For a long time, data center teams treated utility interconnection like a slow administrative process. It was important, certainly, but it sat somewhere upstream of the “real” engineering work. The project team would submit the load request, wait for studies, coordinate with the utility, and eventually the service design would settle into place.
That mindset is outdated.
For AI data centers, utility interconnection is no longer paperwork. It is one of the main design constraints on the project. In some cases, it is the constraint that decides whether the project is viable at all.
The reason is simple: AI has turned data centers into unusually large, unusually fast-moving electrical loads. The U.S. Department of Energy reported that data centers consumed about 4.4% of total U.S. electricity in 2023 and are projected to consume approximately 6.7% to 12% by 2028. Total U.S. data center electricity use increased from 58 TWh in 2014 to 176 TWh in 2023, with projected demand of 325 to 580 TWh by 2028. That is not a minor planning adjustment. That is a national infrastructure shift.
The impact shows up locally first. A single AI campus can request tens or hundreds of megawatts. In some regions, those requests arrive faster than utilities can add transmission, transformers, substations, generation capacity, and planning staff. PJM’s 2026 load forecast adjusted multiple zones specifically for growth in data center load and projects its summer peak load to increase by 65,733 MW over the next ten years.
That creates a new reality for owners, engineers, EPCs, and manufacturers: the grid is not a fixed input anymore.
The grid changes underneath the project.
The project may not change, but the electrical boundary conditions do
The cleanest way to understand the problem is to imagine a 120 MW AI deployment in early design.
The owner has secured a site. The preliminary electrical model is built around utility data available at that time. The design team specifies service entrance equipment and downstream UL891 switchboards based on assumed available fault current. Procurement starts moving because long-lead electrical equipment cannot wait forever. The project schedule depends on releasing gear early.
Then the utility finalizes its interconnection plan.
To serve the load, the utility may decide it needs a larger transformer, a reinforced feeder, a different substation configuration, or an upgraded service path. None of those changes mean the facility’s load requirement changed. The data center still needs 120 MW.
But the utility’s method of serving that load changed.
That is enough to alter the electrical design.
Moonshot’s existing interconnection article explains the cascade clearly: when the utility’s final design increases available short circuit current at the service entrance, breaker interrupting ratings must be revalidated, bus bracing requirements can increase, arc flash calculations can shift, PPE classifications may change, and service entrance equipment may need to be upgraded. What looked compliant during schematic design can require reevaluation once final utility data arrives.
This is where a lot of teams get burned. Early modeling is not useless, but it is often incomplete. It is a placeholder based on the best data available at the time. The problem starts when the project treats that placeholder as if it were final.
A service entrance that looks straightforward at 35 kA behaves very differently at 65 kA. That difference is not academic. It can change breaker selection, switchboard rating, bus bracing, arc flash incident energy, selective coordination, and inspection posture.
And if that information arrives after equipment has been released, the cost of change goes up fast.
Why utilities are forcing the issue
Utilities are not slowing projects down just because they are bureaucratic. Many are trying to absorb an entirely new class of load.
Traditional enterprise data centers were large, but they usually fit inside known planning patterns. AI facilities are different. They often have higher sustained utilization, faster load ramp-up, less load diversity, and much tighter uptime requirements. Moonshot’s interconnection post makes the important point that AI facilities compress significant demand into a single site, which pushes utilities to strengthen upstream infrastructure. Stronger infrastructure often means higher available fault current at the secondary side of the transformer.
NERC is now studying this issue at the reliability level. In its March 2026 white paper on emerging large loads, NERC concluded that existing reliability standards, interconnection requirements, and planning processes are inadequate to address the risks posed by emerging large loads such as data centers. NERC specifically identified gaps around interconnection processes, planning, resource adequacy, disturbance ride-through, stability, power quality, security, resilience, and load modeling.
That should get every project team’s attention.
This is not just a utility queue issue. It is a standards, planning, and reliability issue. NERC’s paper notes that large-load interconnection processes need thorough system study, analysis, review, and approval because the impacts of emerging large loads can extend beyond the immediate system to which they are connected.
That means utilities and grid operators are going to keep asking harder questions. They will want better visibility into load behavior, ramp rates, power quality, ride-through capabilities, curtailment flexibility, and source configuration. Data center developers should expect more scrutiny, not less.
For the owner, this changes the schedule logic. A project is not “ready” just because the building design is ready. It is not “ready” just because the electrical rooms are laid out. It is ready when the utility boundary conditions, service entrance strategy, fault studies, protection scheme, and equipment ratings are aligned.
That alignment is becoming the critical path.
The redesign problem nobody wants to own
The worst version of this problem happens after procurement.
The one-line is issued. The gear is released. The switchboards are in fabrication. Then the utility’s final study changes the available fault current or service configuration. Now the design team has to ask an ugly question: does the equipment still match the real condition?
Sometimes the answer is yes. Sometimes the existing design had enough margin.
Other times, the answer is no.
That is where a “utility update” becomes a project event.
A redesigned service entrance can trigger changes to main breakers, feeder breakers, bus bracing, switchboard short circuit current rating, equipment layout, arc flash labels, coordination studies, cable sizing, and installation drawings. If the equipment is already fabricated, the issue may require rework or replacement. If the equipment is installed, the project may face field modification, inspection delays, or field evaluation. If the facility is live, the problem becomes even more painful because every change now touches outage planning and operational risk.
This is the part that gets underestimated. Electrical redesign after procurement is not just an engineering cost. It can affect:
- Long-lead equipment availability
- Submittal approvals
- Factory schedules
- Installation sequencing
- Inspection timing
- Commissioning windows
- Energization milestones
- Revenue tied to compute availability
In AI infrastructure, delayed power is not just a construction delay. It is delayed compute revenue.
That is why interconnection uncertainty is so dangerous. It turns what should be a front-end design variable into a late-stage schedule threat.
Morgan Lewis described the broader market issue well in a 2026 analysis of data center energy strategy. The firm noted that interconnection processes are emerging as a key bottleneck, with projects facing grid congestion, delays in interconnection queues, significant system upgrades, and regulatory approvals across federal regulators, regional grid operators, state utility commissions, and local authorities. Each layer can affect timing and feasibility.
That is exactly why the old “procure early and adjust later” approach has become risky. On a normal project, early procurement protects the schedule. On a high-density AI project, early procurement without enough utility certainty can lock in the wrong assumptions.
A familiar failure pattern
The failure pattern is usually not dramatic at first. It looks ordinary.
A developer is under pressure to secure equipment because lead times are brutal. The design team has preliminary utility data. The manufacturer is asked to quote the switchboard package. Everyone knows the utility study is not technically final, but the schedule cannot wait. The team adds what feels like reasonable margin and releases the gear.
Months later, the final utility plan arrives. The transformer is larger than expected. Available fault current is higher. The service entrance configuration changes. The switchboard rating that looked acceptable during schematic design is now too close to the line, or no longer adequate. Coordination studies need to be revised. Arc flash labels need to be updated. The AHJ starts asking questions because the installed equipment has to match the actual available fault current, not the preliminary design story.
No one was reckless. No one ignored the problem. Everyone was trying to keep the project moving.
But the project treated utility data as stable before it actually was.
That is the mistake.
The grid is no longer a background condition. It is a moving design input.
The switchboard is where the grid meets the building
This is where Moonshot’s role becomes important.
A switchboard manufacturer can be treated as a downstream fabricator. In that version of the project, the manufacturer receives a mostly finished design, prices the package, produces submittals, and builds what was specified. That approach works when the design assumptions are stable and the electrical environment is conventional.
AI data centers are not conventional.
The switchboard sits at the point where utility reality becomes facility reality. If the available fault current changes, the switchboard has to answer for it. If transformer sizing changes, the switchboard feels it. If expansion planning changes, the switchboard either supports it or constrains it. If breaker substitutions are made under lead-time pressure, the switchboard becomes the place where procurement decisions collide with coordination studies.
That means Moonshot should not be positioned as a company that simply builds switchboards after the engineering is done.
Moonshot should be positioned as an upstream engineering partner that helps customers protect the design before fabrication locks in the wrong assumptions.
Moonshot’s existing interconnection article already makes this point: bringing manufacturing partners in early can help align short circuit ratings before fabrication, select realistic bus bracing, validate breaker interrupting margin, build coordination studies on confirmed fault levels, and design expansion headroom intentionally.
That is the commercial argument.
Early involvement is not just better engineering. It is schedule insurance.
What early involvement actually changes
Early involvement does not mean every utility variable becomes perfectly known. It means the project team designs around uncertainty instead of pretending uncertainty is gone.
A better process asks sharper questions before the switchboard package is released:
- What utility data is confirmed, and what is still preliminary?
- What available fault current range should the equipment be designed to withstand?
- What transformer sizes or feeder configurations are still possible?
- What happens if the utility increases service capacity to support future phases?
- Are downstream UL891 switchboards rated for the installed condition or just the early model?
- Does the breaker strategy preserve coordination if the available fault current shifts?
- Is expansion headroom being designed intentionally, or merely assumed?
Those questions can feel tedious during early design, but they prevent much more painful questions later.
The important part is not to overbuild blindly. Oversizing everything can create its own problems, including higher cost, larger gear, more space, and higher fault energy. Moonshot’s interconnection article points out the tension clearly: oversizing for growth can increase fault exposure, while undersizing constrains expansion.
That balance is the engineering challenge.
The right answer is not always “bigger.” The right answer is better alignment between utility studies, NEC requirements, UL assembly ratings, fault current exposure, coordination strategy, arc flash analysis, and long-term expansion goals.
That is not a procurement exercise. That is system design.
What this means for owners and EPCs
Owners and EPCs need to stop treating utility coordination as a parallel workstream that eventually feeds the electrical team. It needs to be integrated into the design and procurement sequence from the beginning.
The practical shift is simple: do not let preliminary utility assumptions quietly harden into final equipment decisions.
The one-line should evolve with utility feedback. Fault studies should be updated as the utility design matures. Switchboard ratings should be checked against realistic future conditions, not just the easiest current case. Procurement should understand which substitutions are harmless and which ones trigger coordination or listing issues. The AHJ should not be discovering the mismatch at the end.
In a high-density AI facility, the costliest electrical issue is often the one that was visible early but not owned by anyone. Utility had one version of the future. Engineering had another. Procurement had a deadline. Manufacturing had a release date. Construction had a sequence. The switchboard became the point where all those assumptions met.
That is why the manufacturer needs a seat at the table earlier.
Not because manufacturers should run the project. Because the people building the equipment understand the consequences of changes that look harmless on paper.
The real takeaway
Utility interconnection is no longer just the first step in powering a data center. It is one of the defining constraints of AI infrastructure.
The utility determines more than when power arrives. It influences available fault current, transformer strategy, service entrance configuration, breaker ratings, bus bracing, arc flash exposure, coordination studies, expansion flexibility, inspection readiness, and energization risk.
Projects that treat those variables as static are going to get surprised.
Projects that align utility studies, electrical engineering, and switchboard manufacturing early will have more control.
That is where Moonshot fits.
Moonshot is not just the company that fabricates the gear after the drawings are done. Moonshot can help customers think through the practical realities of fault current, UL891 ratings, breaker strategy, bus bracing, fabrication timing, and future expansion before the project becomes expensive to change.
That is the difference between a downstream vendor and an upstream engineering partner.
If you are evaluating a large-load interconnection, reviewing transformer strategy, or trying to determine whether your switchboard package can survive final utility conditions, this is the conversation to have early.
https://moonshotus.com/request-form/
Because the grid sets the boundary conditions.
Your design determines whether the project can live inside them.
Sources:
https://moonshotus.com/the-real-electrical-impact/
https://www.nerc.com/globalassets/our-work/guidelines/reliability/white-paper—assessment-of-gaps.pdf
https://www.pjm.com/-/media/DotCom/library/reports-notices/load-forecast/2026-load-report.pdf

