Under the Radar: From GPUs to Gas Pipelines – The Unseen AI Supply Chain

The accelerating adoption of artificial intelligence has created the most energy-intensive computing cycle in modern history. Training foundation models and running large-scale inference systems requires dense clusters of graphics processing units (GPUs), often operating continuously across multiple geographies. This is not a one-time trend or a temporary infrastructure cycle. It is the foundation of a generational energy demand shift that is already underway.

Goldman Sachs estimates that power consumption from data centers will rise by one hundred sixty percent by 2030. This demand is largely driven by artificial intelligence training and inference workloads, which use exponentially more power than traditional cloud computing. Unlike earlier tech buildouts, the scale of electricity required to run artificial intelligence models cannot be deferred, throttled, or reduced through software efficiencies. This new workload is physical. It is thermally intensive, geographically dispersed, and entirely dependent on baseload power that must be stable and scalable.

While the long-term future may include advanced nuclear, long-duration batteries, and upgraded transmission systems, the reality today is simple. The power needed to support artificial intelligence must come from existing assets or be built quickly. Renewable sources alone cannot meet the demand due to intermittency, land constraints, and lack of utility-scale storage. As a result, natural gas has reemerged as the dominant short and medium-term solution to power artificial intelligence infrastructure.

Natural gas generation offers dispatchable, high-reliability baseload power that can be sited near data center clusters and brought online much faster than nuclear or high-voltage transmission. Developers and hyperscalers are turning to gas-fired plants to provide the electricity needed to train models, operate cloud platforms, and maintain consistent uptime across inference clusters.

At the heart of this transition are Fuel and Gas Supply Agreements, or FGSAs. These long-term contracts between gas producers, midstream operators, and utilities or power plant operators serve as the foundation for financing and building new gas-fired generation. FGSAs ensure stable pricing and delivery volumes over ten to twenty-year terms, providing the confidence necessary to support new infrastructure development. For investors, companies that can deliver consistent volumes into FGSAs represent a critical piece of the artificial intelligence energy supply chain.

Several small-cap natural gas producers and midstream firms are particularly well positioned to benefit from this demand shift. Three names stand out due to their asset footprints, financial discipline, and growing relevance in the AI-power ecosystem.

Summit Midstream Corporation SMC is a small-cap natural gas and crude oil midstream company that owns and operates strategically located gathering and processing assets across the United States. Summit's operations span the Williston Basin, the Piceance Basin, the DJ Basin, the Barnett Shale, the Delaware Basin, and the Utica Shale. While the company recently exited its Utica assets, it retains a diversified and underappreciated footprint in basins where data center-related gas demand is rising sharply.

Summit's core business involves gathering, compression, and processing of natural gas and crude oil for upstream producers under long-term, fee-based contracts. These contracts are largely volumetric in nature, which provides stable cash flow independent of commodity prices. Summit's infrastructure is positioned close to existing and proposed gas-fired generation projects in the Permian and Rockies regions. These areas are receiving renewed attention from developers seeking to collocate power generation with artificial intelligence data center buildouts.

Summit's recent transactions have improved its financial condition significantly. In 2024, the company sold its Utica gathering system to MPLX for over six hundred twenty-five million dollars in cash. That transaction allowed Summit to reduce debt and reposition capital toward higher-return projects in the Permian and DJ basins. As of the first quarter of 2025, Summit had over three hundred million dollars in unrestricted cash and a four hundred million dollar revolving credit facility. This provides ample flexibility to fund growth tied to artificial intelligence-related gas infrastructure, including potential equity participation in FGSAs and Peaker plant development.

Summit's assets are particularly relevant to the growth of behind-the-meter generation, where data centers and hyperscalers collocate directly next to gas-fired plants. These systems reduce dependence on strained transmission systems and require reliable midstream gathering and delivery. Summit's long-term fee-based contracts and strategic locations make it a compelling candidate for future growth driven by artificial intelligence infrastructure demand.

Antero Midstream AR is a mid-cap natural gas and water infrastructure company with operations concentrated in the Appalachian Basin. It provides gathering, compression, processing, and water handling services to Antero Resources, its primary customer and a leading producer in the Marcellus and Utica Shales. The company operates over five hundred miles of gas gathering pipelines and owns a large network of compression stations, processing facilities, and freshwater delivery systems.

Antero Midstream's gas volumes are directly tied to Antero Resources' production, which is largely hedged and supported by long-term drilling inventory. This provides stability and predictability for investors. While Antero's business model is anchored in the Appalachian region, it is uniquely positioned to deliver into the eastern United States power corridor, where artificial intelligence-driven data center expansion is increasing faster than transmission capacity can accommodate.

The Mid-Atlantic and Southeastern United States are experiencing an influx of hyperscale data center construction, particularly in Virginia, North Carolina, and Georgia. These locations have favorable tax structures, relatively low energy costs, and access to major population centers. However, the power grids in these regions are under strain, with interconnection wait times extending up to five years. Natural gas generation offers the only near-term path to meeting this load growth, and Antero Midstream is one of the few firms with the assets and throughput capacity to support this effort.

Antero Midstream's infrastructure is underpinned by long-term service agreements, including cost-of-service contracts and minimum volume commitments. This model supports reliable cash flow generation and has enabled the company to pay a consistent dividend while maintaining capital discipline. As more utilities and private developers pursue long-term FGSAs to anchor new gas-fired plants, Antero Midstream stands to benefit from rising throughput volumes and expanded infrastructure requirements. Its ability to deliver clean, dry gas into key demand centers positions it as a vital component of the artificial intelligence energy chain.

Epsilon Energy EPSN is a small, independent natural gas producer with operations concentrated in the Marcellus Shale. It also maintains non-operated positions in the Anadarko Basin and legacy interests in the Permian Basin. Epsilon is distinguished by its conservative balance sheet, high free cash flow generation, and vertical integration of midstream infrastructure.

Epsilon owns and operates its own gas gathering systems in northeastern Pennsylvania. This includes pipelines and compression assets that allow the company to deliver gas directly to local utilities and pipeline interconnects. This control over midstream functions provides Epsilon with a level of margin stability and operational flexibility that is uncommon among producers of its size.

The company is entirely unlevered, with no long-term debt, and has consistently returned capital to shareholders through dividends and share repurchases. Its production profile is flat and intentionally capital efficient, focused on maximizing return on invested capital rather than growing volumes for volume's sake.

Epsilon is uniquely well suited to participate in smaller-scale FGSAs and Peaker plant arrangements in the Northeast. As data center developers seek to diversify their power supply and reduce interconnection risk, Epsilon's ability to deliver physical gas under flexible structures becomes highly attractive. The company has already demonstrated its ability to contract with local industrial and utility customers and is actively evaluating opportunities to support on-site or microgrid-style natural gas generation projects.

Given its strong cash position, debt-free balance sheet, and control over both production and delivery infrastructure, Epsilon offers a rare combination of defensiveness and optionality. If even a fraction of the forecasted power buildout linked to artificial intelligence emerges in its operating footprint, Epsilon could see significant valuation uplift from a low base.

Artificial intelligence may be the most transformative technology trend of our lifetime, but it is also the most power-hungry. Meeting this demand requires natural gas, not as a bridge fuel, but as the primary source of dispatchable baseload electricity over the next decade. Fuel and Gas Supply Agreements are the foundation of this buildout, and the companies that control the molecules, pipelines, and delivery systems will become critical enablers of the artificial intelligence economy.

Summit Midstream Corporation, Antero Midstream, and Epsilon Energy are three companies with distinct business models and asset footprints, but they share a common advantage. They are positioned to deliver reliable volumes of natural gas into power-constrained regions at a time when hyperscalers, utilities, and developers are scrambling to secure long-term energy contracts. In the rush to power artificial intelligence, these companies may turn out to be the quiet but essential infrastructure winners.