AI Data Centers Are Driving a U.S. Gas Buildout — A 6-Step Playbook for Leaders

How AI Data Centers Are Rewriting the U.S. Gas Map — and What Leaders Should Do

Executive summary: Rapid AI-driven expansion of data centers has triggered a wave of proposed gas-fired power projects across the U.S., creating operational convenience but meaningful climate and reputational risk. The proposals are large—if everything were built the gas fleet could grow by nearly half—but many projects are speculative and face supply-chain, permitting, and finance constraints.

What the numbers say

Global Energy Monitor (GEM) tracked a dramatic spike in gas projects explicitly tied to data centers: proposals rose roughly 25× year-over-year. GEM recorded about 85 gigawatts (GW) in the U.S. gas pipeline in early 2024, with roughly 4 GW explicitly linked to data centers; by 2025 more than 97 GW of tracked demand had ties to data-center projects. For scale, the U.S. currently operates roughly 565 GW of gas-fired capacity — the development pipeline could add as much as ~252 GW, nearly a 50% increase if every proposal were built.

Developers are already building nearly 30 GW, while about 159 GW remain in preconstruction. But GEM and other analysts warn these headline totals are an upper bound: many proposals are exploratory, some lack confirmed turbine suppliers, and several projects are effectively being shopped to multiple utilities or listed to preserve options.

“About 18 months ago we began seeing a surge in proposals specifically for data centers; many are still speculative,” said Jenny Martos, a research analyst at Global Energy Monitor. “We’re in an explosion-of-proposals phase—what actually gets built remains uncertain.”

Why AI data centers drive demand for gas-fired power

Large AI workloads—training massive models and serving latency-sensitive applications—concentrate power demand in specific locations and at predictable rates. Hyperscale cloud services and GPU-heavy clusters need continuous, reliable electricity. When grid upgrades or transmission interconnections are slow, data-center developers often turn to on-site gas turbines, combined heat and power (CHP), or purpose-built gas plants for speed and resilience.

Definitions for quick reference: 1 GW ≈ enough power for ~0.7–1 million homes depending on region; PPA = power-purchase agreement (a contract to buy power); CHP = combined heat and power (on-site generation that captures waste heat).

Climate and regulatory stakes

The climate implications are material. Natural gas accounted for about 35% of U.S. energy-related CO2 emissions in 2022, according to the U.S. Energy Information Administration (EIA). Burning gas emits less CO2 than coal per unit of energy, but large-scale additional gas consumption still raises CO2 totals substantially. Complicating the comparison further is methane — unburned natural gas that leaks across production, transport, and distribution.

Methane is far more potent than CO2 over short timeframes: on a 20-year horizon its global warming potential is roughly 80 times that of CO2 according to major assessments. Oil and gas systems are estimated to cause a large share of global methane leaks, and expanding fossil infrastructure increases near‑term warming risks unless leakage is aggressively controlled.

“The scale of the proposed build-out has major implications for emissions and climate outcomes,” said Jonathan Banks, senior climate adviser at the Clean Air Task Force. “While burning gas emits less CO2 than coal, large‑scale gas use still means substantial CO2; methane leaks narrow the climate advantage.” He added that AI demand is persistent and the central challenge is reducing the environmental footprint of these facilities.

Policy matters. Recent federal actions have eased or delayed certain power-plant and methane regulations while encouraging data-center growth. In some regions, coal retirements have been postponed to preserve capacity. That alignment—rapid demand growth plus looser oversight—risks locking in elevated emissions profiles unless countermeasures appear.

Supply-side constraints and why the 252 GW figure is not destiny

Turbine shortages: Two-thirds of GEM-tracked projects worldwide lack a confirmed turbine manufacturer. Major turbine suppliers have long lead times; procurement bottlenecks can delay or cancel projects.
Financing and permitting: Securing capital and siting approvals remains a major hurdle. Municipal and state permitting timelines vary, and community opposition can stall projects.
Grid interconnection and transmission limits: Building transmission is slow and costly; developers often choose on-site generation for speed, but that creates the carbon-lock risk.
Operational efficiency and demand-side changes: AI model efficiency improvements, workload optimization, and better server utilization can blunt projected demand growth.

These constraints mean executives should treat the current pipeline as a signal of intent and risk, not as a locked-in outcome.

Scenario sketch: emissions outcomes

Optimistic: Many speculative projects stall; hyperscalers expand renewables and storage; methane rules strengthen. Result: modest increase in gas use, limited near-term warming impact.
Most likely: Some gas plants are built to meet immediate needs; utilities and data centers use a mix of PPAs, on-site renewables, and gas firming. Result: elevated CO2 and methane emissions but with growing corporate mitigation efforts.
Pessimistic: Large portion of the pipeline is built, methane regulation remains weak, and coal stays online to meet peak needs. Result: material increase in near‑term warming from combined CO2 and methane releases.

A pragmatic six-step playbook for business leaders

Quantify exposure. Ask your energy team to map current and projected data-center power needs, identify which sites are at risk of relying on new gas-fired capacity, and quantify potential CO2 and methane exposure.
Prioritize clean procurement. Target renewable PPAs, corporate renewable tariffs, and on-site solar + storage where economics allow. Metric: % of data-center load under renewable PPA or on-site clean generation.
Insist on methane clauses. Require suppliers to meet strict methane-detection and repair standards, and include independent verification. Metric: methane leakage rate thresholds and third-party monitoring.
Design for efficiency. Invest in model-level efficiency, hardware refreshes that increase compute per watt, workload scheduling, and free‑cooling where possible. Metric: kWh per inference/training step or utilization improvements.
Favor flexible grid solutions. Prioritize grid upgrades, demand-response, and seasonal storage over dedicated fossil capacity; include sunset clauses or repowering commitments if gas plants are used.
Engage policy and stakeholders. Support sensible methane regulation, transparent emissions accounting, and community-benefit agreements to anticipate regulatory and reputational pressure.

Case vignette: choosing between speed and carbon risk

A cloud provider faced a choice: commission a nearby gas-fired plant to ensure immediate capacity for a new AI training hub, or delay launch by 18 months to complete a transmission upgrade and procure a long-term renewable PPA. The gas option promised faster time-to-market but would likely operate for decades; the company chose a hybrid: short-term leased capacity paired with a binding plan to reach 100% renewable procurement for that site within five years, along with methane-performance clauses for any interim fuel. The blended approach reduced immediate risk while avoiding permanent carbon lock-in.

What boards and energy teams should ask today

How much of our projected AI/data-center load is exposed to new gas-fired capacity?
Have your energy team map prospective suppliers and plants tied to your projects.
What are our procurement targets for renewables and storage, and can they be accelerated?
Set measurable goals and timelines (e.g., % under PPA within X months).
Do our fuel contracts include independent methane monitoring and remediation clauses?
Specify performance standards and penalties for non-compliance.
Have we stress-tested regulatory scenarios (stricter methane rules, carbon pricing) against our capital plans?
Quantify stranded-asset risk and contingency options.

Methodology & caveats

GEM’s tracker compiles project proposals, announcements, and filings; its totals include many preconstruction and speculative projects. The ~252 GW addition is an upper-bound aggregation of proposals across stages. Two key caveats: many projects lack confirmed turbine suppliers or financing, and data-center demand is often shopped to multiple utilities, which can double-count intent. Treat GEM’s numbers as a risk signal rather than a firm forecast. For more, see the Global Energy Monitor tracker and EIA emissions data at EIA.

Quick takeaways for executives

AI data centers are a major new driver of proposals for gas-fired power — the pipeline is large but uncertain.
Climate risk comes from both additional CO2 and potent methane leakage; regulatory choices will influence outcomes.
Supply-chain constraints (notably turbine lead times) and financing/permitting hurdles mean the headline pipeline is not destiny.
Act now with procurement, methane controls, efficiency, and grid-first strategies to avoid locking in emissions or stranded assets.