AI's Energy Thirst: Why Gas Pipes Are the Next Critical Bottleneck
By Nikhil Raghavan · Reporting from San Francisco ·
Natural gas scarcity won't be determined by underground reserves, but by aging infrastructure and processing limits.
The idea that AI will trigger an energy crisis, specifically by creating a "knife fight" for natural gas in 2028, was the most jarring takeaway from the latest episode of Invest Like the Best. While the premise—that massive computational demand meets constrained infrastructure—is compelling, the real story isn't just about depletion; it’s about how deeply ingrained systemic bottlenecks are.
On the podcast, Matt argued that US natural gas storage levels are projected to drop materially below historical records by mid-2028, with scarcity worsening through 2029. He highlighted that while AI compute/data centers are a massive new demand driver, the core issue is not necessarily physical depletion underground; rather, it's "the ability to service the demand within a given timeframe, constrained by infrastructure, regulations, and contracts." This diagnosis—that the problem is software, not geology—is critical.
The discussion tracked how US export capacity has already grown significantly (from early shipments to about 15 BCF/day today) but faces hard limits. Matt estimated that while existing acreage offers a maximum deliverability of 128–132 BCF/day, the required demand from AI could reach 12 to 15 BCF/day by the early 2030s in an unmitigated scenario. Speaker A added weight to this structural risk, noting that current market complacency assumes gas is abundant, leading to underinvestment and a severe mispricing of near-term supply risk.
The Middleware Constraint: Why Pipes Matter More Than Reserves The most insightful part of the discussion was the detailed breakdown of rate limiters in gas delivery. It’s easy for an outsider to assume that if there's gas in the ground (like the Permian or Appalachia), it will flow to the market. But Matt pointed out that the primary constraints are far more complex: processing natural gas, dealing with NGLs and sulfur compounds, meeting pipeline specifications, and the sheer difficulty of building new gathering pipes from wellhead to processor.
This emphasis on midstream bottlenecks is where a policy analyst needs to focus. The risk isn't simply that we will run out; it's that the existing infrastructure—the "middleware"—cannot handle the exponential load demanded by AI compute, even if resources are physically present. Furthermore, interstate pipelines remain difficult due to regulations and monopolies, creating regional choke points regardless of how much gas is pumped into the ground.
The Nuclear Imperative for Computational Power If natural gas is facing a structural crunch that could drive prices toward $8-$10/MCF by 2030—potentially forcing utilities to choose between powering AI compute or maintaining affordable consumer bills—the immediate policy implications become stark.
The podcast repeatedly circled back to the long-term solution: nuclear power. While Small Modular Reactors (SMRs) were discussed, Matt strongly advocated for large-scale commercialization of established designs like the AP1000, arguing that government action must fund multiple units to "derisk" the supply chain and achieve necessary gigawatt scale by 2033/2034.
The current economic model is deeply flawed because it assumes a flat natural gas cost curve out to the 2030s, leading companies to deploy tens of billions into short-sighted assets—distributed gas generation sets for AI power—without sufficient financial contracting to secure physical supply. This bet on future abundance is dangerously misplaced.
The confluence of these factors leaves us with a clear picture: the energy sector has been optimized for fossil fuels and rapid growth, but it has failed to build the resilient, non-fossil backbone required by the data age. The current narrative that simply requires "more gas" ignores the physical limitations of pipelines, processing plants, and regulatory approval times.
The technological revolution driven by AI is not merely an energy demand increase; it is a systemic stress test for global infrastructure. Solving this crisis demands treating energy supply as a critical national asset—one requiring massive, coordinated public investment in reliable, scalable power sources that bypass the inherent volatility of fossil fuel pipelines and markets.