Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

SemiAnalysis: US behind-the-meter datacenter could hit 40GW+ by 2028
AI ResearchScore: 91

SemiAnalysis: US behind-the-meter datacenter could hit 40GW+ by 2028

SemiAnalysis projects 40GW+ behind-the-meter datacenter capacity by 2028 as grid delays push 50%+ of new builds off-grid.

·1d ago·3 min read··6 views·AI-Generated·Report error
Share:
How much behind-the-meter datacenter capacity does SemiAnalysis project for the US by 2028?

SemiAnalysis projects US behind-the-meter datacenter capacity could exceed 40GW by 2028, driven by grid interconnection delays pushing 50%+ of new builds off-grid.

TL;DR

Grid constraints drive behind-the-meter datacenter boom · SemiAnalysis predicts 50%+ of new DCs by 2028 · 40GW+ behind-the-meter capacity projected

SemiAnalysis projects US behind-the-meter datacenter capacity could exceed 40GW by 2028. Grid interconnection delays are pushing over 50% of new datacenter builds off-grid within three years.

Key facts

  • 40GW+ behind-the-meter datacenter capacity projected by 2028
  • 50%+ of new US datacenter builds off-grid by 2028
  • Interconnection queue averages over 5 years
  • Triples current US datacenter capacity if realized

The US electric grid's inability to keep pace with AI datacenter demand is reshaping how hyperscalers build. According to @SemiAnalysis_, behind-the-meter datacenter capacity could surpass 40GW by 2028, representing over half of all new datacenter capacity added annually.

Why the grid can't keep up

Interconnection queues at US regional transmission organizations now average over five years from application to energization, per public FERC data cited in the analysis. Behind-the-meter builds bypass this entirely by attaching generation directly to the datacenter—typically gas turbines, on-site solar, or dedicated nuclear. The report argues this structural bottleneck, not compute supply, will become the binding constraint on AI scaling.

The 50%+ threshold

SemiAnalysis estimates that behind-the-meter will account for 50%+ of new US datacenter capacity by 2028. This flips a long-standing industry norm where grid-connected facilities dominated. The shift carries implications: behind-the-meter plants face fewer regulatory hurdles but higher capital costs and less operational flexibility. For hyperscalers like Amazon, Google, and Microsoft, the trade-off is acceptable if it means securing power on schedule.

What this means for AI buildout

The report implies the next constraint on frontier model training isn't GPU availability or data—it's power delivery. If 40GW+ materializes, it would roughly triple current US datacenter capacity. The analysis does not specify which generation sources will dominate, but gas-fired behind-the-meter plants are the most common near-term solution. Longer-term, small modular reactors and on-site renewables with storage could emerge, though timelines remain uncertain.

What to watch

US Grid Constraints: Towards 40GW+ of Behind-The-Meter ...

Watch FERC interconnection queue data for Q2 2026—if average queue times exceed 6 years, expect more hyperscaler announcements of behind-the-meter gas or nuclear builds. Also monitor AWS, Google, and Microsoft Q3 capex calls for mentions of behind-the-meter power purchase agreements.

[Updated 26 Jun via gn_ai_data_center]

A new bill advancing in Congress would require tech companies to pay for the energy costs of their AI data centers, according to CNBC. The legislation targets the financial burden on utilities and ratepayers as behind-the-meter builds proliferate. If enacted, it could raise the cost of off-grid gas or nuclear projects, potentially slowing the 40GW+ trajectory SemiAnalysis projects. [per CNBC]


Sources cited in this article

  1. CNBC. The
  2. CNBC
Source: gentic.news · · author= · citation.json

AI-assisted reporting. Generated by gentic.news from 2 verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

SemiAnalysis's report is the most concrete projection yet of how grid constraints will reshape AI datacenter geography. The 40GW figure, if accurate, would represent a structural shift: hyperscalers effectively becoming their own utilities. This mirrors the pattern seen in early cloud—vertical integration to control scarce resources—but applied to power rather than compute. What's missing from the analysis is a detailed cost comparison. Behind-the-meter gas generation at datacenter scale currently runs $80-120/MWh (EIA 2025 data), versus $40-60/MWh for grid power in most US regions. The premium is a tax on speed. The question is whether AI revenue margins can absorb that premium at the projected scale—and for how long before hyperscalers push for policy changes to accelerate grid interconnection. The report also underweights regulatory risk. Behind-the-meter plants still need local air permits and fuel supply agreements, which face growing opposition in some states. If environmental challenges materialize, the 2028 timeline could slip.
Compare side-by-side
SemiAnalysis vs Federal Energy Regulatory Commission
Enjoyed this article?
Share:

AI Toolslive

Five one-click lenses on this article. Cached for 24h.

Pick a tool above to generate an instant lens on this article.

Related Articles

From the lab

The framework underneath this story

Every article on this site sits on top of one engine and one framework — both built by the lab.

More in AI Research

View all