If you have any questions about the content of this article or want to discuss your behind-the-meter power needs, please visit Prime Power Inc.’s website, click here. Prime Power Inc. is an IPP providing both land siting/development services and on-site power solutions.
Executive Summary
- The existing grid was not designed for the scale and speed of power demanded by AI data centers. These data centers continue to scale their deployments and increase power densities, requiring expensive grid upgrades that can take 5-7+ years to build. After being connected, several grid operators, including ERCOT and PJM are reserving the right to force load curtailments and are even contemplating the right to “kill switch” the load entirely, forcing a data center shutdown or transitioning to on-site generation
- An on-site natural gas-fired power solution can be constructed much more rapidly in 18-24 months with the ability to scale solutions to multi-GW applications. This better matches the AI data center communities’ “Speed to Power” philosophy.
- On-site power can operate completely islanded from the grid, with ability to offer 15-20 year PPAs in a fully islanded configuration across the full term. Operational capabilities can exceed well beyond that term with adequate maintenance and component replacements. The solution can serve as a bridge solution if a grid-connection is provided in the future, then transitioning to transmission wholesale participation, emergency power services to the data center, and/or partial PPA arrangement.
- Building and operating behind-the-meter power solutions in an islanded Microgrid architecture is complex and requires the proper mix of generation equipment, AI load modulation control, robust operational & maintenance strategy, and a team with execution experience for islanded power. High power quality and stability can be maintained, and the right, seasoned development teams can deliver on such a solution
- Recent federal policy guidance encourages AI hyperscale companies to build their own co-located generation. This includes the White House’s Ratepayer Protection Pledge, where AI companies, including Amazon, Microsoft, Google, OpenAI, Oracle, and xAI signed the pledge to protect rate payers by ensuring AI data center energy demands do not raise electricity costs on American households. Co-located, behind-the-meter generation solves the issue by serving the load directly, without requiring grid upgrades that often fall on the rate payers (general public).
- Introduction
The explosion of data centers has created the most significant increase in power demand in modern history. AI data center developers and their tenants have traditionally relied on the grid to connect their facilities however, the volume of large load requests across the country has dramatically constrained the grid’s capability to keep up. Bloomberg projects peak demand attributable to data centers could reach 106 GW by 2035 which would represent more than 15% of the total U.S. demand.
Many analysts currently believe that data center growth should continue to rely on the grid. They argue that the best solution is to implement more regulatory policy, increase distributed energy storage, incorporate demand response strategies, and grid-enhancing technologies. While these methods have historically had success in relieving grid constraints, the booming AI load growth is proving them to be vastly inadequate. The existing grid was not designed to support that rapid speed, concentration, and scale of power demanded by modern AI data centers.
Federal polices, including the White Houses’ Ratepayer Protection Pledge, encourage data center developers to pursue local or on-site generation resources, rather than solely relying on the electric grid. On-site generation can significantly increase speed to power, protects rate payers from price hikes, mitigates electric grid curtailments, and ensures more data centers can be deployed across the U.S. to compete with other nations, including China. This article covers other benefits of on-site generation, including a technical overview of how on-site generation technology can reliably serve AI data center load completely islanded from the grid over a 20+ year period.
- The Economic Importance of Data Center Deployment
The rapid deployment of data centers has become a matter of national economic and geopolitical importance. Advanced computing infrastructure underpins artificial intelligence (AI), cloud services, financial systems, logistics networks, and much of the modern digital economy. The United States and China are currently engaged in a strategic race to dominate AI capabilities, which rely heavily on access to large-scale compute capacity and reliable power. Training frontier AI models requires massive computational resources; for example, OpenAI’s GPT-4 is estimated to have required tens of thousands of GPUs and millions of GPU-hours of compute during training. As a result, the availability of high-density data centers with abundant electricity has emerged as a critical bottleneck for AI development. U.S. policymakers increasingly view data-center capacity as strategic infrastructure analogous to semiconductor fabrication plants or energy systems.¹
Beyond national competitiveness, data centers are foundational to the broader economy because they support the digital services that drive productivity and innovation across industries. Cloud computing, e-commerce, financial services, advanced manufacturing, healthcare analytics, and AI-driven automation all rely on scalable data-center infrastructure. The economic footprint of this sector is already substantial. According to Harvard economist Jason Furman, information-processing equipment and software investment, largely tied to AI data center buildout, accounted for roughly 92% of U.S. GDP growth in the first half of 2025, despite representing only about 4% of total GDP.² Further, studies estimate that each direct data-center job supports more than six additional jobs elsewhere in the economy (*), reflecting the sector’s strong multiplier effects across digital and infrastructure industries. At the same time, demand for compute is accelerating rapidly due to generative AI and large-language-model workloads, which can require 5–10× more power per rack than traditional cloud workloads.³ This surge in demand is expected to significantly expand the economic importance of data-center infrastructure over the coming decade.
- The Grid Connection Challenge
Grid Connection Challenge #1 – Clogged Interconnection Queues
The first challenge to connecting to the grid is the massive volume of large load requests sitting in the interconnection queues of the grid operators (ISO/RTO) across the U.S. The clog of projects has constrained the ISO/RTOs and electric utility providers capability to efficiently process, study and ultimately build the necessary upgrades to connect these project. The grid operator in Texas (ERCOT), for example, has an incredible volume of large load requests. As of March 2026, there were 238,629 MW of large load requests in the ERCOT queue, representing 391 different projects. 225 of these requests were received in 2025 alone and it is estimated that 75% of the requests are data centers.
The grid operator in the northeast, PJM, covering 13 states in the northeast, does not publish a large load request queue, however it is estimated that there are as many as 200 requests currently in the queue, representing approximately 32 GW, with 90% of those requests being data centers.
To address the queue issue, several of these grid operators are forming new policies intended to clean up the queue and reduce the number of non-serious or duplicative requests. ERCOT enacted Senate Bill 6, which aims to address these items, proposing higher financial security requirements up to $100,000 per MW, with at least half being non-refundable. These new rules are still being drafted under draft rule-making 58481, with an unclear timeline on when they will become effective and if projects already in the queue will be subject to the new rules. Although these types of policies should help reduce the queue problem, they are mostly early in development, and remain unclear if ISO/RTOs will force stricter requirements on projects already in the queue. The ERCOT and PJM rules are believed to become effective by the end of 2026 however that remains mostly unclear.
SPP, the grid operator which serves 14 states in the Midwest, has implemented a new High Impact Large Load Generation Assessment (HILLGA) process that studies large loads paired with nearby generation. This is a separate dedicated queue process that allows for an expedited study process of as little as 90 days. The load and the generator are studied in parallel and receive expedited study treatment as SPP is recognizing the capacity benefit of the local or on-site generation. The generation reduces the impact that the large load has on the local grid. Other grid operators and utilities are also considering such programs.
Grid Connection Challenge #2 – Long Interconnection Timelines
A common industry consensus suggests current the timelines to connect to the grid include 2-3 years for studies and 3 to 5 years for grid upgrades. This produces an overall timeline of 5 to 7 years from initial large load application to energization date. However as large load requests continue to surge and utilities struggle to perform the necessary upgrades, many cases produce an overall timeline of up to 10 years. This mismatch between AI data center implementation and transmission grid upgrade schedules is driving AI companies to consider on-site generation to meet their expansion ambitions.
After the large load customer submits their request to the serving utility or ISO/RTO, the utility will perform a series of studies typically including steady-state, short-circuit, facilities and other supporting studies. These studies quantify the load impact to the local grid and the schedule & cost to perform required grid upgrades. Following these studies, an Interconnection Agreement is executed to move into the grid upgrade construction period where additional financial securities and cost in aid of construction (CIAC) is due.
| Year 1 | Year 2 | Year 3 | Year 4 | Year 5 | Year 6 | Year 7 | |
| AI Data Center Construction | Planning & Engineering | Construction | |||||
| Co-Located Generation | Planning & Engineering | Construction | |||||
| Grid Interconnection | Studies | Studies | Engineering | Construction | Construction | Construction | Construction |
Transmission grid operators are transitioning to “cluster” study processes where large load requests are studied together, rather than a serial first come, first served basis. This is intended to speed up the study timelines. ERCOT has followed suit and is proposing their first “Batch Zero” beta cluster study process which is tentatively set to be in place in Q3 2026. Despite these efforts to speed up the study timeline, this does not address the volume of grid upgrades that need to be performed to accommodate the projects. TSP driven engineering and construction timelines are expected to remain up to a 5-7 year process or more, depending on how many projects continue to make it through the queue into the future. It does not appear that the volume of requests will slow down anytime soon.
Grid Connection Challenge #3 – Forced Curtailments
Grid operators must protect their system from abnormal system conditions and will utilize curtailment of non-critical large load customers. This means ISO or the serving utility reserves the right to call upon the large load customer to reduce their power demand to assist in alleviating the emergency system condition. Large loads can significantly contribute to the following conditions:
- Thermal Overload – too much power demand can cause a transmission line to exceed its thermal rating, leading to overheating and line sag. This is the most common reason for forced curtailments
- Inadequate Local Generation Capacity – forced curtailment may occur when extreme weather events or concurrent maintenance on local generators create a capacity shortfall.
- Voltage Stability & Reactive Power Limits – large loads the scale of AI data centers can contribute to voltage destabilization in certain vulnerable areas of the grid. Reactive power support can help bring voltage back to within acceptable limits, however when this is inadequate, forced load curtailments must occur to prevent complete voltage collapse.
- Frequency Deviation – sudden swings in generation or load impacting the supply/demand balance can throw the grid frequency outside of acceptable limits. If the frequency drops below a common threshold of 59.7 Hz, grid operators must immediately drop demand. ERCOT is beginning to impose frequency ride-thru and load shedding capabilities to aid in maintaining normal frequency range.
All together these potential issues create an uncertain situation for large loads that intend to rely on the grid for their supply. As demand increases and grid operators struggle to balance their systems, curtailment may become standard operating procedure that impacts grid reliability and ultimately operating revenues for AI data centers.
Grid Connection Challenge #4 – Consumer Power Price Growth
The rapid growth of electricity demand, driven in large part by grid-connected AI data centers, is beginning to influence wholesale power prices and long-term electricity planning. After more than a decade of relatively flat electricity demand in the United States, utilities and grid operators are now forecasting significant load growth due to electrification and large-scale computing infrastructure. According to the International Energy Agency, global data-center electricity consumption is projected to increase from ~415 TWh in 2024 to ~945 TWh by 2030, driven primarily by AI workloads. In many U.S. regions, hyperscale data centers now request 100–1,000 MW interconnections, adding step-changes in demand that tighten generation reserves and require major investments in generation and transmission infrastructure. These dynamics can place upward pressure on both wholesale energy prices and capacity market prices. For example, capacity auction prices in markets operated by PJM Interconnection increased 10x in 2 years [2024-2026] as demand forecasts rose and supply additions lagged. As a result, electricity costs in high-growth regions are increasingly shaped by the pace of data-center development, making access to reliable and cost-stable power a critical constraint for digital infrastructure expansion.
- On-Site Power Generation – Overview & Benefits
On-site natural gas-fired generation provides a reliable alternative to grid supply and achieves a faster schedule to power. As discussed in the technology overview section, the generation equipment, medium voltage architecture and Microgrid controls are designed to support the AI load, effectively replicating grid power supply and capable of long-term islanded operation. Implementing fully islanded natural gas generation solutions provide:
Deployment Capability
- Speed to power. An on-site power solution can be deployed in two years or less following some pre-planning and providing an NTP to the group delivering the power solution such as Prime Power Inc.
- On-site power solutions can increase the number of AI data center deployments. Many sites across the country have access to on-site power resources such as natural gas but no grid infrastructure or grid capacity. Onsite generation allows data center operators to expand site selection options in an environment of limited suitable sites with large grid access.
Project Economics
- A dedicated power solution that can offer reliable power over a 10-20 year PPA contract with operational capability well beyond 20 years with adequate maintenance and component replacements. Long Term Service Agreements (LTSA) from the generator, BESS and other OEMs can support the full term of the PPA and any tail participation in the wholesale market. These LTSA’s can provide guaranteed uptime/availability guarantees and liquidated damages, which support and strengthen the overall on-site power solution offering and contract terms to the data center.
- Onsite generation offers a structural hedge against rising grid electricity prices by decoupling large loads from transmission constraints and capacity market volatility. While wholesale power prices in many U.S. regions are increasing due to load growth and infrastructure bottlenecks, forward curves for U.S. natural gas remain relatively low and stable, reflecting abundant domestic supply from shale basins such as the Permian and Marcellus. This divergence allows BTM projects to lock in long-term fuel costs while avoiding exposure to escalating capacity and transmission charges embedded in retail power prices.
- Onsite generation gives developers control and economic ownership of system management equipment. Grid-connected projects often bear significant “system upgrade” costs, such as transmission expansions and interconnection upgrades, which can be socialized over time and create a free-rider dynamic where early movers disproportionately fund infrastructure that later users benefit from. In contrast, on-site generation systems are more modular, faster to deploy, and easier to scope, enabling tighter cost control and shorter development timelines. As a result, BTM gas-based power solutions can provide more predictable, lower-cost electricity for large, power-dense loads such as data centers, particularly in constrained markets where grid expansion is lagging demand growth.
- A flexible on-site power solution that can either:
- Provide islanded power to the data center over a 10-20 year PPA. No connection to the grid is necessary. Alternate PPA tenors can be discussed.
- Provide initial bridging power as a fully islanded solution prior to a grid connection. Once the grid is connected, the generation asset can transition to a hybrid offering normal power or emergency backup power to the data center, while also offering energy to the transmission system wholesale market (exporting power).
Site Operations
- A reliable source of natural gas fuel supply that can be sized to support the entire AI facility load. Obtaining a firm transport supply contract bolsters reliability, which has been shown to offer 99.9%+ reliability over a 15-20 year contract. If desirable, an on-site backup fuel solution can be offered.
- Protection from grid disturbances and forced curtailments. Whether fully islanded or connected to the grid in parallel with the facility, the on-site power solution can provide sustained, reliable power. If ultimately a grid connection is added in the future, the on-site solution can provide supplemental power if the grid operator imposes load curtailment.
- On-Site Power Generation – Technology and Load Control
An on-site power solution will typically resemble the high level diagram below where various components are interconnected at medium voltage (13.8 or 34.5 kV) to deliver power to the data center campus. These technologies and control strategies will be explored in greater detail in a future publication by Prime Power Inc.
- Natural Gas Generators
The generators serve as the backbone and primary source of power in an islanded system, typically generating at 13.8 kV although there are variations. Often referred to as “prime movers”, the generators typically take the form of reciprocating engines, combustion turbines and/or fuel cells, with future consideration for SMR’s. Each type of technology has pros and cons, which will be explained in a future publication by Prime Power Inc. The generators are typically set up in a simple-cycle arrangement where they deliver power. However Prime Power Inc. can consider alternate arrangements such as combined heat and power or a combined cycle approach. These are strategies to recover heat from the generators to make useful heating or cooling for the data center facility. Although this does improve the overall efficiency of the system, it can impact economics in a negative way, so performing an economic evaluation on the options upfront is critical.
The generators often include emissions control technology, such as oxidation catalysts or selective catalytic reduction (SCR) systems to achieve successful air permitting in particular regions. Federally registered pollutants (NOx, CO, VOCs, etc.) are eliminated by 90% or more.
Generator equipment availability in the market is extremely thin. Having line of site or control on the generator supply is critical to ensuring power delivery schedule. Legitimate power developers in this data center space are the ones who have a clear pathway to generator delivery, and are willing to put forward their own capital to secure the equipment. Prime Power Inc. pursues this strategy of securing generation equipment, gas supply and placing interconnection requests as part of the overall strategy to create a strong project that is NTP-ready with a firm schedule to COD.
- Battery Energy Storage System (BESS)
The BESS can provide several potential use cases with the primary function in this application being smoothing the AI load profile. This is discussed in more detail in the next section. The BESS are typically pad-mounted in outdoor rated enclosures, similar in resemblance to a shipping container with improved aesthetics. Other use cases include injecting power to optimize generator loading & efficiency, generator black-start assistance, potential wholesale market participation if the system becomes grid-connected, and emergency power delivery to the data center if grid-connected and utility power goes down. The design can also consider alternative forms of load smoothing strategies including capacitors and flywheels which each offer their own distinctive capabilities in load smoothing support. These are less common in serving Tier III or IV data centers, but can be considered.
- Medium Voltage Switchgear and Distribution
The MV design and architecture will follow the general theme and intent of the Uptime Institute’s reliability standard. The power plant solution can be adjusted to align with a data center customer’s preference for Tier I, II, III or IV design. Regardless of the specific design standard, the power solution shall include redundant generators, properly sized BESS, MV redundant MV equipment and redundant power distribution pathways that can accommodate concurrent maintenance and unforeseen component failures. A common approach is to generate power at 13.8 kV, then step-up the voltage to 34.5 kV for delivery to the data center campus.
- EMCS & Microgrid Controls
First and foremost, a behind-the-meter power (BTM) solution that provides speed to power ahead of a grid connection must be capable of fully islanded operation. Without the grid as a voltage and frequency reference source, the generation, battery storage system and inverters must create its own grid, with stabilized voltage and frequency. An Energy Management Control System (EMCS) acts as a supervisory and optimization layer that provides economic dispatch, forecasting capabilities (load, fuel usage, etc.) an operator interface, analytics and reporting. A Microgrid Controller is responsible for maintaining a coordinated and synced operation across the fleet of equipment in island mode as well as operation in grid-parallel mode should a future grid connection be pursued. This controller dispatches individual generators, maintains stability (voltage and frequency), and manages the AI load smoothing function. In summary, the EMCS instructs the Microgrid controller what should happen and the Microgrid controller sends/receives control signals to dispatch equipment and maintain power quality.
- Load Smoothing and Power Quality
A behind-the-meter power solution must address each of these challenges to ensure reliable power delivery to the data center. This requires an engineering and execution team with deep experience designing and building behind-the-meter power solutions that have the capability to island without a grid connection. There are approximately 700 Microgrids operating in the U.S., and the teams that have this capability, including Prime Power Inc., are able to adapt these systems to provide a reliable generation solution for a data center.
- AI Step Loads
Some analysts argue that sudden AI load changes can easily destabilize an islanded on-site power system. While this does present a challenge, this is precisely where a hybrid solution of modular generators, battery energy storage and advanced Microgrid controls are able to accommodate the load ramps. BESS is able to inject power with a sub-second response (milliseconds), creating a buffered and smooth load profile for the generators.
The natural gas generators themselves are able to handle step changes in load within certain tolerances, to assist in meeting the swing in load as well. Combustion turbines offer high rotational inertia, which is able to absorb the initial mismatch between mechanical input and electrical output. Generally the turbines can handle a 10-20% step load change in island mode with modest frequency deviation. From there, a 10-30% ramp rate per minute in output can be expected. Reciprocating engines are able to handle step changes of typically 20-30% of rated capacity, immediately increases torque in response, and offers 20-50% per minute ramp rates.
Although the bulk transmission grid is often considered large and stable, it should be noted that the increasing penetration of renewables is reducing the rotational inertia of mechanical generators and the grid stability they provide. Certain areas of the grid are now considered weak and less stable, and this issue is growing, not shrinking. An on-site power solution paired with battery storage, can not only compete with the grid, but can outperform it so as long as the design incorporates the proper equipment, controls, and an execution team that can commission and operate the system.
- Power Quality
Non-linear, ramping loads from the AI data center can present certain power quality conditions, particularly the potential for harmonic disturbances which the design of the on-site power solution must consider. When properly designed and accounted for, predictable performance of the generation facility can be expected, whereas a grid connected sites can inherit uncontrolled upstream distortion.
The battery energy storage system is an inverter-based system, meaning they inherently offer power conditioning capability. The BESS inverters are able to actively filter harmonics, which performs in real-time and eliminates the need for large passive filters. Also, hyperscale data centers will often employ UPS systems that aid in decoupling the IT load from the on-site power plant. This significantly reduces harmonics back feeding to the generation facility.
Bottom line, harmonics are an engineering problem that can be solved, and islanded power systems can generate and deliver more predictable power with higher power quality than the grid in many cases.
- Fault Current Protection
Fault current is simply the higher level of current that occurs when the system is short-circuited or experiences a ground fault. The legacy grid was built around synchronous generators and high fault levels. For this reason, most legacy protection schemes will utilize high fault current design to ensure protection devises (relays, fuses) are able to easily sense the 10-30x normal current.
An islanded power system can experience lower fault current conditions, primarily from the BESS which is an inverter-based resource and offers less fault current contribution (1.1 – 2x normal current). However, given the proposed mix of natural gas generators and BESS, available fault current is increased beyond the minimal levels offered by the BESS. Additionally, a differential protection scheme designed for lower fault current is employed, which incorporates more sensitive relay settings and adaptive, intelligent settings to detect low magnitude faults. Similar to maintaining power quality, these are engineering items that can be easily solved, and doing so is not a new concept. Microgrid systems have been employed for decades, and are being adapted to serve AI loads successfully
- Summary
AI data centers have become now become strategic and economic infrastructure, providing critical support to our financing systems, economic growth, and national competitiveness. The scale and speed of deployment has left the grid both vulnerable and unable to keep up with the necessary buildout to support data center deployments. Traditional reliance on the power grid has created an unprecedented volume of large load requests in the interconnection queue. The massive backlog in projects to be studied and ultimately accommodated for with grid upgrades has create 5-7+ year timelines for data centers to get connected. In some cases, this timeline can reach to 10 years or more in certain pockets of the country, particularly near urban centers where data center develops seek to deploy AI computing.
On-site generation provides the reliable alternative to the grid, offering rapid deployments in as little as 18-24 months, with ability to scale to multi-gigawatt applications. An islanded power solution can support a long-term strategy, with an IPP such as Prime Power Inc delivering a white-glove turnkey development and construction service and wrapping the power delivery under a 15-20 year PPA. If a grid connection is ultimately desired, the solution can be modified to a shorter tenor PPA, with transition to the grid and the power solution delivering transmission wholesale participation, emergency back-up services to the data center, and/or partial PPA arrangement.
Modern designs for islanded Microgrids ensure reliable power delivering, matching reliability metrics pursued by the data center. The power architecture will include redundant generators, redundant power feeders, and a Microgrid control system ensuring synchronized operation in both island or grid-parallel modes of operation. Power quality and stability are maintained through a combination of load smoothing equipment, typically battery storage, with inverter-based technology providing power conditioning and harmonics filtration. UPS systems provide further conditioning and assist in separating the IT from the generation facility. Technical challenges associated with islanded power systems are known and well understood. Maintaining stable power, voltage and frequency can be effectively addressed through established engineering practices.
To summarize, behind-the-meter power generation is the solution to support the ongoing rapid deployment of data centers. Scalable on-site power infrastructure can be deployed, and speed to power can be achieved.
About the Author:
Erik Norgren
VP, Development & Engineering
Mr. Norgren is a registered professional engineer (PE) and brings 15 years of expertise in both natural gas power and renewable development including BESS. Erik has developed both behind-the-meter and front-of-meter projects at DG and utility-scale. He is a subject matter expert in islanded generation solutions and natural gas power solutions including simple-cycle and combined heat and power. His works spans across DoD installations, NASA, private IPPs developments, and behind-the-meter C&I. Erik’s experience across different client and technology types ensures that Prime Power is able to offer customer-fit, reliable power solutions for our clients.