The Grid Interconnection Crisis
What's happening in grid queues, transformer supply chains, and transmission permitting — and why it is the macro constraint behind datacenter growth 2026–2028.
Standfirst: Three years ago, "build it and they will come" was the playbook for hyperscale datacenter development. Now the bottleneck has shifted from land, power purchase agreements, and construction crews to something far more fundamental: the physical grid itself. Projects are stuck in interconnection queues totaling over 2,000 gigawatts nationwide. Transformer lead times have stretched to four years. In Texas alone, the datacenter interconnection queue has nearly quadrupled in a single year. This is not a cyclical tightness that will ease when capex rises or permits fast-track. It is structural.
The Queue: 2,061 Gigawatts of Paper Capacity
As of 2026, the U.S. interconnection queue contains 2,061 GW of generation and storage projects seeking grid connection [1]. That figure sounds enormous — until you learn that it represents a decline from roughly 2,300 GW at the end of 2024, driven almost entirely by withdrawals rather than projects advancing to energization. Berkeley Lab's "Queued Up:" research documents a long-term pattern: approximately 75% of queued projects never reach commercial operation, pulled out by financial sponsors who no longer find their assumptions viable once the real costs of transmission builds and queue delays materialize [1].
The average queue wait for projects actually energized in 2024 was 55 months — more than four and a half years from application to first electrons [1]. That timeline makes sense only if you understand the architecture of the problem: each project undergoes feasibility, system impact, and facility studies; a positive outcome does not guarantee transmission build-out within the developer's budget window; and the iterative nature of study updates means applicants constantly revise their positions, further delaying everyone else in line.
The International Energy Agency estimates global stalled projects at approximately 2,500 GW across all regions, confirming that the United States is one node in a worldwide crisis rather than an outlier [1]. The composition matters: solar accounts for roughly 773 GW of the U.S. queue, storage 749 GW, and gas-fired generation 253 GW — but the datacenter wave is reshaping what comes next.
FERC Order No. 2023: Reform Without Relief
In July 2023, the Federal Energy Regulatory Commission issued Order No. 2023, described as the largest U.S. interconnection reform in two decades. It replaced the old serial "first-come-first-served" study process with a cluster-based "first-ready-first-served" model, mandated firm deadlines, financial deposits, and site control requirements. In theory, it addressed the self-reinforcing withdrawal cascade that had plagued queues for decades [5] — by raising the cost of free-riding on study results while reducing the incentive to withdraw only when conditions worsen.
Two years in, the reform has not yet turned the numbers. Queue sizes remain near record levels, and the 55-month median wait persists. Order No. 2023's effects may appear in the cohort of projects that apply after full implementation, but the existing backlog must still run its course — a pipeline problem whose resolution requires time, not just policy.
Transformers: The Worst Link in the Supply Chain
If the interconnection queue is the administrative bottleneck, transformer manufacturing is the physical one. A generator step-up (GSU) transformer — the critical piece of equipment that steps voltage from a power plant or datacenter's on-site generation to transmission levels — now carries an average lead time of 144 weeks, approximately 2.8 years [3]. Power transformers broadly average 128 weeks. Distribution transformers lag at around 52–65 weeks depending on configuration [3].
These are not normal lead times. For comparison, the pre-pandemic baseline for GSU transformers was measured in months, not years. The escalation reflects a confluence of factors: surging demand for datacenter-grade equipment, raw material constraints, and a supply base that cannot scale quickly because building new transformer manufacturing capacity takes two to four years from announcement to first units.
Demand for GSU transformers specifically rose 274% between 2019 and 2025 according to Wood Mackenzie survey data cited by industry publications and verified through CoBank sector reviews [3]. Overall power transformer demand grew 119% over the same period. Manufacturers including Eaton, Siemens Energy, Hitachi Energy, and others have committed approximately $2 billion in new or expanded capacity since 2023 — substantial investment, but insufficient to close the current gap within the relevant timeline [3].
Material Scarcity: The GOES Problem
A less visible but equally constraining factor is grain-oriented electrical steel (GOES), the specialized silicon steel used in transformer cores. GOES costs have doubled since January 2020 [3]. Perhaps most critically, Cleveland-Cliffs is the sole domestic producer of GOES in the United States, creating a single-source dependency for American transformer manufacturers who cannot simply pivot to an alternative supplier when production rates hit maximum capacity [3].
The pricing picture tells the same story. Power transformer prices are up approximately 77% since 2019. Distribution transformers have surged 78–95%. Generator step-up units, ironically the least volume-intensive category, have risen about 45% — a modest relative increase that belies the absolute cost pressure on individual projects requiring multiple GSU installations [3].
Medium-voltage switchgear, a component often overlooked outside the procurement team but essential for every substation and distribution connection, now faces procurement times of 52–80 weeks according to active contractor reservation data from Terrapin Construction Group, with pad-mount distribution transformers at 40–65 weeks [3]. These timelines mean that even when a datacenter gets grid connection approval, the equipment required to make that connection physically operable may not arrive for another 18 to 24 months.
Regional Breakdown: The United States
Texas: Where Datacenter Demand Meets a constrained Grid
Texas presents the starkest case study in concentration risk. ERCOT's large-load interconnection queue grew from 63 GW in September 2025 to 226 GW reported by September 2026 — a nearly fourfold increase in twelve months, driven overwhelmingly by datacenter demand [4]. That snapshot has since been overtaken: by June 2026 the ERCOT large-load queue had reached approximately 466 GW of requested capacity, roughly 90% of it data centers [4a], and ERCOT documents and Governor Abbott cite large-load requests soaring from about 48 GW in 2023 to more than 474 GW — a "ghost demand" problem that led Texas in 2026 to become the first major US hub to freeze new large-load grid connections while auditing requestors [4b]. To put the scale in perspective: even the older 226 GW figure represents approximately twice Texas's entire current peak electricity demand of roughly 80 GW. This is an unprecedented concentration of AI-driven load in a single balancing authority.
The implications extend beyond ERCOT's borders. Suppliers serving Texas datacenters — Eaton, Caterpillar, Vertiv — report order backlogs stretching years into the future. Eaton's data center backlog reached 307 GW as of Q2 2026, equivalent to 15 years at 2025 build rates [6]. Caterpillar's Power & Energy segment reported Q2 2026 sales surpassing $20 billion, with customers placing orders through 2030 and the company's total order backlog sitting at $72 billion with approximately 59% in power and energy [7].
Vertiv, the leading provider of datacenter power infrastructure and cooling systems, raised full-year 2026 guidance to $14.0 billion in revenue — a 37% year-over-year increase — with deferred revenue of $3.634 billion representing advanced customer payments for contracted but undelivered equipment [5]. GE Vernova's electrification segment booked $2.4 billion in datacenter orders in Q1 2025 alone, exceeding all of its annual guidance with a book-to-bill ratio near 2.5 [6].
Eaton's CEO Paulo Ruiz stated on the Q2 2026 earnings call that the company's data center organic revenue grew 65% during the quarter, significantly outperforming the underlying market growth rate of 23% [6]. And the company just acquired Boyd Thermal to complete its "grid to chip" portfolio — covering everything from the utility substation down to semiconductor-level liquid cooling [6].
What the Numbers Mean
Taken together, these figures tell a consistent story: the tier-one power equipment suppliers are experiencing demand acceleration that far exceeds historical cycles. Caterpillar's customers booking through 2030 suggests that the current crisis extends well beyond 2026–2027. Vertiv's deferred revenue of $3.6 billion implies massive contracted work that, even assuming perfect execution, cannot be delivered faster than the supply chain allows.
The EEI reports that its member companies invested $208 billion in 2025, up from $139.8 billion in 2020 — a 49% increase in five years [3]. S&P Global Regulatory Research projects $1.295 trillion in utility capital expenditure between 2026 and 2030, an upward revision from $1.169 trillion just four months earlier — and that figure represents planned spending, not actual delivered infrastructure [3]). EIA data shows distribution capex rose $31.4 billion (a 160% increase) between 2003 and 2023 alone, with $6.5 billion of that growth occurring between 2022 and 2023 in a single year [3]).
But here is the critical gap: planned capex does not equal installed capacity. The EEI projects $1.1 trillion in cumulative capex for 2025–2029 across its membership, and transmission investment alone reached $32.6 billion in 2024 with projections of $39.9 billion for 2025 [3]). Spending trillions on planning, engineering, and materials procurement is only useful if the manufactured components — transformers, switchgear, cable — arrive on schedule. And they are not.
Regional Breakdown: Europe and China
Europe: €584 Billion Required Before Decade's End
The European situation is structurally similar but quantitatively distinct. The European Commission estimates €584 billion in grid investment required before the end of the decade — a figure that underscores the scale of underinvestment accumulated over the past two decades of slow renewable integration and political resistance to new transmission corridors [2].
In December 2025, the EC proposed the European Grids Package, which includes a controversial "tacit approval" mechanism: if local authorities do not issue a permit within two years for standard projects or three years for complex ones, the permit is automatically granted [2]. This administrative shortcut addresses the permitting bottleneck but does nothing to resolve the manufacturing timeline constraint. You can fast-track a permit for a substation, but if the transformers take four years to manufacture, the tacit approval changes nothing for projects starting today.
The Connecting Europe Facility for Energy is being raised from €5.84 billion (2021–2027) to €29.91 billion (2028–2034) — an increase that represents roughly 50× the prior envelope [2]. While dramatic proportionally, this is still a small fraction of the €584 billion total requirement.
China: The Counterexample
China provides the only example of grid expansion operating at velocity comparable to demand growth. State Grid Corporation committed 4 trillion yuan (approximately $580 billion) for the 2026–2030 period — a 40% increase over its prior five-year plan — targeting 15 new ultra-high-voltage (UHV) transmission lines [1]. During the 2021–2025 plan, China's UHV network grew from approximately 28,000 km to over 40,000 km, operating 45 UHV projects (19 alternating current plus at least 23 direct current) [1].
The key metric where China excels is the approval-to-energisation cycle: approximately two years, compared to the Western 10–20 year permitting cycle [1]. Changji-Guquan, the world's longest UHV link at 3,293 kilometers operating at 1,100 kV, can carry 12,000 megawatts between remote generation and coastal load centers. China adds roughly 12,000 km of UHV annually; the West adds perhaps 1,000 km of new high-voltage lines per year amid litigation, environmental review, and local opposition.
This disparity matters for the global compute supply chain because China's ability to expand grid capacity at speed reduces the risk premium on building datacenters there — but U.S. export controls on advanced semiconductors and training infrastructure largely nullify that advantage for frontier AI workloads. The result is a paradox: China can build grids fast, but Western restrictions limit the equipment that would justify the biggest datacenter investments.
The Demand Trajectory
All of the supply-side constraints above are accelerating precisely because demand is compounding exponentially. According to S&P Global and Commodity Insights, datacenter grid-power demand is projected to rise 22% in 2025, then nearly triple by 2030 [7]. Global datacenter capacity is expected to nearly double from approximately 100 GW today to 200 GW by 2030 [7]. SP Global's own calculation places total grid-power demand at 143 GW by 2030 — concentrated in "a handful of regions rather than spread," amplifying local grid stress where it matters most [7]).
Goldman Sachs estimates that approximately $720 billion in grid upgrades is required through 2030 to accommodate this growth [7]). That figure sits somewhere between the EEI's cumulative projections and the EU's standalone requirement — suggesting the global investment gap is even wider than any single-region analysis reveals.
Reconductoring: The Underutilised Option
Not every solution is mired in multi-year manufacturing delays. One rapidly deployable intervention deserves serious attention: reconductoring. Replacing conventional aluminum conductors with advanced composite-core designs in existing rights-of-way could roughly double American transmission capacity by 2035 without requiring any new land acquisition, corridor approvals, or litigation [1].
The economics are compelling: the approach could cut wholesale electricity costs by 3–4%, save approximately $85 billion in system costs by 2035 and $180 billion by 2050, and deliver capacity gains in months to a year — compared to the 10–15 years required for greenfield new-line construction [1]. It substitutes materials science for legal process, which is arguably the most valuable capability in American infrastructure development right now.
Reconductoring has received relatively limited public policy attention, likely because it operates invisibly inside existing corridors and generates fewer ribbon-cutting opportunities than new line projects. But in a context where transmission expansion is fundamentally constrained by permitting and siting rather than engineering, leveraging the thousands of miles of existing right-of-way offers the fastest path to meaningful capacity gains available in the current timeframe.
Forecast Scenarios
Based on the evidence assembled above, here are three plausible pathways for the grid-interconnection landscape through 2028.
Base Case: Continued Bottlenecks Through 2027–2028
Manufacturers' committed $2 billion in expansion proceeds at planned pace, delivering incremental capacity in 2027–2028 but not closing the current gap. Transformer lead times stabilize at 120–140 weeks (still elevated but no longer increasing). ERCOT queue growth moderates to 300–350 GW as some projects withdraw on their own while grid upgrades begin absorbing load. Reconductoring pilots achieve early deployments but remain geographically limited. China continues aggressive UHV buildout. Interconnection reform effects begin appearing for cohorts applying post-2025 but legacy backlog persists.
Probability: 50–60%
Bull Case: Manufacturing Scaling Outpaces Demand
Unexpectedly rapid ramp of new transformer capacity — particularly if Siemens Energy's Charlotte facility reaches full capacity ahead of the 2027 projection and if additional manufacturers accelerate investment commitments beyond the current $2 billion. Demand growth remains strong but manageable: datacenter build-out stays near SP Global's 22% annual growth without accelerating unexpectedly. Reconductoring achieves broader adoption, adding meaningful transmission capacity faster than anticipated. EU grid package receives accelerated funding.
Probability: 20–25%
Bear Case: Cascading Delays
Multiple simultaneous failures: GOES supply disruptions constrain transformer output despite expanded factory floors, causing lead times to stretch beyond 160 weeks. ERCOT queue balloons past 300 GW as virtual sinking compounds physical constraints. Multiple utilities face material cost escalation that triggers regulatory disputes, delaying recoveries and deterring investment. Demand accelerates: AI model training clusters grow so large that power-per-rack specifications exceed current grid-tie limits in major markets. A combination of transformer shortages, switchgear unavailability, and queue paralysis forces several hyperscalers to defer or cancel planned builds.
Probability: 15–20%
What This Means for Compute Buyers
For organizations planning datacenter deployment, GPU fleet acquisition, or cloud infrastructure strategy, the grid crisis translates into several non-negotiable realities:
Timeline Risk Is Now the Primary Schedule Risk
Permitting, construction, and equipment procurement remain important, but the interconnection queue and transformer supply chain set the outer bounds of feasibility. A project submitted to the queue today should expect at least 55 months — possibly 70–80 months given recent queue congestion — before first power. Any deployment plan built on a 24-month timeline is aspirational, not realistic, unless the site already has confirmed grid access.
Action: Prioritize greenfield sites with pre-approved interconnection or locations where grid upgrades are already underway. Avoid brownfield sites dependent on queue position.
Budget for Escalation
Transformer price increases averaging 77% since 2019 have not bottomed. GOES doubling since January 2020, switchgear lead times of 52–80 weeks, and the single-source GOES dependency create ongoing upside risk on power infrastructure costs. Contracts signed today for delivery in 2027–2028 should include escalation clauses reflecting this trajectory.
Action: Model 25–40% escalation on power infrastructure line items over typical project horizons. Consider forward contracting transformer and switchgear procurement alongside site acquisition.
Supplier Relationships Matter More Than Ever
When Eaton reports 307 GW of datacenter backlog, Vertiv carries $3.6 billion in deferred revenue, and Caterpillar's customers are booking through 2030, the implication is that access to supply depends on how deeply embedded your organization is in the manufacturer relationship ecosystem. New entrants to the procurement queue may find themselves last in line for allocation even if budgets are fully funded.
Action: Secure supply commitments early — ideally concurrent with site selection, not after. Multi-year supply agreements provide both capacity assurance and potentially favorable pricing before late-cycle escalation.
Geographic Diversification Is Not Optional
With ERCOT carrying a large-load queue that has ballooned from 226 GW (Sep 2025–26 reporting) to ~466–474 GW of requests — and Texas now freezing new large-load connections altogether — and interconnection bottlenecks affecting virtually every major RTO, concentrating datacenter investment in a single geographic region creates systemic risk. Regions with proactive grid investment (parts of the Southeast, certain EU jurisdictions pursuing aggressive grid modernization, Chinese provinces with UHV backing) offer better near-term prospects.
Action: Map grid health metrics — queue size, transformer availability, utility capex trajectory, regulatory environment — alongside traditional site-selection criteria (land cost, energy cost, talent pool) into a composite grid-health score.
Secondary Markets Will Erode
When primary supply dries up, secondary markets emerge. The HBM3E export ban prediction (March 2027 deadline) illustrates exactly how quickly hardware scarcity can create black-market premiums. The same dynamic will apply to transformers, switchgear, and ultimately to completed datacenter space itself — operators with guaranteed grid access will command massive rent premiums over those still waiting in queue.
Action: Treat grid-connected capacity as a strategic asset class, not just operational infrastructure. Securing a site with confirmed interconnection today is analogous to buying server racks in a secondary market before the primary supply channel closes.
The additional source registry for this edition is maintained in the evidence ledger at data/platform-data.json; methodology and as-of conventions: methodology.
Key Metrics at a Glance
| Metric | Value | Source |
|--------|-------|--------|
| U.S. interconnection queue | 2,061 GW (shrank from ~2,300 GW) | Berkeley Lab / OsakaWire |
| Global stalled generation + storage | ~2,500 GW | IEA Electricity 2026 |
| Average queue wait (2024 cohort) | 55 months | Berkeley Lab EMP |
| Withdrawal rate from queue | ~75% | Berkeley Lab EMP |
| GSU transformer lead time | 144 weeks (~2.8 years) | Wood Mackenzie via Foley & Lardner |
| Power transformer average lead time | 128 weeks | Wood Mackenzie via CoBank |
| Transformer price increase (since 2019) | ~77% (power), 78–95% (dist.) | Wood Mackenzie / Foley & Lardner |
| GOES cost increase (since Jan 2020) | ~100% (doubled) | Industry publications |
| U.S. GOES single-source dependency | Cleveland-Cliffs only | Foley & Lardner analysis |
| ERCOT large-load queue growth | 63 GW → 226 GW reported (Sep 2026); ~466 GW by Jun 2026 per ERCOTqueue; >474 GW requests cited by ERCOT/Abbott; new connections frozen | Latitude Media / ERCOTqueue / BNN Bloomberg |
| Data center grid-power demand (2030) | Nearly triple 2025 level | S&P Global / 451 Research |
| Global datacenter capacity (2030) | ~200 GW (from ~100 GW today) | SP Global |
| Goldman Sachs grid upgrade estimate | ~$720B through 2030 | SP Global citing GS |
| EU grid investment required | €584B before 2030 | European Commission |
| China UHV investment (2026–2030) | 4 trillion yuan (~$580B) | State Grid Corporation |
| China UHV network (2021–2025) | 28,000 km → 40,000+ km | ChinaTalk / State Grid |
| China approval-to-energisation cycle | ~2 years | State Grid data |
| U.S. utility capex (2026–2030) | $1.295T (up-revised from $1.169T) | S&P Global Regulatory Research |
| EEI member investment (2025) | $208B (+49% from 2020) | EEI |
| Transmission investment (2024) | $32.6B; projected $39.9B (2025) | EEI / FERC tracking |
| Manufacturer capacity commitments (since 2023) | ~$2B total | Eaton / Siemens / Hitachi Energy filings |
| Reconductoring potential (by 2035) | ~2× U.S. transmission capacity | Analysis via OsakaWire |
| Vertiv FY2026 guided revenue | $14.0B (+37% YoY) | Vertiv Q2 2026 earnings |
| Vertiv deferred revenue (Q2 2026) | $3.634B | Vertiv Q2 2026 earnings |
| Eaton DC organic revenue growth (Q2 2026) | +65% vs 23% market growth | Eaton Q2 2026 earnings |
| Eaton DC backlog | 307 GW (= 15 years at 2025 build rate) | Eaton Q2 2026 earnings |
| GE Vernova DC electrification orders (Q1 2025) | $2.4B, BtB ~2.5 | GE Vernova disclosures |
| Caterpillar P&E sales (Q2 2026) | >$20B, backlog through 2030 | Utility Dive / Caterpillar |
Sources
[1] OsakaWire Intelligence: "The Grid Is the Bottleneck — 2,061 GW Stuck in Queues," Aug 2026. Comprehensive deep-dive intelligence report tying interconnection queues → permitting → transformer supply chain → copper/raw materials → workforce. Primary aggregation of Berkeley Lab, IEA, FERC data.
[2] Same source — sections 02 & 07 cover EU Grids Package, tacit approval mechanism, and Connecting Europe Facility funding increases.
[3] Foley & Lardner LLP / National Law Review (Scott D. Ellis): "The Grid's Bottleneck Isn't Just Money. It's Materials." Sep 2026. Authoritative analysis of utility capex vs. procurement capacity gap. Primary source for transformer pricing indices, GOES/Cleveland-Cliffs dependency, Terrapin Construction procurement data, manufacturer investment commitments (Eaton $340M SC, Siemens $150M NC, Hitachi $1B+ VA).
[4] Latitude Media: "ERCOT's Large Load Queue Has Nearly Quadrupled in a Single Year," Dec 2025. Primary source for the ERCOT queue expansion snapshot (63 GW → 226 GW).
[4a] ERCOTqueue (via our supply-intel dataset, entry GRID-ERCOT-LL-QUEUE-466GW-JUNE-2026): ERCOT large-load interconnection queue at ~466 GW of requested capacity by June 2026, ~90% data centers.
[4b] BNN Bloomberg / Reuters mirror (via our supply-intel dataset, entry GRID-ERCOT-LARGE-LOAD-FREEZE-2026), Sep 2026: Texas freezes new large-load grid connections; large-load requests up from ~48 GW (2023) to >474 GW per ERCOT documents and Governor Abbott.
[5] Motley Fool: Vertiv Q2 2026 earnings call transcript. Guidance raise, deferred revenue figures, net leverage position.
[6] Motley Fool: Eaton Q2 2026 earnings call transcript. Data center segment growth, backlog figures, Boyd Thermal acquisition.
[7] S&P Global: "Data center grid-power demand to rise 22% in 2025, nearly triple by 2030," Oct 2024. Original source for demand trajectory and $720B grid upgrade estimate.
[8] FERC: Interconnection reform explainer. Order No. 2023 details and implementation status.
[9] Utility Dive: Caterpillar Q2 2026 results coverage, Aug 2026.
[10] OsakaWire section 04: "Copper, Steel and Skilled Hands" — reconductoring analysis and advanced conductor potential assessment.
[11] OsakaWire section 05: "Two Years in China, a Decade in the West" — China UHV investment commitment and construction timeline data.
The additional source registry for this edition is maintained in the evidence ledger at data/platform-data.json; methodology and as-of conventions: methodology.