BlogSite-owner offtake

Data Center Substation Capacity: What AI Buyers Need

Before a buyer can size compute for your site, they need to know what your substation can actually deliver. Here is how to describe it clearly and what to have on file.

Picture a site owner with an old industrial parcel and a substation at the fence line. An AI infrastructure buyer calls and asks a simple question: how much power can you deliver, and how firm is it? The owner reads the transformer nameplate off a photo from years ago. The buyer goes quiet, then asks for the utility capacity letter, the single-line diagram and the protection study status. None of it is in one place. The conversation stalls for weeks while the owner chases paperwork that should have been ready from the start.

This playbook is for owners of land, buildings or grid connections who want to host AI compute and need to describe their data center substation in terms a buyer can verify. You do not need to be an electrical engineer. You do need to understand what a buyer is checking, why nameplate figures rarely settle the question, and which documents turn a promising site into a credible one.

1. What a data center substation actually does

A substation is the bridge between the utility’s transmission or distribution network and the equipment on your site. For an AI data center, it has a few core jobs, and a buyer will ask about each one.

  • Step-down transformation: power transformers reduce utility voltage to a level the site’s distribution equipment can use.
  • Switchgear: breakers and switches route power, isolate faults and let crews take sections out of service for maintenance.
  • Protection: relays detect faults and trip the right breaker fast enough to protect equipment without blacking out more of the site than necessary.
  • Metering: revenue meters record what you draw, and they often mark the boundary between what the utility owns and what you own.
  • Controls and communications: the systems that let the utility, the site operator or both monitor and operate the equipment.

When you describe your substation, walk through these functions in order. A buyer reading your intake should be able to picture the path power takes from the utility line to the pad where compute would sit.

2. Nameplate, firm capacity and what reaches the connection point

This is where most site descriptions go wrong. There are three different numbers that owners tend to blur together, and buyers treat them very differently.

  • Nameplate rating: what the transformer or switchgear is built to handle under defined conditions. It tells you the ceiling of the equipment and says little about what the grid will supply.
  • Firm capacity: what the substation can deliver with its largest single component out of service. This is the figure a buyer cares about for a load that runs around the clock.
  • Deliverable capacity at the connection point: what the utility has agreed, in writing, to supply to your site at a defined point, on a defined schedule, under defined conditions. This is the number that actually sizes a deployment.

A large transformer on a weak feeder is a weak site. A modest transformer with a firm utility commitment can be a strong one. When you write your intake, state which of these three you are quoting and where it came from. If you only know the nameplate, say so plainly and flag that utility confirmation is still needed. Buyers trust owners who separate verified facts from assumptions far more than owners who round up.

For a broader view of how AI loads differ from conventional ones, see our guide to AI data center power requirements for site owners.

3. Redundancy: how buyers think about N, N+1 and dual feeds

AI compute runs continuously, and an unplanned outage can interrupt long training jobs or customer workloads. Buyers will ask how your substation behaves when something fails or needs maintenance.

  • N: you have exactly enough equipment to serve the load. If one transformer or breaker fails, some or all of the load goes down.
  • N+1: you have one more component than the load needs, so a single failure or a planned maintenance outage does not interrupt service.
  • Dual feeds: the site is served by two independent utility lines, ideally from different sources or substations, so the loss of one line does not take the site offline.

Describe your configuration honestly. A site with a single feed and a single transformer can still be useful, especially for workloads that tolerate interruption or as a first phase, but the buyer needs to know so they can plan backup, phase the build or price the risk. Note whether any redundancy is shared with other customers, because a backup path that also serves a neighboring plant is weaker than one dedicated to your site.

4. Transformer and switchgear lead times are schedule risk

Large power transformers and medium-voltage switchgear are built to order, and lead times can be long. If your substation needs a new transformer, an upgrade or an additional breaker lineup, the procurement queue can set the critical path for the entire project.

Buyers want to know three things here:

  • Whether any major equipment must be purchased or replaced before the site can carry the planned load.
  • Who is responsible for ordering it: the utility, you or the buyer.
  • Whether an order has been placed, a slot reserved or a quote received, and what the supplier has said about delivery.

Do not guess at delivery timing. Share what the supplier or utility has told you in writing and label anything else as an estimate. Our post on how to disclose transformer lead time to a GPU offtaker covers the wording in more detail.

5. Protection and coordination studies

Before a utility energizes a new or expanded load, engineers typically run protection and coordination studies. These confirm that relays and breakers will clear faults correctly, that a fault on your site will not trip the utility’s upstream equipment, and that settings on both sides work together.

AI loads raise specific questions. Large power electronics, fast load swings and dense rack power can affect how protection behaves. A study may require new relay settings, additional equipment or changes on the utility side, any of which can add cost and time.

  • State whether a study has been requested, is in progress or is complete.
  • Note who is performing it and who is paying for it.
  • Share any findings or required upgrades already identified.
  • Flag if the study was based on a different load profile than the one now proposed.

6. Who owns and controls the substation

Ownership shapes almost everything else: who maintains equipment, who can switch it, who pays for upgrades and how quickly changes can happen. Substations generally fall into a few patterns.

  • Utility-owned: the utility owns, operates and maintains the substation. You get reliability and professional maintenance, but upgrades follow the utility’s process and schedule.
  • Customer-owned: you own the substation from the utility’s metering point onward. You have more control and flexibility, along with responsibility for maintenance, staffing, spares and compliance.
  • Shared or mixed: the utility owns the high-voltage side and you own the rest, or the substation also serves other customers on your property or nearby.

Be specific about the ownership boundary and who has switching authority. A buyer planning continuous operations needs to know who they call when something trips at night. For a fuller treatment, read how to disclose substation ownership and control.

7. Space for a future substation bay

Buyers rarely plan for a single phase. If the first deployment works, they will want to grow, and growth usually means more transformer capacity, another feeder or a new substation bay.

  • Mark open ground near the existing substation that could hold an additional bay or transformer.
  • Note setbacks, easements, wetlands or zoning limits that could block expansion.
  • Identify the route for a second utility line, if one is possible.
  • Confirm access for heavy haul equipment, since transformers arrive on specialized trailers that need firm roads and turning room.

Expansion room is a real asset. A site with modest firm capacity today and a clear path to more can be more attractive than a larger site that is boxed in.

8. Documents to prepare and how buyers map power to modules

Assemble these before your first serious buyer conversation. Each one answers a different question, and none substitutes for another.

  • Single-line diagram: shows the path from the utility connection through transformers, switchgear and protection to the load. Mark it current or as-built, and note its date.
  • Utility capacity letter: the utility’s written statement of what it will deliver, where, when and under what conditions. This carries more weight than any internal estimate.
  • Study status: interconnection, protection and coordination studies, with dates, responsible parties and findings.
  • Equipment ratings: nameplate data for transformers, breakers and switchgear, plus age, condition and maintenance records.
  • Ownership and operating documents: boundary drawings, switching agreements and maintenance contracts.
  • Site plan: pad locations, access routes and expansion area.

Once a buyer has these, the work turns to fit. Pacific treats verified power as the starting point. As described on Pacific’s AI factory sites and powered land page, a viable site begins with verified power, a workable placement plan and clear operating interfaces. Pacific confirms the available feed, connection point, timing and delivery obligations, then map them to a planned compute configuration. Pacific’s module reference includes direct-to-chip liquid cooling, networking, storage and bare-metal compute, and modules land on a prepared pad followed by site commissioning and integration.

To check whether the parcel can hold the planned equipment, see our site owner's land requirements guide.

The compute side of that mapping is concrete. Different GPU platforms draw and reject heat very differently, so deliverable capacity determines which configuration and how many nodes a site can support. If you want a sense of what sits at the end of that chain, Pacific’s GPU compute catalog of dedicated bare-metal nodes lists the GPU types Pacific offers, from H100 SXM nodes to GB200 NVL72 and GB300 NVL72 configurations by request. Pacific evaluates every site case by case, and deployment time is always a site-specific target.

Pacific works with owners of stalled or unused power to turn verified capacity into deployed AI infrastructure. Pacific Intelligent Technologies, Inc. reviews site intakes and will tell you plainly what is verified, what still needs utility or engineering confirmation and what fit looks like. Book a site review call with Pacific to walk through your substation and power documents.

FAQ

Is my transformer nameplate enough to describe my data center substation?

No. Nameplate tells a buyer what the equipment can handle, but it says nothing about how much the utility will deliver or whether the site stays up when a component fails. Share nameplate data alongside a utility capacity letter and a clear note on firm capacity, and label anything unconfirmed as an assumption.

Do I need redundant feeds to host AI compute?

Not always. Redundancy affects which workloads fit and how a buyer plans backup and phasing. A single-feed site can still work for some deployments or as a first phase. What matters most is describing your configuration accurately so the buyer can plan around it.

What if my protection study has not started yet?

Say so in your intake. An unstarted study is a known gap a buyer can plan for. Note whether you have requested it, who would perform it and whether the utility has indicated the load profile it needs. Hiding the status only delays the conversation once diligence begins.

Continue on the mothership

This satellite stops at the playbook. Transactions, specs, and comparisons live on pacificmachines.com. If the next step is a human, book 30 minutes with Harper.

Book 30 min