New Datacenters Can’t Launch Because of Transformers Shortage
For two years, the industry has agreed on the diagnosis. AI data centers cannot get power because the interconnection queue is broken. Fix the queue, and the power follows.
Regulators have now acted on that diagnosis. In June 2026, the Federal Energy Regulatory Commission issued show cause orders to every jurisdictional grid operator in the country — PJM, MISO, SPP, CAISO, ISO New England and NYISO — directing them to justify their large-load interconnection rules or revise them, with responsive tariff filings due within 60 days. PJM has launched an expedited interconnection track. ERCOT is processing large-load applications in batches.
The queue is being addressed. The data centers still will not get power on schedule, because the queue was never the binding constraint.
Announced Capacity Is Not Energized Capacity
One figure reframes the entire discussion. Of roughly 12 GW of US data center capacity announced for 2026, less than half has broken ground. The reason cited is not permitting and not queue position. It is the unavailability of high-voltage transformers, switchgear and grid-tie infrastructure.
A grid operator can be compelled by regulatory order to complete an interconnection study within 90 days. No order can compel a transformer core into existence.
The Transformer Position in Numbers
Large power transformers and generator step-up units are the chokepoint, and the position has deteriorated sharply:
- Lead times have moved from 6 to 12 months before 2020 to 24 to 48 months in 2026, with some large units extending further.
- Standard power transformers now average approximately 128 weeks to delivery. Generator step-up units average approximately 144 weeks.
- Roughly 80% of the large power transformers used in the United States are imported.
- Cleveland-Cliffs is the only domestic US producer of grain-oriented electrical steel, the specialized material transformer cores require.
Those last two facts deserve to be read together. The United States imports four out of every five large transformers it installs, and the specialized steel behind them has a single domestic source.
Utilities have adjusted accordingly. Firms that historically procured equipment a year in advance are now ordering transformers up to five years ahead, for projects they anticipate rather than projects they have committed to. Defensive procurement at that scale is a clear signal of what the industry expects.
Why This Constraint Will Not Yield to Regulation
Each previous bottleneck in this build-out yielded to something. Capital resolved the compute shortage. Regulation is now compressing interconnection timelines. Neither of those instruments builds a steel mill.
The shortage is structural rather than cyclical. Three demand curves — general electrification, renewables buildout and now AI data centers — are rising simultaneously against manufacturing capacity that remained flat through a decade of underinvestment. New transformer factories take years to commission. New steel capacity takes longer still. And a transformer plant cannot outrun its steel supply regardless of how full its order book becomes.
The binding constraint has migrated from administrative process to physical goods, and it is now migrating again — upstream, toward raw materials. Anyone forecasting data center energization from announced capacity, or from interconnection queue position, is measuring the wrong variable.
How Developers Are Responding
Consider the position of a developer facing a four-year wait for a generator step-up transformer and associated substation equipment. The rational response is to select a generation technology that does not require them.
This is the substantive reason on-site power is expanding. Fuel cells, reciprocating gas engines and linear generators connect at distribution voltage — 480 volts natively, or up to 34.5 kV with a modest step-up — and therefore sidestep the constrained equipment class almost entirely. Bloom Energy’s servers connect to existing natural gas pipelines rather than waiting for a substation.
In our assessment, this makes the on-site power advantage more durable than the regulatory headlines suggest. Interconnection reform is arriving. Transformers are not.
There is a regulatory consequence that compliance teams should register. That same distribution-voltage connection is what keeps most of this generation outside NERC’s Category 2 registration criteria, which require connection at 60 kV or above. Substantial quantities of inverter-based generation are being installed below the reliability perimeter, with consequences for load forecasting and system planning that we examine in detail in our Bloom Energy analyst report.
Where Chinese Manufacturing Relieves the Constraint
Chinese manufacturers identified the shortage and moved into it at scale.
Chinese transformer exports were reported up approximately 27% year on year in early 2026, at record trade values. TBEA and Sieyuan Electric are among the manufacturers executing large foreign orders, and European order growth has been reported well above 100%. When a Texas AI data center encountered voltage-related downtime, it reportedly bypassed local grid timelines entirely by outbidding competitors for custom transformers sourced directly from China.
For a buyer that needs heavy high-voltage hardware on a compressed timeline, Chinese manufacturers are currently the fastest source available globally. That is a procurement reality rather than a policy position.
The asymmetry this creates
The strategic implication is less widely appreciated, and it runs in an unexpected direction.
Chinese transformer exports represent a more significant long-term challenge to on-site power suppliers than a Chinese fuel cell would. On-site generation commands a premium because grid connection is slow. Grid connection is slow because high-voltage equipment is unavailable. Every Chinese transformer delivered into a Western data center project therefore shortens the timeline of the alternative to on-site generation.
A Chinese entrant into the stationary fuel cell market would compete with Bloom’s product. Chinese transformer exports compete with Bloom’s reason for existing. The second is the more consequential, it is already occurring at record volume, and the first is not imminent.
This applies across the sector — to Mainspring Energy, to the gas turbine developers, and to every other on-site generation supplier. It is a market-level observation rather than a company-specific one.
Why a Chinese Equivalent Is Not Imminent
Given what Chinese manufacturers achieved in solar photovoltaics and battery cells, the question is a reasonable one. The comparison breaks down on engineering rather than on politics.
China’s fuel cell industry is built around hydrogen PEM systems for commercial vehicles — buses, logistics trucks and forklifts. These are compact units optimized for vibration tolerance, intermittent duty cycles and rapid cold starts in sub-zero conditions. Bloom Energy builds stationary solid oxide systems operating at approximately 800°C for thousands of hours continuously.
These are not adjacent products sharing a common platform. Converting one into the other is not a product extension but a separate engineering program, and Bloom has been executing that program since 2001.
Sector conditions compound the mismatch. China’s listed fuel cell pure-plays have been contracting rather than expanding, with revenues declining and losses widening through 2025 as domestic subsidy cycles tightened. Companies in that position reduce cost and preserve cash; they do not open new fronts in foreign markets requiring different technology.
Three further barriers would remain even if a competitive stationary product existed:
| Barrier | Why it binds |
|---|---|
| Fuel infrastructure | Bloom’s underrated advantage is that its servers operate on the existing natural gas pipeline network. China’s fuel cell ecosystem is hydrogen-first. A Western data center cannot operate a hydrogen server without reliable large-scale hydrogen delivery, which does not exist at the required scale or price. |
| Critical infrastructure treatment | Hyperscale data centers hosting cloud infrastructure, corporate data and AI model training are treated as critical infrastructure. Foreign-managed power systems with proprietary monitoring and control software at the center of such a facility would face significant procurement and regulatory resistance. |
| Domestic content incentives | US clean energy deployment benefits from substantial domestic-content-linked tax incentives. A buyer selecting imported equipment forfeits benefits that materially affect project economics. This is a subsidy differential rather than a tariff, and it is harder to work around. |
The Compliance Exposure Inside a Procurement Decision
This is the point at which a supply chain question becomes a compliance question, and it is the section we would direct compliance readers toward.
Consider the facts in combination. Approximately 80% of large US transformers are imported. There is one domestic source for the specialized steel. Western lead times extend to four years. Chinese suppliers can deliver faster than anyone, and buyers under schedule pressure are going to them.
Now add the regulatory direction of travel. A September 2025 Federal Register notice addressed revisions to supply chain risk management reliability standards concerning equipment and services produced or provided by entities identified as risks to national security. That sits alongside NERC’s existing CIP-013 supply chain risk management standard, which governs controls for BES Cyber Systems and carries penalty exposure of up to approximately $1 million per day, per violation.
A registered entity resolving a four-year equipment lead-time problem by procuring from the fastest available foreign supplier may be creating a supply chain compliance problem in the process. Schedule pressure and CIP-013 obligations currently point in opposite directions, and the procurement decision is usually made by people who do not own the compliance risk.
To be clear, importing transformers is routine, and at 80% of the installed base it is unavoidable. The point is narrower and more practical. Sourcing decisions taken under schedule duress are precisely the decisions that later appear in audit findings, and regulatory attention on foreign-sourced bulk power system equipment is increasing rather than diminishing.
Four steps for compliance teams
- Establish scope with procurement. Vendor selection for BES-connected equipment sits inside your supply chain risk management program, not outside it as a purely commercial decision.
- Document the risk assessment at the time of the decision, not retrospectively, for any equipment sourced from a jurisdiction subject to heightened scrutiny.
- Track the supply chain standards revisions actively. This is a moving target and the direction is toward more restriction.
- Extend the same scrutiny to embedded software and remote monitoring. That is where the cyber exposure in modern grid equipment actually resides, rather than in the steel and copper.
What to Watch
- Transformer lead times. The master variable for the entire build-out. Movement back below roughly 24 months would materially change the competitive position of on-site generation.
- Domestic steel and transformer manufacturing announcements. The only durable fix, and the slowest to arrive.
- Chinese export volumes, and any restriction placed on them. Currently the relief valve, and simultaneously the compliance exposure.
- Gas pipeline capacity in regions with heavy on-site deployment. When developers select on-site gas, the bottleneck relocates rather than disappears.
- Skilled labor costs for high-voltage commissioning and the electrical trades. The least tracked and most consistently underestimated constraint in the build-out.
Our Assessment
The equipment bottleneck persists through at least 2028. Interconnection reform will compress administrative timelines meaningfully, but it will not deliver the energization dates it implies, because the hardware is not available. The gap between reformed process and delivered connection is likely to become a visible source of frustration, and possibly of further regulatory intervention.
On-site generation therefore retains a genuine timeline advantage for longer than the regulatory record suggests. That advantage is regionally uneven, and it is being eroded from an unexpected direction — not by a competing generation technology, but by foreign manufacturing capacity making the grid alternative viable again.
The bottleneck to watch, in the end, is not electrical at all. It is a steel mill.
For the full analysis — including the ranked bottleneck register, the regional differences between PJM and ERCOT, and what this means for your build schedule — please contact us.
GO-IBR.com Research is a market intelligence and advisory firm covering the energy sector — regulatory compliance, cybersecurity, and the application of AI to grid operations. This article is general educational and market information, not legal, compliance or investment advice. Lead times, export figures and supply estimates are drawn from public reporting, vary by source and specification, and should be treated as indicative.
