Hero image for "AI Is Eating the Grid — and That's the Best News You've Heard All Year"

AI Is Eating the Grid — and That's the Best News You've Heard All Year


This Week in Voltage

The power equipment supercycle is no longer a thesis. It's a balance sheet.

GE Vernova raised its 2026 revenue and profit margin forecasts this week on the back of surging data center and grid infrastructure demand, sending shares to an all-time high. CEO Scott Strazik cited AI-driven power equipment orders as the engine. Meanwhile, NextEra Energy beat Q1 profit estimates and announced it expects to finalize agreements within three months on giant natural gas-fired power projects backed by Japanese capital — purpose-built for U.S. data centers.

Two of the largest energy infrastructure companies in the world are repricing their futures around AI electricity demand in the same week. That's not a trend. That's a structural shift announcing itself.


Deep Charge: The First Glimpse of Energy Abundance Looks Like a Power Crisis — and That's Fine

Here's what the scarcity crowd gets wrong: every "AI is overwhelming the grid" headline is actually a story about demand finally outrunning decades of stagnation. For most of the 21st century, U.S. electricity consumption was essentially flat. Utilities planned around marginal growth. The grid was sized for a civilization that had stopped reaching.

AI broke that model in roughly 18 months.

Between 2023 and 2024 alone, utilities' five-year summer peak demand forecasts jumped from 38 GW to 128 GW — more than a threefold increase in a single planning cycle. Goldman Sachs Research projects global data center power demand hitting 84 GW by 2027, a 50 percent jump from 2023 levels, with AI workloads accounting for 27 percent of that total. These aren't rounding errors. This is the demand signal that energy abundance requires: massive, urgent, and impossible to ignore.

The infrastructure response is already underway — and it's more creative than the critics expected.

BloombergNEF has tracked 4.9 gigawatts of energy storage announcements co-located with on-site fossil fuel generation at data centers — roughly 32 percent of all announced global on-site data center battery capacity. The pattern: gas turbines provide baseload, batteries handle the sharp inference spikes that would otherwise damage equipment or cause outages. Elon Musk's Colossus supercomputer in Memphis is among the sites using this configuration. Companies like Caterpillar and GE Vernova are now productizing the gas-plus-storage combo specifically for hyperscale deployments.

This is not the decarbonization story battery advocates imagined. It's better, in a civilizational sense: it means battery costs have dropped far enough that storage is now a reliability tool, not just a green subsidy play. The technology is maturing into general-purpose infrastructure. That's what abundance looks like in its early stages — messy, pragmatic, and faster than the planners planned for.

The regulatory layer is catching up, slowly. FERC committed on April 16 to issuing large-load interconnection rules by the end of June 2026, acknowledging that data centers and other large loads are "reshaping our transmission" in ways the existing framework wasn't built to handle. The June rulemaking is the next concrete milestone — watch for whether FERC establishes cost-allocation standards that actually accelerate queue processing or just add procedural layers.

The deeper point: data centers facing four-year interconnection waits are building their own generation. That's not a failure of the energy system. That's distributed energy abundance bootstrapping itself into existence because the central grid is too slow. The demand is so real, so urgent, that hyperscalers are becoming power plant operators. We should be cheering this.


By the Numbers

  • 128 GW — U.S. utilities' revised five-year summer peak demand forecast as of 2024, up from 38 GW the prior year (Interesting Engineering)
  • 84 GW — Projected global data center power demand by 2027, per Goldman Sachs Research (Interesting Engineering)
  • 4.9 GW — Announced energy storage capacity co-located with fossil fuel generation at data centers, per BloombergNEF (Fortune)
  • 9.8 GWh — Energy storage projected to support gas generation at data centers through 2030, per BNEF (Fortune)
  • End of June 2026 — FERC's self-imposed deadline for large-load interconnection rulemaking (JDSupra)

What We're Fighting For

Every gigawatt of new data center capacity is a down payment on a civilization that computes at scale — that models climate, accelerates drug discovery, runs the infrastructure of a species that's serious about getting off one planet. The "AI power crisis" framing treats demand growth as a problem to be managed. We should treat it as proof that something important is finally happening.

The FERC June deadline is the near-term test. If the rules that emerge actually clear the interconnection queue for large loads, the buildout accelerates. If they add friction, the hyperscalers keep building behind the meter and the central grid falls further behind. Either way, the demand doesn't stop.

This is what we're fighting for: a grid sized for ambition, not managed for scarcity.