Susquehanna plant, creating broader economic benefits for the region.

Perhaps the most significant feature of the Amazon-Talen partnership is that it extends well beyond a conventional electricity purchase. Alongside the long-term PPA, the companies announced plans to evaluate the deployment of small modular reactors (SMRs) and explore engineering upgrades that could increase the output of the existing Susquehanna plant. Rather than simply reallocating existing electricity, these initiatives are intended to expand the region’s supply of reliable, carbon-free generation. The agreement therefore combines long-term electricity procurement with a strategy for future generating capacity.

More broadly, the Amazon-Talen agreement illustrates how long-term PPAs can strengthen the economic viability of existing nuclear facilities by providing stable and predictable revenue streams. In competitive electricity markets, where merchant nuclear plants depend on volatile wholesale prices, such contracts can reduce the risk of premature retirement while creating incentives to maintain, upgrade, and expand existing assets. A similar approach underpins Microsoft’s agreement with Constellation Energy to support the planned restart of the Crane Clean Energy Center (formerly Three Mile Island Unit 1), demonstrating how long-term corporate electricity procurement can both preserve existing clean generation and stimulate investment in additional capacity.

Taken together, the Amazon-Talen and Microsoft-Constellation agreements point to a broader transformation in electricity markets. Large technology companies are no longer passive purchasers of electricity; they are becoming active participants in financing, coordinating, and expanding electricity infrastructure. As these arrangements mature, they are likely to encourage a broader range of bilateral contracting models that better match the investment horizons of electricity producers and large consumers.

The data center economy is reshaping the relationship between digital infrastructure and the electricity system. Unlike previous waves of digitalization, the current expansion is creating a scale and concentration of electricity demand that can no longer be treated as a marginal addition to existing loads. The challenge now reaches beyond producing more electricity to delivering it to the right locations, through the right infrastructure, on timelines that match technological change.

Meeting the electricity needs of the data center economy therefore requires a policy framework that facilitates timely investment and effective coordination. Regulatory processes must become more predictable, interconnection procedures more efficient, transmission planning more forward-looking, and electricity markets better equipped to support investment in generation, storage, and network infrastructure. The Department of Energy’s draft 2026 National Transmission Needs Study identifies a pressing need for additional transmission infrastructure due to load growth, including new generation interconnection, new load interconnection, and congestion relief. At the same time, policy should encourage innovative bilateral procurement models, including long-term PPAs, co-located arrangements, and dedicated generation, that better align incentives between electricity producers and large consumers.

Ultimately, success will depend on closer coordination among utilities, grid operators, regulators, technology firms, power producers, investors, equipment manufacturers, permitting authorities, landowners, and large electricity consumers. Data center deployment increasingly depends on their ability to align investment decisions across the electricity value chain. The speed of data center construction matters, but so does the speed with which generation, transmission, interconnection capacity, and supporting infrastructure can expand to serve it. The evolution of more flexible bilateral contracting between electricity producers and large consumers will be an important part of that transition, complementing traditional electricity markets rather than replacing them.

The countries that lead the next generation of digital infrastructure will combine technological leadership with institutional capacity. Advanced large language models and chips will matter, but so will the ability to build, finance, and coordinate the electricity infrastructure needed to power the full data center ecosystem. The AI race is part of a larger electricity race. Countries that cannot deliver megawatts at the pace they deploy computing capacity risk becoming followers rather than leaders, regardless of their technological capabilities.