AI electricity demand is expected to surge threefold within five years, but the US power grid is still "not ready", and natural gas is "taking the lead".
```
Artificial intelligence is pushing the U.S. power grid to its design limits.
According to Goldman Sachs Research, global data center electricity demand is expected to reach 84 GW by 2027—a 50% surge from 2023—with AI loads accounting for 27% of that. In the U.S. alone, utilities' forecasts for peak summer demand over the next five years have jumped from 38 GW to 128 GW, more than tripling in a single planning cycle. The implication: the U.S. grid is already near its carrying capacity, even before the AI wave fully arrives.
Facing this gap, natural gas is currently the only option that can meet demand in both speed and scale. Entergy is investing $3.2 billion to build three natural gas power plants totaling 2.3 GW, dedicated to powering Meta’s new data center in Louisiana. NextEra Energy, the largest renewable energy developer in the U.S., is working with ExxonMobil to build a new 1.2 GW natural gas plant in the Southeast. NextEra CEO John Ketchum describes this trend as the AI industry turning to "BYOG"—build your own generation.
However, the cost of massively betting on natural gas is becoming apparent. The clearing price for PJM’s capacity market (which covers much of the U.S. Mid-Atlantic and Midwest) for the 2026-2027 delivery year has jumped to $329/MW, up more than tenfold from $28.92/MW two years ago. The capital cost for new combined-cycle gas turbines has also nearly doubled, to around $2,000/kW. At the same time, these plants—designed to operate for about 30 years—will profoundly affect the U.S.’s carbon reduction commitments.
Natural Gas Becomes the Only Practical Option
Currently, about 40% of U.S. electricity comes from natural gas, with renewables and coal making up the rest. AI data centers require stable, GW-scale, 24/7 power—a weak point for the current grid.
Renewables face hard constraints: the median wait time for new solar and wind projects to connect to the grid exceeds four years. By contrast, natural gas is cheap, abundant, and supported by a nationwide pipeline network, allowing new plants to come online within three to five years.
The three plants Entergy is building for Meta’s Louisiana data center will require 2 GW of power just for the computing load. The NextEra-ExxonMobil joint plant is based on a “build infrastructure first, then attract tenants” logic. John Ketchum’s “BYOG” comment marks a shift in the relationship between energy companies and tech giants—from “supplier and customer” to “co-builders of infrastructure.”
Systemic Grid Engineering Overhaul Ahead
Traditional grid engineering relies on predictable loads—seasonal peaks, industrial cycles, and population growth can all be modeled years in advance. AI has completely broken this logic.
Training large language models means thousands of GPUs running at high loads for days or even weeks, and then suddenly shutting off; every AI inference can trigger instantaneous power spikes in the hundreds of megawatts (“inference spike”). This load behavior is unprecedented, directly impacting utilities’ dispatch curves and reserve capacity models.
To address this, gas peaker plants—designed for short bursts of high output—are now being deployed right next to data center campuses, absorbing inference pulses that baseload plants cannot match in response time.
Meanwhile, Texas grid operator ERCOT has developed an all-new “Adjusted Large Load Forecast” methodology to distinguish between real demand and “phantom demand” created by speculative grid connection applications.
Transmission Bottlenecks and Cost Surge
The physical grid is also under strain. Transmission investment in much of the U.S. has continually declined since 2015, leaving the system running close to its limits even before the AI demand boom. DOE’s National Transmission Needs Study has found transmission congestion to be severe in many U.S. regions even prior to this surge in demand.
Demand-side data is equally alarming: Texas’s CenterPoint Energy reports that, between the end of 2023 and the end of 2024, large-load grid connection applications have increased by 700%; in Virginia, there are still up to 50 GW of data center projects waiting in the interconnection queue.
Market prices reflect this structural tension: PJM capacity prices have soared from $28.92/MW to $329/MW over two years; the capital cost for new combined-cycle gas turbines has almost doubled to about $2,000/kW. These costs will eventually be passed through to downstream customers in the form of higher electricity prices.
Natural Gas Lock-in Effect and Carbon Emissions Risk
The natural gas plants currently being approved and built are not just transitional infrastructure—they have an average lifespan of about 30 years, meaning they will operate past nearly all major net-zero targets.
The full lifecycle carbon emissions from natural gas are about 490g CO₂/kWh. In the U.S. South, power companies plan to add about 20 GW of new gas capacity over the next 15 years. In Virginia, South Carolina, and Georgia, 65% to 85% of the projected load growth comes from data centers.
Methane leakage further worsens the emissions problem: natural gas infrastructure (drilling, pipelines, compression) continues to leak methane, which over 20 years has about 80 times the warming impact of CO₂. This issue is evolving into a policy fault line between energy companies, large tech firms that have pledged net-zero, and regulators who have yet to clarify the climate impact of AI’s energy use.
Policy and Alternative Paths: Distant Waters Can’t Quench Immediate Thirst
Several structural policy tools are underway, but none can solve the current supply-demand shortfall.
On storage, the 2022 Inflation Reduction Act (IRA) provides a 30% tax credit for standalone storage systems coming online after 2024, covering the system itself and offering financial incentives for operators and utilities to invest in battery systems.
On nuclear, as a zero-carbon, stable baseload option, it is getting more attention. Google has reached an agreement with NextEra to restart the 615 MW Duane Arnold nuclear plant for 24/7 carbon-free power.
Transmission upgrades remain the hardest challenge. DOE studies show significant gaps in transmission capacity in almost every U.S. region, and closing these gaps will require years of coordinated investment and regulatory reform.
Currently, the growth in AI-driven demand has far outpaced the underlying infrastructure buildout—natural gas plants, transmission upgrades, storage deployments, and nuclear restarts are all lagging behind. How to bridge this gap will depend on whether policy coordination and proactive investment can come before the costlier consequences of power shortages, delayed data center construction, and rising electricity rates set in.
Risk Notice and DisclaimerThe market involves risk, so investment should be cautious. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article fit their particular situation. Investments based on this are at your own risk. ```