As artificial intelligence reshapes the global economy, the hyperscalers driving the transformation face surging demand for robust but also rapidly evolving digital infrastructure.
The International Energy Agency (IEA) now projects USD3.9 trillion of data center investment globally between 2026 and 2030.1 Yet delivering the enormous volumes and quality of electricity these facilities require is proving increasingly difficult through existing networks. Waits for grid connection in major markets can now stretch to years, costs are escalating unpredictably, and aging grid networks are already under strain from the broader struggle to keep pace with the needs of the energy transition.
Against this backdrop, data center developers are turning to “behind-the-meter” (BTM) powering in which a dedicated generation facility is sited at or near the data center, either bypassing the public grid entirely or relying on it only as backup.
While grid connection is likely to remain the preferred approach for large-scale projects in most regions, the case for BTM solutions to power new data centers is strengthening. Here we explore the forces driving this shift and the challenges BTM frameworks present for market participants.
Data center electricity demand and grid challenges
The sheer scale of data center energy needs is striking. According to the IEA2, data centers accounted for just under 1.5% of global electricity consumption in 2024. That figure grew by a further 17% in the year to 2025, with total consumption projected to roughly double by 2030. The energy demands of developing and operating large language models, for example, are huge: according to MIT Technology Review, training OpenAI’s GPT-4 model reportedly consumed around 50GW hours of energy, enough to power San Francisco for three days.
Analysts estimate that electricity demand from data centers will grow to around 3% of global electricity demand by 2030. At the facility level, demand is scaling just as dramatically. New cutting-edge data centers specifically designed for AI routinely exceed 100MW, and projects such as the OpenAI Stargate initiative are targeting up to 10 GW—reportedly roughly equivalent to the peak power demand of New York City.
But this demand (and the accompanying pressure on governments to facilitate the digital infrastructure required for AI) comes at a time when electricity networks are already under pressure, contending with the electrification of transport and industry and higher energy usage as a result of rising living standards, amid a broader global realignment around energy security (see our article, Drive for energy security creates a more complex landscape for businesses).
Networks in many major markets were originally designed and built decades ago, in an era of significantly lower populations and overall energy demand. In the United States, much of the power grid dates from the mid-twentieth century and was designed for a centralized generation model serving a fraction of today’s load.
In Europe, national grid systems developed in the post-war period face analogous constraints: they were not designed for the large-scale integration of intermittent renewable generation, let alone the concentrated, gigawatt-scale demand profiles of modern hyperscale data centers.
In 2023, the IEA estimated that global investment in electricity grids needed to reach approximately USD600 billion per year by 2030 to meet national climate and energy targets (roughly double their levels at the time), with a significant proportion directed at replacing aging infrastructure and expanding network capacity.
For governments and grid operators, this has created a compounding problem: not only is new generation capacity required but the grid infrastructure necessary to deliver it is itself in urgent need of upgrade and expansion, with reinforcement programs that carry their own multi-year lead times.
On top of this, generators and power users are increasingly struggling to secure timely grid connections. In some major European cities, it can now take seven to ten years to obtain one. The situation in parts of the U.S. can be equally strained. In regions such as the PJM Interconnection, which spans much of the U.S. East Coast, the queue faces severe delays due to a massive backlog in generation requests, a situation made more difficult by the conflicting priorities of various political constituencies. These challenges are only exacerbated by data center growth.
In response, governments and grid operators worldwide are introducing reforms, including removing speculative projects from grid queues and prioritizing key initiatives.
For example. the UK has recently adopted a “first ready and needed, first connected” model to replace the traditional “first come, first served” approach. In the EU, the European Commission’s Grids Package is aimed at addressing such concerns by making grid connection queues more transparent (i.e., requiring grid operators to publish information on who is in the queue and how long it takes to connect, among other things, in a bid to reduce delays and improve fairness) while accelerating permitting for grid upgrades.
In the U.S., the Federal Energy Regulatory Commission (FERC) and various regional grid operators (independent system operators (ISOs), and regional transmission organizations (RTOs), in addition to state regulatory commissions and legislatures, are all trying their hand at new regulations designed to increase grid stability, control cost, manage the queue process and meet the rising demand for power.
The pressing problem for data center developers though is that the reforms now under way in many jurisdictions will take time to deliver tangible results, and in cases where the local jurisdiction requires BTM or tethered power generation, make energy supply more difficult. “Speed-to-power” and management of local regulation is therefore a key reason why BTM frameworks are attracting increasing interest as a means of securing reliable energy supply, if they can overcome potential obstacles that can delay the route to market.
Challenges facing new-build BTM frameworks
In a BTM framework, a dedicated power-generation facility (whether based on renewable sources, natural gas, or, potentially in the future, scalable nuclear technologies) is co-located with the data center, providing electricity directly to the facility without relying on the public grid network.
This model offers clear potential advantages: it can sidestep congested grid queues, reduce the power user’s exposure to volatile wholesale power prices and certain other network and policy costs and give developers greater control over their power supply, including by acting as a bridge to a longer-term grid-connection. The BTM model can also address growing public concern that grid-connected data center development drives up household utility costs and places additional strain on regional power systems.
However, pursuing a BTM framework introduces its own set of challenges, spanning the choice of the power supply, the commercial structure, and the applicable regulatory landscape.
Type of power supply
While sustainability considerations remain important, securing a dependable power supply which meets the requirements of new data centers is a primary concern for data center developers. As both the technology landscape and regulatory environment continue to evolve, developers must carefully assess the available options along with their site-specific constraints.
For some years, the U.S. tech giants have sought to power greenfield data centers with renewable power, generally wind or solar PV. These companies, particularly Google and Amazon, pioneered frameworks underpinned by long-term corporate power purchase agreements (PPAs), physical and virtual, to effect a hedge in respect of rising power prices, boost renewable energy financing worldwide, and demonstrate “green” credentials.
But in an AI-driven, hyperscale-sized world, these types of arrangements fall short. In addition to the question of whether there is sufficient land available to deliver the power density required, data center developers must reconcile the intermittency of renewable generation with the need for reliability: the so-called “five-nines” rule that a data center requires at least 99.999% uptime.3
In practical terms, if the BTM power supply experiences an outage (or even a disruption to the power quality required by GPU clusters), the data center itself goes down unless adequate back-up arrangements are in place, raising the question of how much redundancy is sufficient without undermining the economic case for a BTM framework. The need for scheduled maintenance of generation assets adds a further dimension, requiring careful planning to ensure adequate continuity of supply.
Solutions for intermittency depend on the commercial structure chosen and may be physical (utilizing supplemental power via a battery energy storage system (BESS) and/or back-up grid connection) and contractual (a considered allocation of guarantees, force majeure and curtailment risks). In reality, the answer is likely to be a combination of these, depending on the project.
Significantly, conventional short-duration BESS may not satisfy “five-nines” and the 24/7 load-matching goals of hyperscale data centers. Long-duration energy storage solutions are becoming more viable, although these technologies are yet to scale and remain costly.
Solutions with built-in redundancy as contingency may also contemplate feeding back excess power into the grid where connection exists, providing a way to offset development costs. The potential for data centers to accommodate more flexible loads to allow demand response participation (in exchange for faster grid connection) is being explored but is not consistent with the current preference of operators to target maximum compute utilization.4
While renewable power will likely continue to make the most sense for parts of Europe, including Scandinavia, and some parts of the Asia-Pacific region, U.S. developers are increasingly pushing forward BTM frameworks with natural gas-fired generation facilities featuring on-demand baseload and peaking capacity. Such facilities can offer critical reliability, albeit with higher emissions than wind or solar PV and with exposure to volatile natural gas prices.
The continued development of small modular reactor (SMR) nuclear technologies, together with associated planning and regulatory regimes, could make SMRs a viable off-grid option for data centers in the medium term (likely from the 2030s).
In principle, SMRs are increasingly seen as a strong match for the energy demands of modern data centers while offering a scalable solution, although technology maturity, cost competitiveness, and regulatory complexity present significant hurdles.
Certainly, the deployment of SMRs will only be available in countries which have elected to support advanced nuclear technologies, an issue which has proven to be politically divisive in many. A further, less-discussed constraint is workforce capacity and expertise; for example, the U.S. may face a shortage of nuclear engineers.
Commercial structure and financing
Commercial structure and associated financing are key considerations in a new build BTM framework, although any approach will be capital-intensive, with the generation facility requiring substantial upfront investment in addition to the data center itself.
In broad terms, there are two commercial models for new build BTM power supply. Under the first, the data center developer contracts with a third-party generator which develops, owns, and operates the generation facility, supplying the data center under a long-term power purchase agreement (PPA).
The commercial arrangements in this model must address the central interdependency between the generation facility and the data centers, as well as any shared infrastructure. Committed data centers’ power demand will be key to the financing of the generation facility, with lenders to the generator focused on the creditworthiness of the data center developer (or its parent guarantor) as offtaker and the revenue certainty of the PPA. Focusing on the differential in useful life of both assets will be essential, and from the generator’s perspective, how to derive return within that constraint via the tariff.
Where the generator is itself leveraged, lenders to the data center developer will also seek comfort tha the generator’s financing arrangements do not create enforcement risks that could disrupt supply. The parties will also need to consider what happens to the generation facility if the data center fails or ceases to operate and the generator has no grid connection or alternative power purchaser – the “stranded asset” risk for the generator.
As the data center developer does not own or control the generation facility, its lenders will focus on the performance obligations of the generator under the PPA, feedstock demand and risk, the remedies available to the data center developer if the generator fails to deliver, and the extent to which interruption of power supply could impair the data center’s revenue and the borrower’s ability to service debt. The tariff structure and certain operational oversight and control over development decisions affecting the tariff in a pure captive world will be a critical focus from the data center developer and its lenders’ viewpoint.
These factors translate to key contractual issues such as the allocation of rights and responsibilities at the site level (particularly where the generation facility occupies a significant portion of the available land in order to service the load), demand risk (including the treatment of curtailment, force majeure and under-utilization) and the allocation of delay risk given the realistic prospect that the construction program for start-up of the generation facility may not align with that of the data center. Long-stop dates, delay liquidated damages and termination rights will need to be carefully calibrated in response. Robust termination and step-in provisions, together with carefully structured security packages, will be essential to making these arrangements financeable.
Under the second model, the data center developer takes on the development of the generation facility itself, either directly from the outset as part of the overarching data center development, or by entering into a build-transfer agreement (BTA) with a third-party developer under which ownership of the generation facility transfers to the data center developer at an advanced stage of construction.
A BTA may allow the data center developer to accelerate delivery of generation capacity by acquiring a project that is already well advanced, but this route is likely still to involve significant upfront expenditure; the purchase price is typically structured in staged payments that, in economic terms, amount to the data center developer funding the later phases of development and construction.
The data center developer also assumes exposure to design and construction risk under the BTA over which it may have limited practical control. BTAs are seemingly less common in the U.S. market, with more traditional joint-venture arrangements transferring generation facilities at the operational stage, as opposed to during construction.
Where the data center developer is undertaking both data center and generation facility development, this will of course have significant cost and timing implications and project-on-project risk.
The developer and its lenders will be focused on contractual risk allocation (and associated credit support) between the developer and its counterparties, in particular with its construction contractor(s), in order to mitigate any delays in data center operations commencing and revenue being generated.
The choice of technology will also affect considerations. For example, gas-fired generation facilities will require fuel-supply infrastructure, with associated land-rights and permits and an allocation of shared facilities if pipelines and other equipment are utilized across projects.
Lenders facing concentrated exposure to a single borrower across both the data center and generation assets will scrutinize the developer’s operational capability and track record in managing power-generation infrastructure alongside its core data center business.
Lenders are likely to require the appointment of an experienced third-party generation facility operator and a carefully negotiated long-term operation and maintenance agreement as a condition of financing (matters on which data center customers will also want reassurance). Stranded asset risk is again a consideration for the generation facility.
An associated issue across both models is the potential bottleneck in procuring key generation equipment, with gas turbine makers recently reporting soaring backlogs and delivery times years after order placement (not counting any supply chain challenges arising from the Middle East conflict).5
Competition with other power-generation developers for essential equipment could lead to less favorable contract terms and increased pricing, on top of the impact of critical component shortages and trade barriers (including U.S. duties on Chinese solar panels and batteries and retaliatory export controls).
This race to purchase gas turbines is leading to limited purchaser leverage and stocking of supplies, which in turn is fueling a secondary market offering even less contractual protection from brokers and resellers. Reports of certain energy solutions providers repurposing jet engines to power data centers demonstrates the current market temperature.
In terms of financing models, the interdependency between data center and generation facility increasingly means that data centers with BTM need to be conceived and financed less as straightforward real estate developments and more as integrated energy and infrastructure projects, with the associated complexity that entails.
Combining the data center and generation facility financing may reduce the complexity of the debt-raising process but may complicate credit issues, not least because lenders accustomed to financing data centers as real estate assets may lack familiarity with the risk profile of power generation, and vice versa. In a truly captive structure, stapled/coordinated financing with the same lender group may in some markets be inevitable.
In any event, lenders will test the sufficiency of the aggregate security to cover the debt service requirements in downside scenarios, whether the exposure is to a single integrated borrower or divided across separately owned assets.
Regulatory regimes not designed for BTM frameworks
Beyond the choice of generation technology, a BTM developer must navigate the regulatory and licensing regimes applicable to power generation and supply in the relevant jurisdiction.
Legacy power systems and associated regulatory frameworks have been largely based around generators connecting to centralized grid infrastructure, with electricity transmitted across public grids and then distributed to end-users.
While almost every country’s power sector exists at some stage of reform to bring it in line with more modern demand and supply dynamics, on-grid powering still represents the norm and BTM frameworks may fall into existing regulatory gaps.6
In some countries, deploying a BTM framework will not yet be feasible due to regulatory constraints. In others, regulatory requirements (such as the need for the generator to hold a license to perform various restricted activities including distribution of electricity to the data center) may create additional obstacles, costs and risks. These considerations require careful, jurisdiction-specific analysis and, in certain cases, first-of-a-kind discussions with system operators and regulators.
Another core issue is the extent to which the data center is legally required to maintain back-up power or grid connection. In Ireland, for example, data centers use about a quarter of the national electricity supply, leading authorities to require on-site back-up power or battery solutions. Conditions may be attached to grid connection, including the allocation of network charges and surcharges. Regulators in several jurisdictions are actively grappling with these issues as part of a wider debate about how to treat new digital infrastructure and its associated impact on electricity costs and public water supplies.
Across Europe, for example, the principle that BTM-generation facilities benefiting from grid backup should contribute more to network costs is reportedly gaining traction. In the U.S., FERC’s December 2025 order7and PJM’s subsequent proposals require large co-located loads above certain thresholds to pay for transmission services, marking a significant departure from longstanding practice.
As this and other cost issues continue to play out, stakeholders will need to remain alert to potential shifts in the financial case for BTM.
Looking ahead
With grid connections becoming harder to secure and connection timelines lengthening across major markets while the impact of reforms is awaited, BTM frameworks are likely to play an increasingly prominent role in powering the rapid growth of data center capacity.
However, BTM is not a panacea. The viability of any given arrangement will depend on a complex interplay of factors: the available generation technology, the maturity and flexibility of the local regulatory regime, the creditworthiness and risk appetite of the parties involved, and the evolving expectations of lenders and investors.
The dynamics of shifting energy security priorities, critical mineral supply chain realignment, diverging sustainability regulation, and geopolitical instability add further layers of complexity for developers operating across multiple jurisdictions.
Footnotes
1. https://www.iea.org/reports/key-questions-on-energy-and-ai
2. https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai; https://www.iea.org/reports/key-questions-on-energy-and-ai
3. The rising challenge of powering data centres | Oxford Economics
4. UK-first trial of AI Grid Technology Successfully Demonstrates the Ability of Data Centres to Adjust Power Needs | National Grid
5. https://oilprice.com/Energy/Energy-General/US-Power-Boom-Triggers-Global-Gas-Turbine-Shortage.html
6. The Emerging Regulatory Framework to Power Data Centers
7. https://www.ferc.gov/news-events/news/ferc-directs-nations-largest-grid-operator-create-new-rules-embrace-innovation-and
