For years, data center development followed a familiar sequence: find the site, secure utility service, design the facility, procure equipment, build the campus, energize the load. Power was critical, but it was usually treated as something the utility would eventually deliver if the project team started early enough and paid for the required upgrades.
That model is starting to break.
The constraint in AI infrastructure is no longer just compute availability. It is not only GPU supply, cooling design, or construction labor. Increasingly, the hard stop is whether the site can get enough power, fast enough, with terms the owner can live with.
The U.S. Department of Energy reported that data centers consumed about 4.4% of total U.S. electricity in 2023 and are expected to consume roughly 6.7% to 12% by 2028. DOE also reported that total data center electricity use climbed from 58 TWh in 2014 to 176 TWh in 2023, with estimates of 325 to 580 TWh by 2028. That is not normal load growth. That is a structural shift in the American power system.
Data centers used to be large customers on the grid. Now, the largest AI facilities are becoming grid events.
That is why on-site power is moving from backup strategy to business strategy. Microgrids, behind-the-meter generation, combined heat and power, battery storage, and grid-parallel systems are no longer exotic concepts reserved for campuses, hospitals, and military bases. They are becoming part of the AI data center playbook because waiting on the grid is becoming a competitive liability.
The utility interconnection process has become a project risk
The old assumption was that interconnection was slow, but ultimately procedural. Submit the application, coordinate the studies, wait for upgrades, pay the bill, and the project eventually gets service.
That assumption is becoming dangerous in high-growth data center markets.
In Northern Virginia, one of the most important data center regions in the world, Unison Energy reported that Dominion Energy’s interconnection backlog has grown beyond 40 GW of proposed projects. The same analysis cites reports of waits up to seven years for new applicants. That is not a schedule delay. That is an existential problem for any project whose economic model depends on bringing AI capacity online quickly.
Texas is moving in a different but equally important direction. Unison’s policy analysis notes that Texas Senate Bill 6, enacted in June 2025, applies to facilities consuming 75 MW or more and requires mandatory curtailment capability during emergencies, along with payment of full interconnection and transmission study costs. The law also requires remote disconnection equipment as a condition of grid interconnection, which is why some observers have called it a “kill switch” requirement.
The message from regulators and utilities is getting clearer: large-load customers cannot assume the grid will absorb them quietly.
That changes the data center development equation. A 100 MW or 300 MW AI campus is no longer just asking for power. It may be triggering transmission upgrades, substation investment, new rate structures, curtailment obligations, cost-sharing debates, local political scrutiny, and utility planning concerns. In some markets, the facility may technically be developable but economically delayed because power will not arrive on the business timeline.
This is why power strategy is now part of site selection, capital planning, and speed-to-market analysis. A site with fiber, land, tax incentives, and water can still be a bad site if the power path is uncertain.
On-site power is not just backup anymore
Historically, data centers thought about on-site generation mainly through the lens of emergency backup. The utility supplied the facility under normal conditions. Generators stood ready for outages. UPS systems bridged the gap. Fuel logistics supported resilience planning. That model still matters, but it is no longer enough for the largest AI loads.
The new question is not only, “How do we survive an outage?”
The new question is, “How do we secure power capacity when the grid cannot deliver on our timeline?”
That is where microgrids and behind-the-meter systems enter the conversation. A microgrid is not simply a generator sitting behind a building. It is a controlled local power system that can integrate generation, storage, and load within a defined electrical boundary. It can operate in coordination with the utility grid, and in some designs, it can disconnect and run independently in island mode.
Unison describes two broad configurations that are especially relevant to data centers. In an islanded configuration, the microgrid can operate independently from the utility, which helps protect against curtailment requirements, interconnection delays, and outages. In a grid-parallel configuration, the facility can operate in coordination with the utility while retaining the ability to island during emergency conditions. In some cases, grid-parallel operation can also allow excess capacity to be exported back to the grid when permitted.
That distinction matters. Islanding is about survival and independence. Grid-parallel operation is about flexibility and economics. The best strategy depends on the market, the utility, the load profile, the generation mix, and the owner’s tolerance for complexity.
CoreSite’s 2026 outlook captures the broader market shift well. It notes that data centers are moving from passive energy consumers to grid stakeholders, co-investing in infrastructure upgrades, enabling load flexibility, and deploying on-site power generation and storage to improve reliability and manage costs.
That phrase, “grid stakeholder,” is the real tell. The data center is no longer just the end of the power chain. In many cases, it is becoming part of the power system.
CHP makes the conversation even more interesting
Combined heat and power, or CHP, deserves more attention than it usually gets in data center conversations.
A CHP system generates electricity on site while capturing useful thermal energy that would otherwise be wasted. In many industrial environments, that recovered heat is used for process loads, domestic hot water, or building heat. In high-density data centers, the opportunity is different but still meaningful: waste heat can support cooling loads or improve the overall energy value of the site when engineered correctly.
Unison notes that when CHP is paired with microgrids, it can provide continuous on-site electricity while capturing waste heat to support cooling loads in high-density data centers. The result is a fuel-flexible system that can support deployment timelines, reduce operational risk, and provide islanding capability during curtailment.
That does not mean CHP is the answer for every AI campus. Natural gas availability, emissions permitting, local policy, heat recovery use cases, sustainability targets, and tenant requirements all matter. CoreSite’s 2026 outlook also notes that natural gas may serve as a bridge in alleviating grid constraints, while some large data center tenants remain cautious because natural gas solutions may conflict with sustainability expectations.
That tension is important. The industry does not have one clean answer. Some projects will use gas as a bridge. Some will pair generation with batteries and renewables. Some will pursue nuclear power contracts. Some will wait for grid upgrades. Some will build hybrid systems that can evolve over time.
The common thread is control.
Owners want more control over capacity, cost, uptime, and timing. On-site power gives them options the traditional utility-only model often cannot.
The electrical design implications are bigger than most teams realize
The mistake is thinking of on-site power as something that gets added next to the building after the main electrical design is already set. That is how projects create coordination problems.
A microgrid changes the entire electrical conversation. It affects source sequencing, fault current, protection coordination, grounding, transfer schemes, metering, controls, power quality, arc flash calculations, maintenance strategy, and future expansion. The switchboard is no longer only receiving utility power and distributing it downstream. It may need to coordinate between multiple sources, multiple operating modes, and multiple fault scenarios.
A facility that can run grid-connected and islanded is electrically more complicated than a facility that only runs from one normal utility source and emergency standby generation. When the system changes operating modes, fault current can change. Available short circuit current may be different on utility power than it is on generator power. Protective devices that coordinate in one mode may not behave the same way in another. Ground fault behavior can shift. Control logic becomes more important. Documentation becomes more important.
That is where switchboard design becomes strategic.
For a grid-parallel or islandable facility, the design team has to think through questions that do not always come up in conventional low-voltage distribution:
- What sources can feed the switchboard under each operating mode?
- What is the available fault current from the utility, from on-site generation, and from combined sources?
- How does selective coordination change when the facility islands?
- What control, metering, and monitoring must be integrated into the assembly?
- How will the system safely transition between utility-connected, islanded, and restored operation?
- Can the electrical architecture expand if more generation, storage, or load is added later?
These are not minor details. They determine whether the system is resilient in practice or only resilient on a one-line diagram.
Moonshot has already written about the way utility interconnection changes electrical boundary conditions. In a high-density AI facility, a larger transformer or reinforced feeder can increase available fault current, forcing breaker interrupting ratings, bus bracing, arc flash calculations, and service entrance equipment to be revalidated. Moonshot’s interconnection analysis makes the core point clearly: utility interconnection is no longer a procedural step, it is a design constraint.
On-site power adds another layer to that same reality. The grid sets one set of boundary conditions. The microgrid creates another. A serious design has to account for both.
Why Moonshot belongs in this conversation early
This is exactly where Moonshot’s role should expand beyond “switchboard manufacturer.”
Moonshot’s UL891 switchboard page emphasizes that its engineers design solutions around supply chain constraints and maximum flexibility, and that the company evaluates site design as part of its sales process to provide code-compliant, cost-effective, quick solutions. Moonshot also highlights the ability to use major breaker manufacturers, which matters in a market where lead times and substitution pressure can create real design risk.
That capability becomes more valuable as on-site power architectures get more complex.
A microgrid-ready distribution system is not just a bigger switchboard. It is a system that must be designed around real operating modes. Utility-connected. Islanded. Generator-supported. Storage-supported. CHP-supported. Possibly export-capable. Possibly curtailed. Possibly expanded in phases as the AI load grows.
If Moonshot is brought in late, the conversation is limited to fabrication. The project team has already made assumptions about source configuration, fault current, breaker strategy, control integration, and future capacity. At that point, the switchboard is forced to fit decisions that may not have been made with manufacturing reality, listing integrity, or long-term flexibility in mind.
If Moonshot is brought in early, the conversation changes.
Moonshot can help align service entrance strategy with distribution architecture. It can help validate how different power sources affect short circuit ratings and breaker selection. It can help design UL891 assemblies around realistic expansion paths. It can help prevent a project from buying “available” equipment that later becomes a constraint when the owner wants islanding, additional generation, battery storage, or different utility terms.
That is how Moonshot becomes an enabler of energy autonomy. Not by making broad claims about microgrids, but by engineering the electrical infrastructure that makes resilient power usable, controllable, and inspectable.
The real takeaway
The rise of on-site power is not a trend because data centers suddenly want to become utilities. It is happening because the old dependency model is becoming too slow and too risky for AI infrastructure.
The grid is under pressure. Large-load regulation is tightening. Interconnection timelines are stretching. Utilities are rethinking how massive customers should pay for upgrades. At the same time, AI facilities are demanding more power, faster, with tighter uptime expectations and less tolerance for uncertainty.
That combination forces a different posture.
The most sophisticated data center operators are no longer asking only, “Can the utility serve this site?” They are asking, “What power can we control ourselves, what power can we coordinate with the grid, and what electrical architecture gives us the most flexibility over the next ten years?”
Microgrids, CHP, on-site generation, batteries, and behind-the-meter systems are all possible answers. None of them work well if the switchgear and switchboards are treated like afterthoughts.
The power strategy has to reach all the way into the distribution architecture.
For Moonshot, that is the opportunity. The company is not just supplying gear for projects that already know what they need. It can help shape the electrical backbone for data centers that need to operate through grid uncertainty, curtailment risk, interconnection delays, and phased AI growth.
If you are evaluating on-site power, planning a grid-parallel system, or designing a facility that may need to island under emergency or curtailment conditions, early electrical coordination matters.
https://moonshotus.com/request-form/
Because energy independence is not created by putting generation on site.
It is created by designing the power system so the facility can actually use it.

