Gas, Grids, and AI: The Infrastructure Race for the Next Computing Paradigm

The infrastructure race for the next computing paradigm is being defined by an exponential demand for electricity. Data centers, the physical engines of artificial intelligence, are projected to consume over 1,000 terawatt-hours (TWh) of electricity by 2030. That represents more than a doubling from the 460 TWh consumed in 2024. This isn't just a growth story; it's a fundamental shift in the energy landscape, creating a long-term, baseload power requirement that will strain existing grids.

In the United States, the scale of this shift is stark. Today, data centers account for roughly 2% of total electricity use. By 2030, that figure could climb to as much as 9%. This surge is driven by the sheer energy intensity of AI workloads, which require thousands of servers running and cooling around the clock. The strain is already visible, with centers pulling power from the grid and contributing to peak demand issues. Meeting this demand will require an abundant supply of reliable, 24/7 energy, making the power infrastructure layer itself a critical battleground.

The strategic response is beginning to take shape. The Trump administration and a bipartisan group of governors have proposed a radical policy move: a one-time $15 billion emergency auction through the PJM grid operator. This would allow data center owners to bid on 15-year power purchase agreements, effectively pre-paying for new generation built specifically to serve them. While the proposal is still under review and faces regulatory hurdles, it signals a clear recognition of the bottleneck. The goal is to accelerate the build-out of new power plants, with natural gas expected to play a growing role in the near term. This policy push underscores that the AI paradigm shift is creating an unprecedented, long-term demand for reliable baseload power, making the energy infrastructure layer a critical near-term play.

The Gas Infrastructure Play: Catalysts and Constraints

The regulatory landscape is shifting to clear the path for natural gas to fuel the AI power surge. Two key actions are removing long-standing bottlenecks. First, the Federal Energy Regulatory Commission (FERC) rolled back Order 871 in April 2025. This rule had mandated a 30-day delay for gas project construction while rehearing requests were pending, even if no opposition existed. Its removal eliminates a major source of scheduling uncertainty and added cost, directly accelerating the timeline for building new power plants.

Second, the Environmental Protection Agency (EPA) is poised to introduce permitting reforms that would allow developers to begin construction on ancillary site structures-like concrete pads and external walls-before securing final air permits. This "flexibility" could shave months off project start dates, a critical advantage when building power plants to meet AI demand. These moves, part of a broader push from the Trump administration to streamline energy infrastructure, create a powerful catalyst for gas-fired generation.

Yet the most immediate constraint is a looming reliability crisis in the very grid that needs this new power. The Mid-Atlantic region, governed by PJM Interconnection, faces a severe capacity crunch. From 2020 to 2025, the region forced nearly 17 gigawatts of reliable baseload power offline. More critically, approximately 40 GW of firm power capacity is at risk of retirement by 2030. This creates a dangerous gap between projected AI demand and the grid's ability to deliver it.

The proposed $15 billion emergency auction aims to bridge this gap by pre-paying for new generation. But the success of this plan hinges on the very regulatory reforms now taking effect. Without them, the accelerated build-out of gas plants cannot happen fast enough to prevent the reliability issues and price spikes that have plagued the region. The gas infrastructure play is therefore a race against time, where policy catalysts must outpace a structural grid deficit.

The Build-Out Gap: Policy Acceleration vs. Physical Reality

The policy push to accelerate power generation is racing against a fundamental physical reality: building a gas plant takes years. The proposed PJM emergency auction, while a bold signal, creates a significant capital commitment that must be paid upfront. Under the plan, data center owners would be required to pay for the new generation built for them, whether they use the power or not. This "take-or-pay" structure is designed to de-risk investment and guarantee funding for new plants, but it also locks in massive, long-term liabilities for developers. The auction is not an immediate fix; Capstone analysts suggest a six to 12 month timeline before an auction could be held, at the earliest.

Even after an auction, the clock starts ticking on a multi-year build-out. The demand for gas turbines has surged to unprecedented levels, with lead times now stretching to 7-8 years. This lag creates a dangerous gap between the policy acceleration and the actual delivery of power. While regulators are removing permitting hurdles, the supply chain for the core equipment is still catching up. As one industry leader noted, "Frankly, we can't make enough gas turbines to support this market". The result is a market where the first-mover advantage is not just strategic-it is existential.

This sets the stage for a winner-take-most dynamic. Developers who can secure pipeline access and navigate the new regulatory landscape quickly will have a decisive edge. Proximity to natural gas pipelines is no longer a nice-to-have; it is a critical site selection factor. As one guide notes, developers should look to put new data centers along major pipeline corridors, where they can "drill a hole and interconnect". Those who move early can lock in the necessary infrastructure and avoid the escalating costs and delays that will plague later entrants. The policy catalysts are clear, but the physical build-out is a long S-curve. The companies that bridge that gap first will own the infrastructure layer for the AI paradigm.

Catalysts, Scenarios, and What to Watch

The gas infrastructure thesis now hinges on a series of near-term catalysts that will validate or break the investment case. The immediate test is the outcome of the proposed PJM emergency auction. This one-time mechanism is the linchpin for de-risking the $15 billion capital commitment needed to build new plants. Its success depends entirely on approval from the Federal Energy Regulatory Commission (FERC), which has already signaled grid reliability as a top priority. A swift green light would confirm the policy acceleration is real. A delay or rejection would expose the plan as political theater, leaving the physical build-out gap wider than ever.

A second, more subtle catalyst is the fate of the EPA's permitting reform. The proposed rule, expected in 2026, aims to provide "flexibility" for developers to start construction on ancillary site structures before securing final air permits. This is a critical piece of the speed puzzle. If finalized and implemented, it would directly address the scheduling uncertainty that has plagued projects. Its absence would mean the regulatory catalysts are incomplete, and the promised acceleration would falter against the same old bottlenecks.

The physical validation of the thesis will be visible in the ground. Investors must monitor the actual build-out of data center clusters in preexisting large hubs like Northern Virginia. The narrative is about AI demand driving power needs, but the proof is in the construction. A surge in new data center announcements and interconnection applications in these established clusters would confirm the demand signal is translating into tangible, on-the-ground activity. Conversely, stalled projects or a lack of new announcements would signal that the policy push is not yet moving the needle.

The key risk is a regulatory or political reversal. The current administration's "energy addition" strategy is a temporary policy stance, not a permanent fix. The proposed auction and permitting changes are executive actions that can be undone. The broader regulatory environment for energy infrastructure remains volatile. Any shift in political power or a change in the FERC leadership could stall or scuttle these initiatives. This creates a window of opportunity that is both narrow and precarious. The thesis assumes the current policy momentum holds, but the underlying infrastructure build-out is a multi-year S-curve. The risk is that the policy catalysts fade before the physical plants are online, leaving the grid vulnerable to the very reliability crisis they were meant to solve.