.svg){kind=avatar}
In previous Green Molecules® Journal posts, we tracked how ERCOT, PJM, and FERC are rewriting the rules of power access, and how regulators are arriving at the same conclusion from different directions: the large load is expected to bring more of its own power. Texas is raising the bar to enter the queue. PJM is trying to stop socializing the cost of demand it cannot yet serve. FERC is planning twenty years out for load that needs power on an eighteen-month timeline.
That gap, between what the grid can plan and what a hyperscaler needs now, is being filled project by project across the country. The practical answer has a name: behind-the-meter generation. In this issue, we look at what is actually getting built, which technologies are capturing the mandate, and why the opportunity sits across a supply chain rather than a single deployment decision.
Behind-the-meter natural gas has moved from a contingency plan to the default power strategy for large-scale AI infrastructure in about two years. Developers have announced roughly 101 gigawatts of onsite natural gas generation capacity across the United States, with more than 57 gigawatts already tied to disclosed equipment orders, meaning turbines and engines actually purchased, not modeled in a pro forma.
To put that figure in perspective, 101 gigawatts is on the order of the entire installed generating capacity of a large industrial nation, being contemplated not for the grid at large but for a single category of customer that barely existed as a load driver five years ago. It is one of the fastest demand reversals the power sector has seen in a generation, and it is happening almost entirely outside the traditional utility planning process.
But announced is not built. Of the capacity tracked, only about 2 gigawatts is operating today. That gap between announced and energized is where the next phase of this cycle will be won or lost, and it is why speed-to-power has become as important an underwriting variable as cost-per-megawatt.
The arithmetic is simple. A dedicated onsite gas plant can be built in as little as eighteen months, against grid interconnection queues stretching five years or longer. For a hyperscaler racing to bring compute online on a customer's contracted schedule, that difference is the entire investment case, the same gap our last issue described between FERC's twenty-year planning horizon and a state-level "prove it first" queue. When the cost of being late is a breached delivery contract with an AI customer, paying a premium for power that arrives on time is not a luxury. It is the rational choice, and it is why capital is flowing toward whatever can be energized fastest rather than whatever is cheapest per megawatt-hour on paper.
The more consequential shift may be structural rather than technological: who is building these plants. Midstream operators, companies that have spent decades moving molecules rather than electrons, are repositioning as direct power providers. Since 2025, strategy has shifted from strengthening the existing gas network toward direct-supply partnerships that pair fuel and generation as a single offering, priced and contracted at the site.
That reframes where incremental gas demand gets monetized: not at the wellhead or city gate, but at the meter of a single, gigawatt-scale customer. It is the Green Molecules® thesis playing out at the point of conversion, where value increasingly sits in owning the last mile where fuel becomes electrons, on the customer's timeline.
For the operators who move and balance the nation's gas, this is a meaningful expansion of the addressable business. Selling molecules into a regulated distribution network is a mature, margin-compressed activity. Selling firm power to a hyperscaler under a long-term contract, with the fuel, the generation, and the reliability bundled together, is a different and more valuable proposition. It turns a transported commodity into a delivered service, and it lets the operator capture a share of the economics that previously accrued to the power producer downstream.
The companies that already own the pipelines, the compression, the storage, and the balancing capability are uniquely positioned to make that move, because the hardest part of siting generation at the edge of the grid is guaranteeing the fuel behind it.
Not every project reaches for the same equipment, and the differences map to where value is accruing. The right way to read this market is not as a contest to crown one winning technology, but as a portfolio in which each option fits a different site, timeline, and risk profile.
Combined-cycle turbines remain the choice for the largest baseload campuses, prized for their efficiency and their ability to anchor a gigawatt-scale site. But tight supply is pushing developers toward alternatives rather than waiting in an OEM backlog, and for many projects the deciding factor is no longer efficiency on a spec sheet but simply what can be delivered and commissioned inside the compute schedule.
High-speed gas engines are gaining ground where speed and flexibility matter most. They start within seconds, ramp quickly, and are fuel-flexible enough to absorb biogas or hydrogen blends as those pathways mature. Deployed in modular arrays, they let a developer scale capacity in increments and keep generating even when individual units are down for service, a resilience profile that maps well to the always-on requirements of AI compute.
Fuel cells are the quieter growth story, carrying a materially lower emissions and noise profile that becomes critical once a project sits near a residential zoning line. They convert gas to electricity electrochemically rather than through combustion, which makes them easier to permit in constrained locations and increasingly attractive where community acceptance is the binding constraint. At least one major manufacturer has reported its backlog more than doubling over the past year, a signal that developers are willing to pay for a technology that clears local scrutiny even where its cost per megawatt runs higher.
Underneath all three sits a layer that matters more than it gets credit for: the systems that make onsite gas infrastructure efficient, transparent, and permittable. This is the part of the opportunity that is easy to overlook and hard to replicate, because it does not show up as a headline megawatt number. It shows up as the difference between a project that clears its air permit and one that stalls, or between a site that wrings full value from every molecule and one that wastes it. Recovering otherwise-wasted energy, measuring and managing emissions, and improving the efficiency of the fuel path are what allow a distributed gas plant to be built at the edge of the grid without importing the problems that get projects blocked.
The regulatory tailwind does not mean this builds without resistance, and that resistance is increasingly priced into financing. Community opposition has contributed to delays or blocks on tens of billions of dollars of projects in the past year alone: a fuel cell facility sited near a residential neighborhood, a pipeline blocked on the way to fueling a gigawatt-scale plant, an air permit stalled long enough to threaten a compute delivery contract. What used to be treated as a late-stage permitting formality is now a front-line underwriting risk, and developers are learning to price the possibility that a project clears every financial hurdle and still fails to clear its community.
Rate-payer politics are catching up too. Because a behind-the-meter project contracts directly with a gas supplier, it can sit outside the reach of state rate regulators, which is precisely why critics argue it lets the best-capitalized loads lock in advantaged pricing while gas-fired generation still sets prices for everyone else. That tension is unlikely to resolve quietly. As more gigawatt-scale loads route around the regulated system, the political pressure to bring them back inside it, through tariffs, exit fees, or new categories of large-load regulation, will grow, and the projects being underwritten today are being built into that uncertainty.
An open question remains underneath all of it: what happens to the asset once the grid catches up. A bridge's residual value depends on what waits on the other side. If onsite generation is a temporary bridge until interconnection queues clear, the economics have to earn out over a shorter life. If it is a permanent feature of how large loads get powered, it is a far more valuable asset. The answer is not yet settled, and it is one of the central variables in how these projects should be valued.
Regulation is pushing large loads to become their own power developers, and gas, in multiple forms, remains the only dispatchable, energy-dense option that can be sited and energized on their timeline. That is not a temporary accident of the current cycle. It is a structural consequence of a grid that cannot plan and build as fast as AI demand is arriving, and it points to a durable role for onsite molecules for as long as that mismatch persists.
That demand does not resolve into one winning technology. It resolves into a supply chain: generation equipment, fuel management, methane and emissions performance, and the efficiency layers that stretch every molecule further, built to operate at the edge of the grid rather than wait on it. The winners will not all be power plants. Many of them will be the companies supplying the parts that make onsite power buildable, permittable, and efficient enough to pencil.
The large loads are showing up with their own answer to power. The opportunity is in the companies building the parts of that answer worth owning.