In August 2026, Amazon confirmed plans for a natural gas facility in Pecos County, Texas, capable of producing up to 7.65 gigawatts of electricity. If completed as planned, it would be the largest power plant in U.S. history — larger than Washington State's Grand Coulee Dam, the current record holder at roughly 6.8 GW.
The plant is not being built for the grid. It is being built behind the meter — on-site, adjacent to the data center it will power, and initially operating independently of the Texas electricity grid.
Amazon is not alone. Nearly 60 behind-the-meter gas projects have been reported since the start of 2025, representing roughly 90 gigawatts of combined capacity. Cleanview, a data center tracking firm, identified around 60 data centers planned with a combined 97 GW of behind-the-meter generation — approximately the total installed capacity of Mexico.
The behind-the-meter model has become the default answer to a simple problem: the grid cannot deliver power fast enough. It is also creating a new set of problems that the industry has only begun to confront.
Why Behind-the-Meter
The attraction is straightforward. Grid interconnection queues in the U.S. now stretch four to seven years. Data centers need power in 18 to 24 months. Building generation on-site offers a way to bring power online without waiting for new transmission infrastructure.
The economics are compelling for developers. Rabobank estimates sustained delivered energy costs at year five of $101 to $107 per MWh for behind-the-meter gas generation, depending on technology — competitive with grid-supplied power, without the queue.
Behind-the-meter also offers regulatory advantages. In Nevada, a developer seeking approval for a 510 MW gas plant at the Tahoe-Reno Industrial Center noted that behind-the-meter power "does not rely on public utility infrastructure and therefore has less oversight by state energy regulators." The power would not have to meet state renewable energy goals.
The scale of the shift is visible in individual projects. Meta is building two 286.5 MW natural gas plants in New Albany, Ohio, to power its Prometheus data center complex. Microsoft has committed to a 2 GW data center campus in Pecos, supported by a 2.67 GW power plant. SpaceX accounts for roughly three-quarters of the 2 GW of behind-the-meter data center capacity operating in the U.S. as of spring 2026.
The Ratepayer Question
The behind-the-meter model was supposed to solve the ratepayer problem. If data centers build their own power, the argument goes, they do not shift grid upgrade costs onto residential customers.
That argument is more complicated than it appears.
Even behind-the-meter data centers typically maintain backup connections to the grid for reliability and supplemental power. Those connections can require infrastructure paid partly by local ratepayers. And the demand for natural gas from dozens of new behind-the-meter plants puts upward pressure on regional gas prices, which raises the cost of gas-fired electricity for everyone — including households.
The White House recognized the political risk. On March 4, 2026, President Trump issued a proclamation establishing the Ratepayer Protection Pledge. Seven leading AI companies and hyperscalers — Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI — signed on, committing to "build, bring, or buy all of the energy needed" for their data centers and "cover the cost of all power delivery infrastructure upgrades required for their data centers."
The pledge was expanded in July 2026, bringing in utilities, data center developers, and state leaders. But the pledge is voluntary and nonbinding. It commits signatories to a principle without giving regulators a tool to enforce it.
The Regulatory Response
FERC has moved to clarify the rules.
In December 2025, FERC issued an order finding that PJM's tariff was "unjust and unreasonable" as it related to co-located load arrangements. The commission directed PJM to establish clear rules supporting co-location of data centers and generation facilities, and to modify its rules for behind-the-meter generation.
FERC rejected arguments that co-location falls outside its jurisdiction. The commission found that it retains exclusive authority over the interconnection of generating facilities to the interstate transmission system, and that co-located arrangements "necessarily involve transmission service."
In June 2026, FERC escalated further. The commission issued six show cause orders directing each of the six FERC-jurisdictional RTOs and ISOs to justify or revise their tariff provisions governing large-load interconnection and co-location arrangements. Each RTO was required to submit an informational report within 30 days.
The commission's position was clear: behind-the-meter generation would operate under FERC's rules, not outside them.
The China Comparison
China faced the same physics. It did not face the same governance problem.
China's approach is called source-grid-load-storage integration. The policy was formalized in the 15th Five-Year Plan for the New Power System, issued in August 2026, which explicitly calls for using computing facilities to develop "source-grid-load-storage integration, green power direct connection, and other new models" to achieve "aggregated green power trading, local consumption, and increased green power share."
The model works differently from behind-the-meter gas generation. Instead of building a dedicated power plant, data centers are integrated into a regional power system that coordinates generation, transmission, load, and storage. The goal is not to bypass the grid — it is to make the grid work better.
The flagship project is in Ulanqab, Inner Mongolia. In July 2025, the region completed China's first green power direct connection project under the source-grid-load-storage model. The project pairs 300 MW of renewable generation (200 MW wind, 100 MW solar) with 45 MW of battery storage, and operates on a "load determines source, source follows load" principle.
In Ningxia, a larger project is underway. Phase one totals 2 GW, with a total investment of 8.7 billion yuan ($1.2 billion). It combines 500 MW of source-grid-load-storage photovoltaic with 1.5 GW of wind, using a dual-track supply system that combines "physical direct supply" with "bilateral trading." The photovoltaic portion was fully connected and began direct green power supply to data centers in February 2026. The wind portion is scheduled for full connection in September 2026, at which point it will supply 2.29 billion kilowatt-hours annually to the Zhongwei cloud base data center.
The structural difference between the two approaches is significant. In the U.S., data centers build their own gas plants because the grid cannot deliver power fast enough. In China, data centers are integrated into a planned power system that coordinates generation and load at the regional level.
What the Divergence Means
The behind-the-meter gas boom reflects a systemic problem rather than solving it.
It is a symptom of a grid connection process designed for individual generation projects competing for transmission capacity, not for gigawatt-scale loads arriving in 18-month windows. It is a symptom of a regulatory framework that has been playing catch-up since the first hyperscale data center plugged in. And it is a symptom of a political system where local communities bear the costs and the benefits accrue elsewhere.
The ratepayer protection pledge acknowledges the problem but does not solve it. FERC's show cause orders will clarify the rules without accelerating the timeline. Amazon's Pecos County plant — the largest power plant in U.S. history — is being built because waiting for the grid would take longer than building a facility that could power a small country.
China's source-grid-load-storage model operates in a different political system, with different property rights, different regulatory authority, and different energy markets. But it addresses the same problem — how to power gigawatt-scale computing — with a mechanism designed for the scale, rather than a workaround built to avoid the queue.
The U.S. is building gas plants behind the meter because that is what the system incentivizes. Until the system changes, the gas plants will keep coming.
Sources:Data Centre Magazine (August 20, 2026); The Nevada Independent (June 26, 2026); Cleanview via New Vision (August 11, 2026); Data Center Dynamics (August 10, 2026); Industrial Info (June 9, 2026); K&L Gates (January 15, 2026); Bloom Energy (July 29, 2026); Data Center Knowledge (April 15, 2026); Federal Register Proclamation 11014 (March 4, 2026); White House Fact Sheet (March 4, 2026); Orrick (July 6, 2026); Bracewell (July 9, 2026); Securities Times (July 23, 2026); The Paper (September 13, 2026); Xinhua (May 6, 2026); National Energy Administration via 21st Century Business Herald (August 3, 2026); Wired (August 25, 2026); Utility Dive (March 17, 2026); Gas Turbine World (August 30, 2026); BigGo Finance (August 26, 2026).
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.
Forecasts from third-party analysts can change with market conditions.
Cost and pricing examples are point-in-time estimates; actual rates vary.
Country and company comparisons rely on public reporting, not operational data.
This sector moves fast; timelines and deal terms may be updated later.
Company deals and regulatory rulings may evolve; verify current status.
AI infrastructure is changing quickly; claims can become outdated soon.
Sources
- Data Centre Magazine (August 20, 2026)
- The Nevada Independent (June 26, 2026)
- Cleanview via New Vision (August 11, 2026)
- Data Center Dynamics (August 10, 2026)
- Industrial Info (June 9, 2026)
- K&L Gates (January 15, 2026)
- Bloom Energy (July 29, 2026)
- Data Center Knowledge (April 15, 2026)
- Federal Register Proclamation 11014 (March 4, 2026)
- White House Fact Sheet (March 4, 2026)
- Orrick (July 6, 2026)
- Bracewell (July 9, 2026)
- Securities Times (July 23, 2026)
- The Paper (September 13, 2026)
- Xinhua (May 6, 2026)
- National Energy Administration via 21st Century Business Herald (August 3, 2026)
- Wired (August 25, 2026)
- Utility Dive (March 17, 2026)
- Gas Turbine World (August 30, 2026)
- BigGo Finance (August 26, 2026).
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.; Forecasts from third-party analysts can change with market conditions.; Cost and pricing examples are point-in-time estimates; actual rates vary.; Country and company comparisons rely on public reporting, not operational data.; This sector moves fast; timelines and deal terms may be updated later.; Company deals and regulatory rulings may evolve; verify current status.; AI infrastructure is changing quickly; claims can become outdated soon.