en.Wedoany.com Reported - The rapid expansion of the artificial intelligence industry is driving a continuous surge in electricity demand from data centers, with power consumption showing explosive growth. The high-speed operation of AI training and inference tasks has also significantly intensified load fluctuations in data centers. Compared to traditional industrial loads, AI computing clusters impose higher requirements on power supply stability, continuity, and response speed, which is forcing an accelerated iteration and upgrade of power infrastructure.
Texas, USA (hereinafter referred to as "Texas"), is a typical sample of this round of industry transformation. The local penetration rate of new energy ranks among the highest in the United States, while it also hosts a large number of newly built AI data centers. The combination of highly volatile new energy output and drastically fluctuating AI electricity loads means the Texas grid is no longer just a transmission channel but is gradually becoming a real-world testing ground for verifying the adaptation of power stability to high-computing loads.
In the past two years, Texas has seen a boom in behind-the-meter (BTM) power source construction. Statistics from research firm Cleanview show that from 2024 to 2025, the scale of officially announced BTM data center projects in the US was approximately 56 GW, of which over 20 GW is located in Texas; in just the first four months of 2026, Texas added another 10 GW of related projects. Currently, the construction pace of AI data centers has significantly outpaced the expansion speed of traditional power grids. Although the Texas grid operator ERCOT is accelerating the grid connection approval for large electricity loads, entirely new transmission lines will not be operational until at least early 2030. This means that projects relying solely on connection to the public grid will face delays extending several years.
Amid this supply-demand mismatch, a new power supply model is gradually taking shape. Data center campuses are building their own natural gas power plants and supporting them with energy storage equipment to achieve partial off-grid operation, with subsequent connection to the main public grid. The industry refers to this approach as "bridge-to-grid." In this industry shift, the role of energy storage has also undergone a fundamental change. In the past, energy storage was often a supporting facility for new energy generation, primarily handling auxiliary tasks like frequency regulation and peak shaving. Now, energy storage is gradually becoming an indispensable foundational configuration in the AI power system. Whether it's natural gas generation paired with storage, wind and solar farms equipped with long-duration storage, or grid-connected data centers adding battery systems, energy storage has transformed from an optional accessory into core equipment ensuring stable computing operations.
The stringent requirements of AI equipment for power supply stability are the core reason energy storage has become standard in local data centers. A single training cycle for a large AI model can last weeks or even months; even millisecond-level voltage or frequency fluctuations can cause training interruptions or server trips, rendering all prior training data invalid and requiring re-iteration. Grid fluctuations tolerable for ordinary commercial and industrial facilities constitute major operational incidents for computing centers. Based on the unique characteristics of computing loads, ERCOT has introduced stricter grid connection rules for high-power electricity loads. In 2025, Texas advanced the SB6 bill focusing on managing large new loads, authorizing ERCOT to strengthen management and requiring electricity facilities above 75 MW to possess greater tolerance to grid disturbances. Facilities that cannot meet these requirements may be prioritized for load shedding during grid emergencies.
Currently, AI data centers in Texas primarily adopt three types of power supply schemes. The first is the "bridge-to-grid" model using natural gas plus energy storage, which is currently the fastest to implement and most widely adopted route in Texas. A key reason this model took off first in Texas is that the state possesses energy conditions unavailable in other major data center clusters, such as abundant low-cost natural gas resources, especially in West Texas adjacent to the Permian Basin, the largest shale oil and gas producing region in the US, where gas prices are significantly lower than in the Northeastern US. The Matador project under construction by US private energy developer Fermi America is a typical case of this model. The energy park has a total planned scale of 17 GW, including 11 GW of natural gas generation capacity, supplemented by 4.4 GW of nuclear, photovoltaic, and energy storage equipment. Another representative project is the collaboration between artificial intelligence data center developer Prometheus Hyperscale, Texas independent power producer Conduit Power, and French utility company ENGIE, where a single campus can provide approximately 300 MW of transitional power supply capacity.
The second mainstream solution is "grid connection plus energy storage." Data centers maintain access to the public grid while equipping large-scale energy storage facilities. Through energy storage, they optimize electricity consumption characteristics, allowing highly volatile AI loads to adapt to the grid's operational rhythm. The Hidden Lakes energy storage project, co-developed by US energy storage developer GridStor and European energy trader Axpo, represents this model. Located in Texas, the project has a planned scale of 220 MW/440 MWh, reached a long-term revenue swap agreement in 2025, and is scheduled for operation in 2026. This revenue swap model financializes energy storage, beginning to deeply integrate with AI electricity demand. What data centers are purchasing is not just the battery itself, but a new energy service system built around stable power supply, real-time regulation, and price risk management.
The third model is a green power computing solution using new energy plus long-duration energy storage. Leveraging Texas's low-cost new energy resources and pairing them with multi-day long-duration energy storage, this approach builds a relatively independent and stably operating new energy power supply system. In 2024, US long-duration energy storage company Form Energy reached a cooperation agreement with AI infrastructure company Crusoe, planning to equip the latter's data center project with up to 12 GWh of iron-air battery energy storage systems. Iron-air batteries can achieve continuous discharge capability for up to approximately 100 hours, possessing multi-day energy scheduling capacity, which can physically break the intermittency constraints of new energy generation.
In traditional power systems, energy storage primarily served short-term grid balancing, with profitability relying on market trading arbitrage. However, in the era of high-load AI power consumption, the core function of energy storage is shifting, gradually transforming to provide stable, predictable power for highly volatile computing loads. On one hand, energy storage is being solidified into infrastructure assets with stable cash flow through financial instruments like revenue swaps and long-term power purchase agreements. On the other hand, energy storage is deeply integrated into the data center energy architecture, becoming a critical guarantee for continuous computing operations. The concentrated implementation of various innovative power supply models in Texas is the result of multiple converging objective conditions. Local new energy installed capacity is high, leading to significant grid volatility; ERCOT operates a pure energy market mechanism without capacity compensation, directly exposing market participants to electricity price and supply-demand fluctuations. After the concentrated deployment of AI data centers, the original power fluctuations were further amplified, making stable power supply a scarce industry resource. Against this backdrop, the previously clear industrial boundaries between generation, storage, grid, and computing centers are being broken down. Data centers are no longer just electricity consumers but are actively engaging in power source configuration, energy storage construction, and electricity price risk management; energy storage is no longer merely auxiliary equipment attached to new energy but is gradually evolving into new infrastructure possessing attributes of power supply assurance, load regulation, and financial characteristics. Texas is just one of the earliest regions where this wave of change has manifested in a concentrated manner; the reshaping of the power system by AI has only just begun.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
