For the last two years, most of the AI infrastructure conversation has been about chips. Who gets the GPUs? Who gets priority access? Who has the capital to buy the next generation of accelerators? Who can build enough compute fast enough to serve the demand?
That story is still true, but it is no longer the whole story.
A data center can have financing, land, tenants, engineering, permits, and GPU commitments and still fail to energize on time because the electrical infrastructure cannot get there fast enough. The constraint has moved upstream from the server cabinet to the electrical yard. Transformers, switchgear, breakers, bus systems, and control equipment are now controlling the schedule on projects that everyone thought were only limited by compute.
The hard reality is this: AI capacity does not exist when GPUs are ordered. It exists when power is delivered.
That is why transformer availability has become one of the most important infrastructure issues in the AI buildout. Reuters reported that average U.S. delivery times in the second quarter of 2025 were 143 weeks for generation step-up transformers and 128 weeks for power transformers, according to Wood Mackenzie. Reuters also reported that U.S. demand for generation step-up transformers soared 274% between 2019 and 2025, driven by grid expansion, renewables, data centers, manufacturing, EVs, and aging infrastructure replacement.
That means a transformer decision can easily become a three-year decision.
For data centers trying to deploy capacity in 18 to 24 months, that math is ugly.
The bottleneck is no longer theoretical
The electrical equipment shortage is not a niche procurement problem hiding in the background. It is now visible enough that policymakers, manufacturers, utilities, and developers are all reacting to it.
The Department of Energy says distribution transformers are facing supply chain constraints that result in long lead times and component shortages, and DOE has convened a working group to reduce transformer production times and improve communication across manufacturers, utilities, distributors, and engineers.
Utility Dive reported in April 2026 that the U.S. government had invoked Defense Production Act authorities around critical domestic energy capabilities, with the White House calling U.S. capacity to produce and deploy transformers, high-voltage transmission components, advanced conductors, power electronics, substations, and grid-supporting manufacturing equipment “dangerously limited.” Utility Dive also reported that distribution transformer backlogs were running a year or more, roughly twice the historical lead time, according to the National Electrical Manufacturers Association.
This is what makes the situation so painful for AI data center developers. The supply chain is not short on one thing. It is short across the electrical stack.
Reuters reported 2025 market shortages of 40% for high-voltage switchgear, 25% for breakers, and 20% for medium-voltage switchgear, according to Wood Mackenzie estimates.
That creates a compounding problem. Even if a transformer slot opens up, the project may still be waiting on breakers, switchgear, controls, or related components. Even if the data center gets utility approval, the physical equipment required to connect and distribute power may not be available on the project timeline.
That is why the old procurement mindset does not work anymore. You cannot treat electrical gear as a normal buyout package when the market is acting like a constrained commodity market.
The false comfort of “we already ordered it”
A common mistake in this market is assuming that placing an order eliminates supply chain risk.
It does not.
It may reduce one risk while creating others. Long-lead equipment orders are often based on incomplete assumptions because projects are trying to secure manufacturing slots before the utility study, final fault current, tenant load profile, and expansion plan are fully settled. That creates a dangerous tradeoff.
Wait too long, and you may lose the manufacturing window.
Order too early, and you may lock in the wrong configuration.
Picture a 200 MW AI campus trying to move quickly. The project team knows transformers and switchgear are tight, so they push procurement early. The early design assumes a certain utility service configuration, transformer size, fault current range, and phasing plan. The project places deposits, secures production slots, and feels like it has protected the schedule.
Then the utility changes the service plan. Or the tenant increases rack density. Or the owner decides phase two needs to come online earlier than expected. Or a breaker family becomes unavailable and procurement proposes a substitution.
Now the project has equipment on order, but the design assumptions underneath that order are no longer clean.
That is the difference between buying equipment and designing infrastructure.
In a normal supply chain environment, projects can absorb more adjustment. In this market, mistakes become expensive because replacement options are scarce. If a transformer is wrong, the replacement may not be months away. It may be years away. If switchgear needs to be redesigned around different ratings, the project may lose its place in the queue. If a breaker substitution disrupts coordination or listing integrity, the team may not have an easy alternative.
The cost of being wrong has increased.
Domestic versus offshore is really about risk concentration
Offshore sourcing can look attractive on paper. The price may be lower. The quoted lead time may seem competitive. The project team may feel like it has found a way around the domestic backlog.
Sometimes that may be true. But in mission-critical infrastructure, the cheapest route is not automatically the lowest-risk route.
Moonshot’s own analysis of domestic versus offshore electrical manufacturing gets to the heart of the issue: in high-density AI environments, electrical distribution sits on the critical path, and delays, listing issues, or quality defects ripple through construction schedules, financing timelines, and energization milestones. Moonshot frames the decision between domestic and offshore fabrication not as geography, but as risk concentration.
That is the right lens.
Offshore procurement can introduce several layers of exposure at once. There is shipping volatility, customs risk, tariff uncertainty, documentation friction, communication delay, rework exposure, and less direct visibility into material sourcing or component substitutions. If a quality issue appears after equipment arrives, the feedback loop is longer. If documentation does not satisfy the AHJ, the delay lands at the worst possible time. If the equipment needs rework, the project is now coordinating across distance, time zones, and logistics constraints.
Moonshot’s domestic manufacturing article also points out that domestic vertically integrated operations allow immediate engineering review, rapid correction, direct communication between design and fabrication, controlled torque verification, and quality checks.
That is not just operational convenience. In today’s market, that is schedule protection.
When electrical gear is scarce, visibility becomes valuable. Speed matters, but controlled speed matters more. A faster quote does not help if the project later gets stuck on documentation, field labeling, rework, or rating mismatch.
The supply chain lesson is straightforward: long lead times make quality and coordination more important, not less.
Solid-state transformers are part of the future, but not a magic escape hatch
The transformer shortage is also one reason solid-state transformers are getting attention in AI data center design. SSTs are not just being discussed because they are more futuristic. They are being discussed because traditional transformer supply chains are slow, custom, and capacity-constrained.
In a pv magazine interview, SolarEdge described solid-state transformers as a response to AI power density. SolarEdge said most data centers today may sit around 10 kW per compute rack, while the next generation entering AI data centers may be around 100 kW per rack, and the following generation could approach 1 MW per rack on the same footprint. The company argued that current infrastructure physically cannot sustain that kind of density.
The same pv magazine article explains why SSTs are different. SolarEdge described an SST as replacing not just the transformer, but also the converter, functioning like a transformer and inverter in one device. It also described a direct medium-voltage-to-DC architecture, with the manufacturer expecting 99% conversion efficiency from its in-development product.
That matters for the supply chain because an SST manufacturing model may look more like power electronics manufacturing than conventional transformer manufacturing. In the article, SolarEdge said conventional transformers are often custom-built and require orders years in advance, while SST designs can use modular, mass-produced subcomponents that may support faster delivery.
That is promising, but it does not mean projects can ignore today’s reality.
SSTs are not yet the default answer for most data center builds. The standards, service ecosystem, AHJ familiarity, protection schemes, and integration models still need to mature. The market will likely move through a hybrid phase where traditional AC systems remain in place while higher-voltage DC and SST-based architectures appear first in specific high-density zones, modular deployments, pilot projects, and next-generation AI halls.
The practical point is not “wait for SSTs.”
The practical point is design today’s infrastructure so it does not block tomorrow’s options.
That means owners should think about transformer procurement, switchboard design, voltage architecture, DC-readiness, space planning, fault current exposure, battery storage, and future conversion equipment as one connected strategy. The projects that win will not be the ones that guess perfectly. They will be the ones that preserve flexibility while others are boxed into rigid designs.
Waiting is now a strategy, and usually a bad one
Some project teams still treat the supply chain problem like weather. It is inconvenient, but eventually it passes. That mindset is risky.
Reuters reported that major manufacturers are investing heavily in U.S. grid equipment capacity. Hitachi Energy announced more than $1 billion in U.S. manufacturing infrastructure investments, Siemens is building a $150 million transformer factory in Charlotte, Hyosung HICO is expanding its Memphis transformer plant, and WEG is investing in specialty transformer manufacturing in Missouri.
Those investments matter, but new factories do not solve current projects overnight. Reuters reported that Wood Mackenzie expects GSU demand to be about 100% higher than supply in 2025, with the gap falling below 10% by 2030 as new capacity comes online.
That tells owners something important: relief is coming, but the market is still tight now.
A developer waiting for the supply chain to normalize may be waiting through the most important years of the AI buildout. Meanwhile, competitors are securing equipment, finding domestic partners, redesigning around available gear, using modular approaches, and locking in the electrical path before GPUs arrive.
The better strategy is not passive waiting. It is constraint-aware design.
Constraint-aware design starts by admitting that the supply chain is part of the engineering problem. If transformer availability is uncertain, the project should evaluate alternate phasing, temporary capacity, modular substations, stocking partnerships, mobile substations, staged energization, breaker flexibility, and switchboard designs that preserve future options. If conventional transformer lead times are too long, the project should investigate whether available inventory, domestic sourcing, or alternate voltage strategies can change the schedule. If SSTs may become viable for later phases, the current design should avoid decisions that make future integration impossible.
This is where Moonshot’s transformer offering becomes relevant. Moonshot states that it is an authorized distributor with direct partnerships with leading transformer manufacturers worldwide, offering pole mount, pad mount, substation transformers, and mobile substations for public utilities, data centers, and solar power. Moonshot also says it has strategic partnerships with stocking partners across the U.S., giving customers access to a large selection of in-stock transformers.
That is not just a procurement bullet point. In this market, access to inventory and stocking relationships can directly affect whether a project energizes on time.
Moonshot’s real advantage is vertical coordination
This is where the Moonshot connection should be direct.
The market does not need another company saying, “We can get gear.” Everyone says that until the gear is late.
The stronger position is this: Moonshot reduces supply chain risk because it connects engineering, procurement, manufacturing, and field reality earlier in the project.
Moonshot’s UL891 switchboard page says its engineering team designs solutions based on supply chain constraints and maximum flexibility. It also notes that Moonshot can use major breaker manufacturers, which gives the company an edge in building UL-listed switchboards and control panels faster and more cost-effectively. Moonshot also says it evaluates site design during the sales process to provide code-compliant, cost-effective, quick solutions.
That is exactly the message this blog should reinforce.
Vertical integration is not just a nice internal capability. It is how you respond when the electrical supply chain becomes the schedule driver. If engineering and fabrication are disconnected, every substitution becomes slower. If procurement and design are disconnected, every available component may create a downstream compliance issue. If manufacturing and field requirements are disconnected, every late change becomes a potential inspection problem.
Moonshot can position itself against that risk.
The argument is not that Moonshot can make all shortages disappear. Nobody can. The argument is that Moonshot helps customers make better decisions inside a constrained market.
That includes:
- Designing around available equipment without compromising ratings or compliance
- Aligning breaker choices with coordination studies and UL listing requirements
- Using domestic manufacturing to reduce logistics volatility
- Leveraging transformer stocking relationships where possible
- Planning around future SST and DC architecture without overcommitting too early
- Reducing the chance that a procurement shortcut becomes an energization delay
That is the commercial value.
In a loose market, buyers can afford to treat switchboards and transformers as procurement categories. In a tight market, those decisions become strategic. Moonshot’s role is to make them strategic before they become urgent.
The real takeaway
The AI infrastructure race is not just a GPU race anymore.
It is a power equipment race.
Transformers, switchgear, breakers, bus systems, and control equipment are now determining which projects become real capacity and which remain announcements. The market is short, lead times are long, prices are under pressure, and domestic capacity is still catching up.
That does not mean every project is stuck. It means projects need to design around the constraint instead of pretending it is temporary background noise.
The winning teams will secure electrical pathways early, validate equipment against real fault current and load requirements, preserve flexibility for future voltage architectures, and choose partners who can connect engineering decisions to manufacturing reality. The losing teams will treat transformers and switchboards like late-stage procurement items and discover too late that the queue controls the schedule.
For Moonshot, this is a direct sales argument.
Domestic manufacturing, vertical integration, breaker flexibility, transformer stocking relationships, and engineering-led design are not secondary differentiators. They are exactly what the market now rewards.
If you are planning a high-density AI deployment, waiting for the transformer market to normalize is not a plan. Designing around the constraint is.
https://moonshotus.com/request-form/
Because in AI infrastructure, the bottleneck is no longer whether you can buy the compute.
It is whether you can power it.
Source:
https://pv-magazine-usa.com/2026/03/26/ai-data-center-developers-eye-solid-state-transformers-for-ai-power-density/
https://www.pv-magazine.com/2026/04/01/solaredge-outlines-path-to-800-v-dc-data-center-infrastructure/
https://moonshotus.com/domestic-electrical-manufacturing-vs-offshore-risk-compliance-and-lead-time-tradeoffs/
https://moonshotus.com/ul-891-switchboards/
https://moonshotus.com/transformers/
https://www.energy.gov/oe/distribution-transformers
https://www.utilitydive.com/news/what-does-trumps-wartime-powers-flex-mean-for-the-transformer-shortage/818159/
https://www.reuters.com/business/energy/grid-equipment-makers-invest-us-ease-sup