How a data center gets power: the grid, substation, interconnection

The grid connection, and why it sets the whole project

A data center's grid connection is how it gets electricity from the utility, and on a large site that connection is a substation tied to the high-voltage grid that steps the voltage down to a level the campus can distribute. Everything inside the fence, the distribution, the backup power, the cooling, depends on power arriving at the property first. That arrival is no longer a given. It is the single hardest thing to secure on a new build and the first thing to settle before any other decision is worth making.

Power availability is now the top site-selection driver for a data center, ahead of land, fiber, water, and tax. A site with cheap land and no path to power is a site you cannot build, while a site with a firm interconnection beats a better-located one that has to wait in a queue. The order has flipped from a decade ago, when power was assumed and location was the contest.

This guide stays upstream of the fence. It covers the grid, the transmission feed, the substation, the interconnection process, and the utility deal, up to the point where the campus distribution takes over. From the substation secondary on down, the path to the rack is the subject of the power distribution chain guide, and the on-site backup that carries the load when the grid drops is the subject of the generator sizing guide. Read those for inside the building. Read this for how the power got to the building at all.

How does a data center get power?

A data center gets power by tying into the utility's high-voltage grid through a substation, which steps the transmission or sub-transmission voltage down to the medium voltage the campus distributes. The path runs from a generating source across the bulk transmission grid, to a substation at or near the site, through the metering and demarcation point where the utility hands off to the owner, and into the campus medium-voltage distribution. From there the on-site transformers, switchgear, and UPS carry it the rest of the way to the racks.

The handoff is the part that matters for who owns what. The interconnection point, often called the point of common coupling, is where the utility's system meets the customer's. Upstream of it is the grid and usually the utility's responsibility. Downstream is the owner's. On a large campus the substation itself can sit on either side of that line depending on the deal, which is the on-site versus utility substation question covered below.

The reason it goes through a substation at all is the load. A normal commercial building takes service at low voltage, a few hundred volts, because its load is small enough to carry that way. A data center campus pulling tens to hundreds of megawatts cannot. At that scale the current at low voltage would be absurd, so the campus takes service at medium or high voltage and steps it down on site, which is the whole reason the substation exists at the front of a data center.

The bulk grid and the transmission feed

The grid a data center ties into is the bulk power system: the generating plants, the high-voltage transmission lines that move power across long distances, and the substations that connect them. Power is moved at high voltage on transmission because higher voltage means lower current for the same megawatts, which cuts the line losses and the conductor size over hundreds of miles. The data center connects to this system at a substation, either on the high-voltage transmission network or on the lower-voltage sub-transmission that feeds a region.

Where on the grid a site connects shapes everything that follows. A site near strong transmission with spare capacity can be energized sooner and larger than a site at the weak end of a long radial feeder. This is why the busiest data center markets cluster where the transmission is strong, and why developers now read the grid map before the real estate map. The available capacity at the nearest point of interconnection is a real, finite number, and it is often already spoken for.

The grid is also not one system but several interconnected regions, each run by an operator that manages reliability and the interconnection process. The capacity to add a large new load, and the studies that prove the grid can carry it without destabilizing, fall to that operator and the local utility. A data center is a large enough load that connecting it can require new transmission, new substation capacity, or upgrades miles from the site, all of which take time and money the project has to carry.

What voltage does a data center take service at?

A data center takes utility service at medium or high voltage because of the size of the load, commonly somewhere from the 13.8 kV and 34.5 kV medium-voltage range up to transmission levels of 115 kV, 230 kV, or higher for the largest campuses. The exact voltage is set by the utility, the size of the load, and what the local grid offers, so it varies by region and project. The principle is fixed: the bigger the load, the higher the voltage it makes sense to take it at.

A small commercial building takes 480 V or 208 V service straight off a utility transformer on a pad or pole. A data center campus cannot, because the current to deliver tens or hundreds of megawatts at those voltages would need conductor and gear that do not exist at sane cost. Take the same power at 115 kV instead of 480 V and the current drops by more than two hundred to one, which is the whole reason large loads connect high and step down on site.

The level you connect at also sets what gear you own and operate. Connect at distribution medium voltage and the utility brings it close and you take it from there. Connect at transmission voltage and you are operating high-voltage substation gear, with the protection, the clearances, and the qualified people that come with it. Confirm the service voltage early with the utility, because it drives the substation design, the demarcation, and the metering, and it is not a number the project gets to pick on its own.

What is a data center substation?

A data center substation is the installation that connects the campus to the utility grid and steps the high voltage down to the medium voltage the site distributes. It carries the high-voltage incoming gear, the step-down power transformers, the medium-voltage switchgear on the secondary, the protective relaying, and the metering. On a large campus it is a fenced yard of its own near the property line, sized and dedicated to that one load, not a shared utility substation feeding a neighborhood.

The dedicated substation is what separates a data center from an ordinary large customer. A campus pulling enough power to rival a small city often justifies, or is required to have, its own substation built for it, sometimes funded by the developer and sometimes by the utility under the interconnection agreement. The transformers in it are large power transformers, oil-immersed and rated in the tens to hundreds of MVA, a different class of gear from the dry-type unit substations inside the building.

Inside the fence the substation does three jobs at once: it steps the voltage down, it protects the grid and the site from each other through the relaying, and it meters the energy for billing. The step-down and the campus distribution that follows are the start of the power chain covered in the distribution chain guide. What is specific to the substation, and to this guide, is that it is the physical and contractual tie to the grid, the place where the utility and the owner meet and where the whole project's power either exists or does not.

Owner-built or utility-owned substation

The substation can be owned by the utility or by the data center, and the line between them, the demarcation, sets who builds it, who maintains it, who operates the high-voltage gear, and where the meter sits. There is no single right answer. It is a deal struck site by site between the utility and the owner, and it shapes the schedule, the cost, and the day-to-day operation for the life of the campus.

A utility-owned substation puts the high-voltage work, the operation, and the maintenance on the utility, with the demarcation at the substation secondary or the property line and the owner taking medium voltage from there. A customer-owned substation puts the build, the gear, and the qualified high-voltage operation on the data center, often in exchange for a faster schedule or a better rate, with the metering moved up to the primary high-voltage side. Primary metering through high-voltage potential and current transformers is common when the customer owns the step-down, because the utility measures what it delivers before the customer's transformer losses.

Get the demarcation pinned down in writing before design, because it decides scope. A surprise late in the project about who is building the substation, who owns the protection scheme, and where the revenue meter lives is a schedule and budget problem at the worst possible point. The one-line, the metering location, and the operating responsibility all flow from that single line on the agreement.

The scale: megawatts to gigawatts

A data center's load is measured in megawatts, and the AI campuses now being planned are measured in gigawatts, which is the number that makes the grid the constraint. A traditional enterprise hall might draw a few megawatts. A large cloud campus draws tens to low hundreds of megawatts. The hyperscale AI campuses announced over the last few years reach into the hundreds of megawatts and past a gigawatt on a single site, a load that rivals a city or a large industrial plant.

The jump in scale comes from the rack. AI compute pushed rack densities from the old 5 kW to 15 kW range into 50 kW, 100 kW, and beyond, and a hall full of those racks pulls what a whole traditional building used to. Multiply that across a campus and the site load lands in territory the local grid was never built to serve without expansion. The detail of how that density reshapes the distribution inside the fence is in the power distribution chain guide. The point here is what it does outside the fence.

At gigawatt scale the data center stops being a customer the grid absorbs quietly and becomes a load the grid has to plan around. Global data center electricity demand has been projected to pass 1,000 TWh, roughly double its early-2020s level, with AI driving most of the growth, though the exact figures move with each forecast. A single campus at that scale can shift a regional load forecast on its own, which is why a new project can trigger new generation and transmission, and why power, not space or capital, is what now limits how fast the buildout goes.

Why does it take years to power a data center?

It takes years to power a large data center because connecting a load that big to the grid requires an interconnection study, an agreement, and often grid upgrades, and the queue to get through that process now runs multiple years in the busiest markets. The building can go up in 12 to 24 months. The power to fill it can take far longer, and the gap between a finished shell and an energized one has become the defining schedule risk on a data center project.

The process itself is the bottleneck. A new large load files into the utility's interconnection queue, the operator studies what the connection does to the grid, what upgrades it forces, and what it costs, and only then is an interconnection agreement signed and construction of the tie and any upgrades begun. Each study takes time, the queue is long, and the upgrades, a new substation bay, a reconductored line, a new transformer with its own long lead time, stack on top. In the hottest markets the wait from queue to power has been reported at four to seven years.

The numbers behind the wait are large. Recent queue reports put well over 1,500 GW of generation and storage waiting in United States interconnection queues, more than the entire installed capacity of the country, and the load queue is filling alongside it. A campus that joins a queue in a constrained market may not realistically draw utility power for years regardless of how fast it is built or how much capital stands behind it. Secure the interconnection first. The schedule lives or dies on it, not on the construction.

Why is power the biggest constraint for data centers?

Power is the biggest constraint for data centers because the demand from AI compute is outrunning the grid's ability to deliver it, so the limit on the buildout is energized megawatts, not chips, capital, or land. Operators have found they can buy the GPUs and raise the money faster than the utility can deliver the power to run them. The binding constraint moved from the silicon to the substation, which is the defining condition of the 2026 build cycle.

Three things collide to make it the constraint. AI rack densities multiplied the load per site. The interconnection queues filled to years-long waits. And the grid itself, the generation and the transmission, was not built for a surge of gigawatt loads arriving all at once in a few markets. Where they meet, a project with money and a design sits waiting on a connection date it does not control, which is a different kind of problem than the industry had when power was assumed.

The response has reshaped how sites are chosen and built. Developers now site campuses where power is available rather than where the market would otherwise put them, chasing regions with spare grid capacity or stranded generation. They split builds into phases the grid can feed in stages. And, increasingly, they build their own generation on site to skip the queue entirely, which is the behind-the-meter trend covered below. Read the constraint plainly: for the foreseeable build cycle, the project that has power wins, and the one that does not waits.

Redundant utility feeds and diverse routing

A data center that counts on the grid for availability takes more than one utility feed, ideally from separate substations on separate routes, so a single utility fault does not take the whole site dark. Two feeds is the common arrangement, and the higher-availability designs run them as a 2N pair, two complete independent paths from the grid where either one alone can carry the site. The feeds enter through the substation and tie into the campus the same way the rest of the redundancy is built, with either path able to fail or be taken out for service.

Two feeds are worth less than they look if they share a common point upstream. Feeds that come off the same substation transformer, or share a duct bank or a right-of-way for the last stretch, both go down to a single fault even though the one-line shows two. Diverse means electrically and physically separate, back to separate sources, which is the same lesson the power distribution chain guide makes about A and B paths inside the fence. Confirm the real diversity with the utility, because a drawing can show two feeds that quietly meet at a single transformer or bus.

The utility side has its own redundancy question, asked the same way as the rest of the chain. A substation with a single incoming line and a single transformer is a single point of failure no matter how redundant the campus is downstream. Designs that need the availability run redundant transformers and redundant incoming lines so the substation can lose one and keep the site up. The Uptime Institute Tier framework that classifies the campus redundancy applies here too, because a Tier rating only holds if the redundancy is continuous from the grid in, not just from the UPS down.

The grid is not 100 percent, so the site backs itself up

No utility grid is perfectly reliable, so a data center carries its own backup power to ride through the outages the grid will have. Even a strong grid with redundant feeds has faults, storms, and the rare wide-area event, and a data center cannot tolerate the seconds of interruption a normal building shrugs off. The grid is the primary source and a good one. It is not a sole source, and a design that treats it as one is a design waiting for the outage that proves it wrong.

The backup is the UPS and the standby generators. The UPS bridges the gap on stored energy the instant the grid drops, holding the critical load for the seconds to minutes it takes the generators to start and accept it. The generators then carry the site on stored fuel until the grid returns. Sizing and selecting that generator plant, the rating, the step load, the fuel runtime, and the redundancy, is the whole subject of the generator sizing guide. This guide makes only the upstream point: the grid feed and the on-site backup are two halves of the same availability question.

How much backup, and how it is arranged, ties straight back to how good the grid feed is. A site with two diverse, reliable feeds carries a different backup posture than a site at the weak end of one feeder. The availability target, set by the Uptime Tier or the owner's requirements, has to be met across both the utility feed and the on-site backup together. Neither one alone is the answer, and the design that gets it right reads them as one system, not two.

The utility deal: capacity, demand charges, and the rate

Securing the power is a commercial deal as much as an engineering one, and the load-serving agreement with the utility sets the rate, the capacity reserved, the demand and capacity charges, and the term the data center is committed to. A campus this size does not buy power off a standard tariff. It negotiates a large-load agreement, and the terms in it run for the life of the site and shape its operating cost as much as the equipment does.

The charges are structured for the size and the risk. Beyond the energy used, a large load pays a demand charge tied to its peak draw and often a capacity charge for the grid capacity held in reserve for it. Utilities facing data center load have moved to protect their other ratepayers by writing in long minimum terms, commonly a decade or more, allowable ramp periods to fill the load, minimum monthly demand charges set at a high fraction of the contracted capacity, and exit fees for walking away from reserved capacity. The exact terms vary by utility and regulator, so confirm them against the actual agreement.

Read the deal as part of the project, not as a back-office detail. The reserved capacity is the number the whole design is built against, the ramp schedule sets how fast the load can actually arrive, and the demand and capacity charges are a running cost for the life of the site. A campus that contracts for capacity it then cannot fill on schedule is paying for power it is not using, which is why the load ramp and the buildout phasing have to line up with what the agreement actually commits to.

Medium-voltage distribution across the campus

From the substation, power moves around the campus at medium voltage before it is stepped down to utilization voltage at each building, because medium voltage carries the load across the site at far lower current and conductor than 480 V could. The substation secondary feeds a campus medium-voltage distribution, commonly somewhere in the 13.8 kV to 34.5 kV class depending on the design, which runs to each data hall or block and into the unit substations that drop it to the 480 V the building plant uses.

This is the stage where this guide hands off to the power distribution chain guide. The medium-voltage distribution, the unit substations, the switchgear, the UPS, and everything down to the rack are that guide's subject in depth. The reason it is mentioned here at all is to close the path: the substation does not feed the racks directly, it feeds a campus distribution that fans the power out to the buildings, and the choice of campus distribution voltage is a real design decision that trades current and copper against gear cost.

On a large campus the medium-voltage distribution is its own redundant system, looped or run as dual feeds so a building can be fed from either side and a section can be taken out for service. That redundancy has to line up with the redundancy of the substation feeding it and the plant inside the building, or the campus has a strong grid feed and a weak link in the middle. The redundancy is only real if it holds continuously from the grid to the rack.

The transformers that step the voltage down

The transformers are what step the voltage down at each stage from the grid to the building, and the substation carries the largest of them. The high-voltage incoming line lands on power transformers that step transmission or sub-transmission voltage down to the campus medium voltage, large oil-immersed units rated in the tens to hundreds of MVA, with their own cooling, their own protection, and lead times now measured in many months to years. After those, smaller unit-substation transformers inside the buildings drop medium voltage to the 480 V the plant runs on.

Each transformation is a deliberate, sized, protected step, and each one has a loss and a failure mode. A substation power transformer running warm has a finite life, and its failure takes out everything fed from it, which is why the availability designs run redundant transformers so either one can carry the load. The long lead time on large power transformers has itself become a schedule constraint on data center builds, on the order of the interconnection wait, so the transformer order often has to be placed long before it is needed.

The downstream transformers, the building unit substations and the floor PDU transformers, belong to the power distribution chain guide, which covers how they are grounded as separately derived systems and built to take the harmonic heat the IT load throws back. The substation transformer is the one specific to this guide: the large step-down at the grid tie that makes the high incoming voltage usable by everything below it, and a long-lead, single-failure stage worth designing redundancy around.

Revenue metering and protective relaying at the tie

At the interconnection, two systems watch the power for two different reasons: revenue metering bills the energy, and protective relaying isolates a fault before it damages the grid or the site. Both live at the substation, and both run off the same high-voltage instrument transformers, the potential and current transformers that scale the high voltage and current down to signals the meters and relays can read. Where the metering sits, primary or secondary side, follows the demarcation and who owns the step-down transformer.

The revenue meter is the cash register of the deal. Primary metering on the high-voltage side measures what the utility delivers before the customer's transformer losses, which is common when the customer owns the substation. Secondary metering measures after the step-down. The difference is the transformer loss, and which side the meter sits on is part of the negotiated agreement, not an afterthought, because it decides who pays for that loss over the life of the site.

The protective relaying is the safety system at the tie. The relays watch the instrument transformers for faults and abnormal conditions and trip the high-voltage breakers to isolate the substation from the grid before damage spreads either way. The protection scheme and its settings are coordinated with the utility's own protection, because a fault on one side must clear without dragging down the other. This is engineering and utility-coordination work, not field guesswork, and the field's job is to confirm the relays are set as the coordination study and the utility agreement require, not as the generic drawing shows.

Power factor and the utility penalty

Power factor is the ratio of the real power a site uses to the apparent power the utility has to deliver, and a poor power factor can earn a penalty on the bill because it makes the utility carry current that does no work. Many utilities bill a penalty when the power factor at the meter falls below a threshold, commonly in the 0.90 to 0.95 range, with the exact threshold and formula set by the tariff. At data center scale even a modest penalty is real money over the life of the agreement.

Modern data center load is generally well-behaved on power factor, because the UPS rectifiers and the variable-frequency drives on the cooling that used to drag it down now mostly run at near-unity input on active front-end designs. That is not a guarantee. Older gear, lightly loaded equipment, and certain operating modes can pull the measured power factor below the tariff threshold, and the meter at the tie is what the utility bills against, not the nameplate.

Where the power factor runs low, the fix is correction at the site: capacitor banks or active correction that supply the reactive demand locally so the utility does not have to carry it across the grid. Whether it is needed is a question the load study and the metered power factor answer, not an assumption. Confirm the tariff threshold and the actual metered power factor before deciding the site needs correction, because correcting a power factor that is already fine is capital spent on a problem the site does not have.

On-site and behind-the-meter generation

On-site generation, building power at the data center behind the utility meter, has become a leading way to get a campus running without waiting years in the interconnection queue. Instead of depending on the grid for primary power, the operator generates some or all of it on site, often called bring-your-own-power, and connects to the grid later or for backup only. It is the most consequential shift in how data centers are powered in this build cycle, driven straight by the queue and the grid-capacity constraint.

The technologies in play are gas turbines, reciprocating gas engines, and fuel cells for primary on-site power, with nuclear small modular reactors talked about for the longer term. Around 50 GW of behind-the-meter gas generation was reported announced in a single recent year, and one industry forecast puts on-site generation as the primary source for more than a third of data center facilities by 2030, up from a small fraction a year earlier, though forecasts like these move. Fuel cells are cited as offering a speed-to-power advantage of several months to a year over heavy turbine orders, which is the metric that matters when the queue is the problem.

Behind-the-meter is not free of trade-offs, and it is not the same as the standby generators in the generator sizing guide, which exist to back up a grid-fed site. Primary on-site generation means the data center is also running a power plant, with the fuel supply, the emissions permits, the maintenance, and the reliability that come with it, and gas pulled from a utility pipe is a utility dependency of its own. The driver is plain: where the grid cannot deliver power on a schedule the project can accept, on-site generation is how the project gets built anyway.

Renewable power and 24/7 carbon-free energy

Data centers buy renewable power mainly through power purchase agreements, long-term contracts to buy the output of a specific wind or solar project, and the leading edge of that has moved from annual matching to 24/7 carbon-free energy, matching clean supply to demand hour by hour. The PPA is the instrument: the operator contracts for a project's output, often before it is built, which both supplies clean energy and helps finance new generation onto the grid. Reported PPA prices have run roughly in the tens of dollars per MWh depending on technology and location.

Annual matching, buying enough renewable energy over a year to equal total use, is the older standard, and its weakness is that the sun and the wind do not line up with a data center's flat, around-the-clock load. The 24/7 approach, formalized in the carbon-free energy compact several large buyers have signed, tracks the load hourly and aims to meet it with carbon-free supply in every hour, which is a far harder target and the reason the largest operators are contracting for firm clean power, storage, and a wider mix.

The honest engineering point is that a data center runs a flat baseload and renewables are variable, so matching them in real time takes either storage, a diverse portfolio of projects, or firm clean generation to fill the hours the wind and sun do not cover. The carbon goal is real and the contracts are large, but the grid mix and the hourly shape are what decide whether the clean energy actually reaches the load when the load is drawing it. Read the PPA for what it firms, not just the megawatts it names.

The data center as a grid asset: demand response

A data center can give power back to the grid as well as take it, by curtailing its load or switching to on-site sources when the grid is stressed, which is the demand-response or flexibility role that has gone from theory to demonstration. The same backup generators, batteries, and on-site generation that protect the site can support the grid during a peak, and a load that can drop or shift on a utility signal is worth more to a constrained grid than one that only ever draws.

The flexibility comes in a few forms. The site can reduce or shift its draw, lean on battery storage for a stretch, or run its on-site generation to take load off the grid during the worst hours. Recent demonstrations have shown facilities curtailing a large fraction of their grid load within a minute of a utility signal and feeding firm capacity back through storage, and one widely cited analysis found that curtailing a small fraction of annual consumption during the most stressed hours could let grids absorb on the order of 100 GW of new load without major new generation. The figures are early and the programs are still forming.

For the operator this is a way to turn a constraint into an arrangement. A data center willing to be flexible during the few worst hours of the year can sometimes connect faster, on better terms, or to a grid that could not otherwise serve a firm load that size. The catch is that the flexibility has to be real and it cannot disrupt the mission-critical workload, which is the line the demonstrations are testing. Whether it fits a given site is a question for the operator, the workload, and the utility, not a blanket answer.

Energizing the service and the substation

The grid connection gets commissioned and energized in coordination with the utility, and it is a milestone with its own schedule, its own acceptance, and its own gating role on the whole project. The substation gear, the transformers, the high-voltage switchgear, the protection, and the metering, gets acceptance-tested before it is ever energized, and then the energization itself is a scheduled event run jointly with the utility, because the tie cannot be made live on the owner's say-so alone.

The testing before energizing is the part that cannot be rushed. The power transformers are tested, the protective relays are tested and their settings verified against the coordination study and the utility's requirements, the metering is verified, and the high-voltage switchgear is checked out, all to the acceptance standards the gear is held to. A relay set wrong or a meter wired wrong is found here, on a de-energized system, or it is found the hard way later. The acceptance testing standards that govern this are referenced below.

Energization is also where the project's schedule meets the utility's. The connection goes live only when the utility's own work is done, its protection is coordinated, and its operations sign off, which is why the energization date sits on the utility's calendar as much as the owner's. A campus can be built, tested, and ready and still wait on the utility to energize the tie. Treat the energization as a jointly owned milestone from the start, and build the schedule around the utility's process, not around the construction finish.

Gigawatt campuses and the grid buildout

The direction of travel is toward gigawatt-scale campuses, and the grid being built to feed them is the largest load-driven expansion the power system has seen in a generation. The AI buildout has put loads on the planning map that did not exist a few years ago, and the grid, the generation, the transmission, the substations, the long-lead transformers, is being expanded, sited, and financed to try to keep up. Whether it keeps up is the open question that runs under every project in this cycle.

The strain is concentrated. The load is arriving fast, in a few markets, in chunks the local grid was not planned for, which is why the interconnection queues filled, why power moved to the front of site selection, and why on-site generation took off. The grid was built to grow steadily with a diffuse load, and it is being asked to absorb a sudden, lumpy, enormous one. The expansion to serve it, new lines, new substations, new generation, runs on timelines measured in years, against demand measured in quarters.

For anyone working a data center project, the practical reading is unchanged from where this guide started. Power is the constraint, the interconnection is the long pole, and the schedule lives upstream of the fence. The campus that secures a firm grid connection, or builds its own power, is the campus that gets built. The one that assumes the power will be there because the building is ready is the one that waits, and the wait is now measured in years, not months.

What to document for the grid connection

The record for a grid connection is the agreement, the one-line, and the as-built condition of the tie, captured so the next person can answer what the site is entitled to, what it is committed to, and what carries the load if a feed or a transformer is lost. A connection whose terms and redundancy were clear at turnover and quietly drifted with a later change is the common way a site ends up paying for capacity it cannot use or running on a redundancy that is no longer real.

Common mistakes

• Committing to a site or starting construction before the interconnection and the power are actually secured.
• Underestimating the interconnection lead time and queue, and building a schedule the utility cannot meet.
• Taking a single utility feed to a critical load, so one grid fault or one substation event drops the whole site.
• Calling two feeds diverse when they share a substation transformer, a bus, or a duct bank back to a common point.
• Treating the substation and medium-voltage scope as a detail instead of a long-lead, single-failure stage to design around.
• Assuming the grid for primary power with no on-site backup sized to ride the outages it will have.
• Ignoring the demand and capacity charges and the contract term, so the running cost and the ramp commitment surprise the project later.
• Leaving the substation ownership, demarcation, and metering location unsettled until late, when changing them is a schedule and budget problem.

Field checklist

Standards and references

Several bodies govern different parts of the grid connection, and naming the right one for the point is the credibility. IEEE work covers substation design and the interconnection of facilities to the grid, including the standards on connecting distributed resources and on protective relaying and coordination. The National Electrical Safety Code, the NESC, governs the high-voltage utility-side installation, the overhead and underground lines, the substation clearances, and the work practices, which is a different code from the NEC that governs inside the fence.

On the grid side, the North American Electric Reliability Corporation, NERC, sets the reliability standards for the bulk power system, and the regional grid operator and the local utility set the interconnection requirements a large new load has to meet. Those requirements, not a generic standard, define the study process, the protection the utility demands at the tie, and the upgrades the connection triggers. The Uptime Institute Tier classification gives the common language for how much of the path, grid feed included, can fail or be maintained with the load up. NETA gives the acceptance and maintenance testing standards the substation gear is checked against.

The voltages, the lead times, the queue durations, and the scale in this guide vary by region, by utility, and by year, so confirm them against the specific utility, the adopted code editions, and the actual interconnection agreement before you rely on them. Cite the controlling document by topic, and let the utility requirements and the project basis of design override any rule of thumb. None of these replace the engineer of record, the utility's own studies, or the manufacturer's requirements for the installed gear.

Units, terms, and abbreviations

The grid connection uses a stack of terms that get used loosely, and the same word can mean different things at the substation and inside the building. Pin the term to the stage before you act on it. Voltage levels run from low voltage at 1000 V and below, through medium voltage at roughly 1 kV to 35 kV, to high and extra-high transmission voltages above that, and power at the campus scale is measured in megawatts and now gigawatts.

Substation

The installation that ties the site to the grid and steps high voltage down to medium voltage

Interconnection / point of common coupling

Where the utility's system meets the customer's, the demarcation between grid and site

Interconnection queue

The line of projects waiting on the utility study and agreement to connect, now years long in busy markets

MV / HV

Medium voltage, roughly 1 to 35 kV, and high voltage, the transmission levels above it

MW / GW

Megawatt and gigawatt, the scale of a data center campus load

Behind-the-meter (BTM)

On-site generation behind the utility meter, used to power a site without the grid queue

PPA

Power purchase agreement, a long-term contract to buy the output of a generation project

Demand charge

A utility charge tied to the site's peak power draw, separate from the energy used

Demand response

Curtailing or shifting load on a utility signal so the site supports the grid at a peak

Power factor

Real power divided by apparent power; a low value at the meter can earn a utility penalty

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• How does power get from the utility to a server?
• How do you size a data center generator?
• Why are data centers building their own power?
• What is selective coordination?
• What megger voltage do I use on busway?