Finding
Electrification now waits on electrons: interconnection and transformers set the schedule for gigafactories, fabs and AI campuses
The heaviest new loads in the US are battery and EV gigafactories, semiconductor fabs and AI data centers, each drawing 50 MW to more than 1 GW at one site. The median US generation project completed in 2025 took 61 months from interconnection request to operation, 2,061 GW sat in interconnection queues at the end of 2025, and large power transformers take about 128 weeks to deliver.
Data centers are the fastest-growing load. Lawrence Berkeley National Laboratory put them at 4.4% of US electricity in 2023 and projects 6.7-12% by 2028. In Texas, large-load requests to ERCOT have outrun the generation and transmission built to serve them (ERCOT large-load interconnection, AustinIO). Builders respond by bringing power with them: behind-the-meter gas turbines and engines, fuel cells and battery storage organized as microgrids.
China plans transmission ahead of load: State Grid and China Southern Power Grid build ultra-high-voltage lines on multi-year plans, and State Grid's annual grid investment has run above 600 billion yuan. That is one reason Chinese gigafactories cite cost, not grid access, as the constraint. In the US the gating items are physical - transformers, switchgear, gas turbines - and procedural: interconnection studies and transmission permits.
Loads
The demand triad
Typical campus loads; individual sites vary widely.
| Sector | Typical campus load | Outage sensitivity | Examples |
|---|---|---|---|
| EV and battery gigafactories | 50-200 MW | High: production stops, formation and dry rooms must restart | Tesla Giga Texas, Panasonic De Soto (Kansas), CATL Ningde, BYD Hefei |
| Semiconductor fabs | 50-300+ MW | Very high: wafer scrap, tool damage, days to requalify | Samsung Taylor, TI Sherman, TSMC Arizona, SMIC Shanghai |
| AI data centers | 100 MW to 1 GW+ | High to very high: stranded compute, interrupted training runs | Stargate Abilene, xAI Colossus (Memphis), Alibaba and ByteDance campuses |
Metrics
Metrics that decide a power plan
| Metric | Definition | Why it matters |
|---|---|---|
| SAIDI / SAIFI | Average outage duration and frequency on the utility feeder | Sets how much on-site resilience is needed: UPS, BESS, gensets |
| Time to power | Months from interconnection request to energization | Often the longest item on a gigafactory or data center schedule |
| LCOE | Levelized cost of energy: lifetime cost per MWh | Compares solar, wind, gas, nuclear and storage on one basis |
| Scope 1, 2, 3 emissions | Direct, purchased-electricity and supply chain emissions | Drives PPA strategy, compliance and customer requirements |
Bottlenecks
What slows energization
| Bottleneck | Impact | Outlook |
|---|---|---|
| Large power transformers | ~128 weeks; step-up units ~144 weeks; US imports about 80% of large units | New US plants from 2027; Wood Mackenzie sees the shortfall falling from 30% (2025) to about 5% by 2030 |
| Grain-oriented electrical steel | About 10 mills worldwide; Cleveland-Cliffs is the only US producer | Baowu, Nippon Steel and POSCO lead capacity; amorphous cores are a partial substitute |
| Gas turbines | GE Vernova, Siemens Energy and Mitsubishi Power order books are full for years | Reciprocating engines, fuel cells and batteries fill the gap |
| Interconnection and permitting | Queue studies and transmission approvals take years | FERC Order 2023 cluster studies; pre-approved corridors |
| Skilled labor | HV electricians and lineworkers limit commissioning | Prefabricated skids, factory-tested blocks, training pipelines |
Sources
Sources and scope
- Scope
- Energy Foundation
- Sources
- LBNL (2024 data center report; interconnection queue data); Wood Mackenzie transformer outlook (2025); EIA
- Updated
- October 2026