On-Site Generation and Bridge Power for Data Centers

Almost every data center has some on-site generation. The question in site selection is which role it plays, because that decides how much fuel, land and permitting the site has to support. A backup generator yard and a permanent power plant are very different neighbors.

The roles can overlap. Bridge equipment is sometimes kept as backup or peaking capacity once the grid arrives, and some campuses are planned from the start to combine grid service with a permanent on-site plant.

Backup generation is sized to carry the critical IT load and the cooling that supports it, usually in an N+1 or 2N arrangement depending on the reliability target. Diesel remains the common fuel because it can be stored on site and the units start and accept load quickly. Gas-fired backup avoids on-site fuel storage but depends on the pipeline staying up during the same event that took down the grid.

At typical unit sizes of a few megawatts each, a large campus needs a very large number of generators. The generator yard becomes a major piece of the site plan, and it drives noise studies, setbacks and air permitting. Emergency units are generally permitted with limits on non-emergency run hours, and the details vary by state air agency and by the emissions tier of the equipment.

For site selection, backup generation is rarely a deal-breaker, but it is not trivial. Check the air quality status of the area, the local noise ordinance and the distance to the nearest homes. Our guide to noise, setbacks and buffers covers how generator yards shape a layout.

Bridge power exists because utility timelines and data center timelines often do not match. A building can be finished well before the transmission upgrades or new substation it needs are energized. Bridge generation lets the project start operating, often at a partial load, while it waits. Our guide to power timelines and interconnection queues explains why that gap appears.

A workable bridge plan answers several questions up front:

• Fuel: is there firm gas supply at the site, or only interruptible service or trucked fuel that suits small loads?
• Reliability: running islanded means the plant itself has to meet the uptime target, which usually means spare units beyond N.
• Load behavior: AI training loads can swing quickly, and engines, turbines and batteries respond differently to fast changes.
• Equipment: who owns it, whether it is leased or bought, and how long the long-lead items take to arrive.
• Permits: an air permit for continuous operation is a different and larger permit than one for emergency backup.
• Exit: whether the units are removed, converted to backup or kept for peaking once grid service begins.

Some campuses plan to generate most or all of their own power for the long term. The plant can run islanded, with no grid connection, or in parallel with the grid, which adds reliability but requires an interconnection and standby service arrangements with the utility. Another approach places the campus next to an existing or new power plant, which brings its own regulatory questions about how the plant’s output, the grid and the data center relate.

The legal framework matters as much as the engineering. States differ on whether a customer can generate its own power on site, whether a third party can own the plant and sell power to the data center, and what rights the local utility holds in its service territory. Wholesale market rules from FERC and the regional ISO or RTO can also apply when the plant connects to the grid. None of this has a single national answer, so it has to be checked state by state and utility by utility.

Plant type affects land and water. Combined-cycle plants are more fuel-efficient than simple-cycle turbines but need more space and, unless they use dry cooling, a meaningful water supply. Our guide to data center water requirements covers how cooling choices drive water needs.

For bridge and prime power, gas supply is what separates a real option from a talking point. A pipeline crossing the county is not the same as available firm capacity delivered at the pressure the equipment needs. The site needs a feasible tap, a lateral and a meter station, plus a contract path with the interstate pipeline or the local gas utility. Natural gas pipeline access covers those steps in detail.

To size the gas need, multiply the generating load by the equipment’s heat rate (the fuel energy needed per kilowatt-hour) and the hours of operation. Heat rate varies by technology and by load, so use manufacturer data for the specific units. Even a mid-sized continuous load works out to a gas demand on the scale of a large industrial customer, which is why many local distribution systems cannot serve it without upgrades.

Trucked compressed or liquefied gas can support small loads or short gaps, but it is rarely practical as the primary fuel for a large campus over a long period.

Under the Clean Air Act, a generating plant is permitted based on its potential emissions. Smaller facilities may qualify for a minor source permit. Larger ones can trigger major source review, known as Prevention of Significant Deterioration in areas that meet air quality standards and Nonattainment New Source Review in areas that do not, along with a Title V operating permit. Nonattainment areas are generally harder, because new emissions may need to be offset.

Permitting usually involves dispersion modeling of pollutants such as nitrogen oxides, which ties stack height, plant layout and distance to property lines together. Emission controls such as selective catalytic reduction can reduce emissions, at added cost and complexity. Stack height can also bring FAA airspace review into play near airports.

Timelines and requirements vary by state and by the size of the plant. Start early, and treat the air permit as a critical path item rather than paperwork.

When we screen a site for bridge or prime power, we look past the grid questions to a second set of checks:

• Gas: distance to an interstate or intrastate transmission pipeline, available firm capacity, delivery pressure and a feasible lateral route.
• Air quality: the attainment status of the area and any state rules that limit new combustion sources.
• Land: room for the plant, fuel handling, switchyard and buffers on top of the data center program.
• Neighbors: distance to homes, schools and other sensitive uses, and the local noise standard.
• Water: supply and discharge capacity if the plant uses evaporative cooling or a steam cycle.
• Rules: whether the state and utility allow the ownership and operating structure the project intends.
• Grid: whether a parallel connection is available, so the plant is not the only source forever.