This is a reduced view. Bottleneck assessments, priority scores, and segment-level constraint counts are from the full analysis. Not included here: sourced evidence citations per node, bottleneck reasoning and supply-demand quantification, business impact assessments, vendor concentration data, three-scenario banding detail, full 34-node priority ranking. Cross-segment constraint network uses modeled connections derived from node-level analysis; precise vendor mapping is not shown.
Context
The constraint on AI power is not generation. It is connection.
The industry conversation centers on whether enough power plants will get built to serve AI datacenters. That is the wrong question. Generation technology is mostly available. Gas turbines ship. Solar panels ship. The binding constraints sit in the physical infrastructure required to connect, stabilize, and protect that power: transformers, reactive power compensation equipment, grid interconnection queue throughput, and the simulation tools that govern who connects and when.
Transformers dominate the chain. Eleven of the 34 constrained nodes are transformer variants, spanning 8 of 15 segments. Extra-high-voltage power transformers (230-345 kV, 200+ MVA) score 9.9 out of 10 on priority: the highest in the entire chain. Lead times for large power transformers now exceed 100 weeks. Every campus interconnection, every BESS tie-in, every generator step-up requires a transformer that may not exist in time.
Grid queue throughput is the second systemic chokepoint. PJM's interconnection queue depends on a single software platform and a finite pool of simulation licenses. Study capacity sits at 17-26 GW per year against 36+ GW of demand. Generation that has been financed and permitted cannot connect.
Reactive power scarcity blocks gigawatt-scale connections. STATCOMs and synchronous condensers required for voltage support at new points of interconnection are in severe shortage. Three of the top 10 priority nodes are reactive power systems.
Segment Map // 15 Chain Layers
Severe
Bottleneck
Equilibrium
Excess Capacity
Structural Chokepoints
Five constraint clusters that bind AI datacenter power delivery
High-Voltage Transformer Supply Chain8 Segments // 11 Nodes
Transformers are the single most pervasive constraint in the entire value chain. They appear in 8 of 15 segments, from load forecasting through decommissioning. The shortage is not one product; it is an entire class of custom-manufactured equipment with 80-120+ week lead times, concentrated among fewer than ten global OEMs.
230-345 kV, 200+ MVA EHV power transformers for grid interconnectionSEVERE9.9
MV/HV step-up transformers for BESS interconnection (34.5 kV to 69-138 kV)SEVERE9.1
345-500 kV, 300+ MVA autotransformers for campus POI (N+1)SEVERE8.9
500/230 kV and 765/345 kV autotransformer banks for bulk substationsBOTTLENECK8.8
EHV condenser bushings (RIP/OIP) rated 230-345 kV for GSU terminalsBOTTLENECK8.7
Generator step-up transformers (50-300 MVA) for reciprocating engine plantsSEVERE8.65
Grid Interconnection Queue and Simulation Throughput4 Segments // 7 Nodes
PJM's interconnection queue depends on a single software platform (QueuePoint/NextGen) and a finite pool of PSS/E and PSLF simulation licenses. ERCOT requires vendor-validated EMT models that most inverter-based resource vendors have not produced. Financed, permitted generation capacity cannot connect to the grid.
Vendor-validated EMT models (PSCAD/EMTDC) for IBR in datacenter microgridsBOTTLENECK9.15
ISO-accepted PSS/E and PSLF simulation suites for ERCOT LLIS (75+ MW)BOTTLENECK8.95
PJM QueuePoint/NextGen interconnection workflow moduleSEVERE8.9
PJM-class AC deliverability and contingency screening engine (TARA)SEVERE7.5
Reactive Power and Grid Stability Systems2 Segments // 3 Nodes
Gigawatt-scale datacenter clusters depress local voltage and reduce grid inertia. The corrective equipment (STATCOMs and synchronous condensers) is in severe shortage. Without reactive power compensation at the point of interconnection, transmission operators will not approve energization.
Large synchronous condensers (200-600 Mvar) for grid-tie voltage support and inertiaSEVERE9.35
230-500 kV STATCOMs (150-400 Mvar) for dynamic voltage support at interconnection nodesBOTTLENECK9.35
200+ Mvar modular-multilevel STATCOM systems (MMC-IGBT) for 230-500 kV POIsSEVERE9.1
Nuclear Fuel Supply (HALEU and TRISO)2 Segments // 3 Nodes
The small modular reactor pathway for datacenter power depends entirely on high-assay low-enriched uranium fuel. Non-Russian enrichment capacity does not exist at commercial scale. TRISO-coated particle fuel fabrication has one qualified production line globally. The SMR-for-datacenters narrative outpaces fuel supply reality by 3-5 years.
Non-Russian HALEU fuel at 15.5-19.75% U-235 (enrichment + deconversion)SEVERE9.3
TRISO-coated particle fuel fabrication (UCO/UN kernels with SiC coatings)SEVERE9.3
19.75% HALEU UF6 enrichment and deconversion feed (U.S., non-Russian)SEVERE8.7
Gas Infrastructure and Fast-Start Generation3 Segments // 5 Nodes
Gas-fired generation is the practical near-term pathway. But the gas supply chain has its own chokepoints: firm pipeline transport into Northern Virginia is at capacity, heavy-duty turbine packages have multi-year lead times, and the secondary market for repowered aeroderivative turbines is supply-constrained.
LM6000-class CF6-80C2 core overhaul and repackaging into 48 MW aeroderivative gensetsSEVERE9.0
Firm pipeline no-notice gas transport + storage deliverability into Transco Zone 5SEVERE8.95
HA/J-class heavy-duty gas turbine packages (7HA.02, SGT6-9000HL, M501JAC)SEVERE8.15
Highest Priority Bottlenecks // All Segments
Showing top 10 of 34 constrained nodes across 15 segments. Full priority ranking not included.
Methodology. Nodes assessed for supply adequacy against projected demand using sourced evidence (regulatory filings, vendor disclosures, OEM data, grid operator publications). Bottleneck status reflects structural supply-demand imbalance, not temporary disruption. Priority scores weight severity, systemic reach, and replaceability. Three-scenario banding (Low/Base/High) applied to each node. Assessment window: December 2023 through November 2025. Current as of December 2025.