The Case for Grid-Connected Data Centers: Cleaner, More Affordable, and Fast Enough

AI is driving rapid growth in electricity demand, and data center developers are increasingly looking for ways to secure power faster than the grid can provide it. In major data center markets, new projects can face years-long waits for grid connection and the generation and transmission investments needed to serve them. That creates a powerful incentive for developers to find alternative sources of electricity.

Increasingly, some developers are turning to behind-the-meter (BTM) natural gas generation. Developers have announced over 110 gigawatts (GW) of BTM gas generation at data centers, equivalent to about 20% of the existing installed gas capacity across the country, with much of the proposed capacity relying on gas-fired technologies that can be deployed more quickly than conventional power plants.

The appeal is understandable: for a data center developer, getting power even a year earlier can be enormously valuable. But speed is only one dimension of the power-supply decision. Data center developers, utilities, regulators, and policymakers also need to consider emissions, electricity costs, and the broader effects of new loads on the power system.

On those measures, grid connection tends to offer advantages for both data centers and other electricity customers. It can mean lower emissions, lower costs for data centers, and lower bills for other customers. Although BTM gas can often provide power sooner today, there are ways to make it easier for data centers to connect to the grid without sacrificing speed.

Cleaner: Grid connection can substantially reduce emissions

Grid power is already supplied by a diverse mix of resources. In 2025, 42% of US utility-scale generation came from renewable and nuclear resources. And the resources now being added to the grid are substantially cleaner still: solar, battery storage, and wind account for 93% of planned US capacity additions in 2026.

The makeup of the grid, combined with the relatively inefficient generation being used for BTM gas, makes grid power the cleaner option. We estimate that only a fraction of the 110 GW of proposed BTM gas will come online by 2030 — in line with EPRI’s medium data center demand scenario. Still, BTM gas serving data centers could produce approximately 39 million metric tons of CO₂-equivalent emissions in 2030. Serving the same load from the grid would produce about 22 million metric tons, a 43% reduction.

The technologies being prioritized for BTM gas can carry additional environmental risks. Many projects favor aeroderivative turbines for faster deployment, even though they require over 40% more fuel per unit of electricity than grid-connected combined cycle gas plants. That increases both combustion-related carbon dioxide emissions and upstream methane emissions per unit of useful energy generated. Other projects rely on reciprocating engines, which are also less efficient than combined-cycle plants, can emit higher levels of hazardous local air pollutants, and leak methane, further increasing their climate impact.

Grid connection does not guarantee low-emissions electricity. The emissions consequences depend on the regional power mix and the resources added to meet new demand. But it creates more pathways to serve that load with an evolving portfolio of generation, storage, and demand-side resources — including options for data center customers to fund incremental clean resources directly.

More Affordable: Grid connection can lower costs for data centers and other electricity customers

Grid power is generally cheaper than building and operating dedicated gas generation. RMI research finds grid electricity costs of $55–$115/MWh across eight large load tariffs. By comparison, recent estimates from Rabobank put the all-in cost of BTM gas generation for data centers at roughly $100–$165/MWh. These figures will vary by location, fuel prices, utility tariff, and project design, but the basic tradeoff is straightforward: Although BTM gas can offer faster access to power, developers generally pay a premium for that speed.

Grid connection can also create affordability benefits for other electricity customers. One way is by increasing the efficiency of the existing grid. Recent Brattle Group research estimates that increasing grid utilization could reduce US electricity bills by $110–$170 billion over the next decade by spreading the cost of existing infrastructure across more electricity sales.

Grid connection can also help fund new resources that provide broader system value. In Minnesota, for example, Xcel Energy’s proposed agreement to serve a new Google data center includes a Clean Energy Accelerator Charge supporting 1,900 MW of clean energy resources. Xcel estimates the agreement could produce roughly $1.1 billion in benefits for existing customers, subject to regulatory approval.

BTM gas largely forgoes these opportunities while creating a new source of fuel demand and associated risks. We estimate BTM data centers could consume approximately 742 trillion Btu (TBtu) of gas in 2030, equal to about 5.5% of the gas used for power generation today, and with much larger impacts in some states. Concentrated demand growth could strain pipelines and fuel markets, contributing to infrastructure constraints, higher prices, and greater exposure to fuel-price volatility.

Grid connection does not automatically lower bills for other customers. Poorly structured tariffs or speculative investments can shift costs to existing ratepayers. But when designed well, grid connection creates an opportunity to turn data center growth into investment in shared infrastructure and resources — an opportunity largely lost when a data center builds an isolated fleet of gas generators.

Fast Enough: Narrowing the speed-to-power gap

Despite these tradeoffs, developers are considering BTM gas primarily because grid connection is often too slow.

Large data centers can be built on timelines that are considerably shorter than the time needed to develop major generation and transmission infrastructure. For companies competing to deploy AI computing capacity, the economic value of reaching operation sooner can outweigh differences in fuel costs or emissions.

BTM gas therefore has a real speed-to-power advantage in some markets today. But what if grid connection became fast enough that developers no longer need to accept such a large tradeoff?

Three strategies, shown in the Exhibit below, are particularly promising.

1. Expand flexible interconnection. Large loads do not necessarily need firm service for their full peak demand from their first day of operation. Under flexible or non-firm service, data centers can connect using available grid capacity and agree to reduce consumption during constrained periods while longer-term upgrades are completed.

That model is beginning to move into practice. In 2026, FERC accepted Southwest Power Pool’s Conditional High Impact Large Load Service, which allows qualifying loads to receive non-firm service before all generation resources and network upgrades needed for firm service are complete. FERC has also directed PJM, MISO, CAISO, ISO-NE, and NYISO to propose their own flexible interconnection frameworks for large loads.

2. Use alternatives to conventional network upgrades. The choice is not always between waiting years for a transmission project and building on-site gas. Batteries, grid-enhancing technologies, advanced conductors, and demand flexibility can sometimes relieve the specific constraint preventing a data center from connecting.

Portland General Electric, for example, has combined customer-funded batteries, advanced analytics, dynamic line ratings, and evaluation of advanced conductors and power-flow controls as part of its large-load strategy. The utility has reported identifying more than 80 MW of capacity that could be made available without building new infrastructure, and construction is already underway on a battery project that will accelerate data center interconnection.

3. Let customers bring capacity to the grid, not just behind their meter. In some cases, large customers can fund or procure incremental resources that allow them to connect sooner. Importantly, those resources do not necessarily need to be located at the data center itself.

“Bring your own resource” frameworks could allow customers to invest in generation, storage, transmission, demand flexibility, or distributed resources elsewhere on the system and receive credit when those resources address the constraints limiting their connection. Georgia Power’s Customer Identified Resources program offers one emerging example, allowing large customers to identify renewable generation projects for utility procurement. Virtual power plants are another potential model: a data center could finance flexible capacity from thousands of homes or businesses and use the resulting system capacity to help support its own electricity demand.

These approaches will not make every data center connection fast. Major new transmission and generation will still be needed as electricity demand grows. But they can expand options for timely service, making grid connection a viable choice for a greater number of data center projects.

Faster grid connection can help capture the benefits of grid-connected growth

Grid connection can offer important advantages for data center developers, electricity customers, and the broader power system. It can provide access to lower-cost electricity, reduce emissions, and create opportunities for data center investment to support shared generation, transmission, and distributed energy resources.

The challenge is timing. When grid connection takes years longer than on-site alternatives, developers may choose BTM gas even when grid-connected service would provide better long-term outcomes.

That makes faster grid connection key. Flexible interconnection, alternatives to conventional network upgrades, and pathways for large customers to fund needed capacity can help shorten the wait for grid service. The faster these pathways become, the easier it will be to capture the benefits of grid connection without sacrificing economic competitiveness.

We would like to thank Sarah Wang, Lauren Shwisberg, and Mark Dyson for their contributions to this article.

Methodology

Our emissions estimates are based on EPRI’s medium data center demand scenario, which projects 596 TWh of US data center electricity consumption in 2030. We assume that BTM generation serves 25% of incremental data center electricity demand. This results in approximately 92 TWh of BTM generation in 2030. Assuming a 75% utilization factor, this would translate into approximately 14 GW of data center capacity served by BTM generation.

For the BTM gas case, we translate that electricity demand into natural gas consumption using a weighted-average heat rate based on the mix of generation technologies deployed across operating, under-construction, and permitted BTM data center projects tracked by Cleanview. We then estimate greenhouse gas emissions based on the CO₂, CH₄, and N₂O emissions rates reported for stationary natural gas combustion by EPA. We convert these to CO₂-equivalent emissions using the global warming potential values listed in IPCC AR6, for a 100-year time horizon. This methodology ensures consistency with the state-level CO₂-equivalent emissions estimated for grid power, which is derived from the long-run marginal emissions rates shown in NLR’s Cambium dataset.