Across the United States, developers are realizing that the fastest-growing part of their projects isn’t concrete or computing resources—it’s the wait. New data centers designed for high-density AI workloads face multi-year interconnection delays as utilities and grid operators struggle to handle record volumes of new load and generation requests. In response, a new strategy is gaining traction: deploying on-site “bridge power” built around modular natural gas generators combined with utility-scale battery storage in a microgrid setup. This hybrid approach can power campuses years before a full grid connection is available while keeping options open for how those assets are used once the interconnection is established.
A Texas pilot announced in early September 2025 defines the model. Prometheus Hyperscale and Conduit Power plan to colocate dedicated gas-plus-battery power plants at existing ENGIE battery sites along the I-35 corridor, with the first liquid-cooled data center scheduled to open in 2026. The deployments are designed to scale up to 300 MW per site, using existing infrastructure to speed up timelines while reserving growth for the bulk power system in the future. The strategy is practical, but it is not without risks. This Energy Brief looks at how these hybrid systems operate, why they are spreading now, and the operational, commercial, and environmental trade-offs of relying on behind-the-meter generation for reliability.
The bottleneck that created a market for bridge power
The starting point is straightforward: interconnection timelines have lengthened just as demand for AI-ready power has surged. National Lab tracking indicates that by the end of 2024, approximately 2.3 terawatts of generation and storage were seeking transmission access, with median wait times for projects reaching operation stretching to around five years for recent cycles. These figures reflect a system under strain at every stage, from study queues to network upgrades. Grid operators confirm this; even in ERCOT—long known as faster than other regions—large-resource interconnections typically take 18–30 months for the study and approval process alone, and the queue of proposed large loads now far exceeds the pace of approvals. The practical implication for developers is that load can be sited and built more quickly than the infrastructure to transmit it.
AI loads increase urgency. Power-dense, liquid-cooled clusters operate quickly and intensely, and regions with many hyperscale sites have already experienced system-level effects. In mid-2024, a voltage event in Virginia caused dozens of data centers to transfer off the grid almost simultaneously, highlighting how multi-gigawatt, sensitive loads can disrupt grid stability if they move together. Reliability authorities have since flagged large loads as a near-term challenge and are exploring procedural reforms to connect them more safely and predictably. Until those reforms take effect and new capacity is added, developers are not waiting.
What “bridge power” is, and what it is not
“Bridge power” in this context refers to a temporary or interim power plant located on or immediately adjacent to a data center site that provides most or all of the facility’s electrical needs until the long-term grid connection and capacity upgrades are finished. The typical bridge design is a microgrid built around one or more rows of modular reciprocating gas engines, each ranging from 5 to 20 MW, combined with a front-of-the-switch battery energy storage system. The engines handle sustained load, while the battery covers rapid ramps, short-term disturbances, and black-starts the island after an outage. Because the equipment is containerized and vendor-standard, it can be installed quickly, scaled linearly, and—according to many contracts—relocated later.
Two clarifications are essential. First, bridge power differs from traditional backup. Conventional diesel or gas backup systems are sized for rare emergencies and only operate a few hours per year for testing. In contrast, bridge deployments are meant to serve as primary power for months or even years. Second, bridge power does not necessarily imply “off-grid.” Some systems energize from day one as self-sufficient islands, while others connect on a provisional, limited-service basis using existing site interconnection rights, with the microgrid operating behind the meter and the battery and controls managing power quality.
How a hybrid gas‑plus‑battery microgrid works
From an electrical perspective, these systems combine steady-state capability with rapid dynamic responses. Gas engines offer high inertia-like behavior, precise frequency control, and good part-load efficiency compared to simple-cycle turbines of the same size. In an AI data hall with fluctuating demand as training jobs increase, the battery acts as a buffer, reducing peak loads and injecting or absorbing power within milliseconds to maintain voltage and frequency within the tight limits required by the servers. When the campus is islanded, the battery’s grid-forming inverters establish a reference waveform and support the site while engines are brought online successively; if a blackout occurs, the battery provides the black-start path to re-energize switchgear and restart the engines in sequence.
Controls manage the entire system. Modern microgrid controllers predict short-term load, monitor state-of-charge constraints, and direct engines and storage to reduce fuel use and wear while maintaining strict power quality limits. When the grid connection is limited, controls can also set export caps or absorb disturbances that might otherwise trip sensitive loads, lowering the risk of sudden, large transfers that have caused concern among grid operators recently. For developers, the key qualities are speed, modularity, and the value of reusing components.
Case study: Texas bridge power at ENGIE battery sites
Prometheus Hyperscale’s Texas plan with Conduit Power and ENGIE demonstrates the real-world application of the model. The partners aim to install modular gas generators and large batteries at selected ENGIE battery storage sites, colocating liquid-cooled data centers at those same locations. They state that the initial capacity could be operational in 2026, with potential to reach as much as 300 MW per site, and additional sites planned for 2027 and later years. Conduit’s role is to develop and manage the hybrid power plant; Prometheus focuses on the AI-ready, high-efficiency compute campus; and ENGIE provides the real estate, energy management, and trading platform supported by its storage assets.
The configuration is important. By building at existing battery sites, the partners can take advantage of site infrastructure and, in some cases, existing interconnection rights, even if the bridge system itself functions as a separate microgrid behind the meter. Conduit has stated that its battery will help smooth the engines’ output and enable black-start capability; the engines would act as baseload for the data halls. The initial term planned for the bridge plant is about five years—long enough to cover interconnection studies and any needed network upgrades. Once a firm grid connection is in place, the engines can either be kept as long-term backup or peak-shaving or redeployed to other projects that require similar bridge service.
Why this is happening in Texas first
Texas serves as an ideal testing ground. ERCOT has the largest fleet of standalone batteries in the nation and accelerates generation interconnection processes more than most regions. It also faces a growing backlog of large loads and has abundant access to low-cost natural gas. Colocating at battery sites reduces development times and promotes operational synergies. The state has also overhauled its air-permitting process with a standard permit tailored for natural-gas-fired electric engines, detailing emissions, operating guidelines, and administrative requirements that reciprocating engine projects can follow. Meanwhile, public agencies and market participants are developing new tools to address the reliability impacts of rapidly expanding data center loads. In essence, the key elements—siting, fuel, storage knowledge, and clear regulations—are all in place.
Speed‑to‑market, structured for flexibility
Time is the most valuable asset that drives power sales. In a market where full interconnection can take over three years and the supply chain for large turbines is backlogged, row-based gas engines and containerized storage can be procured and installed alongside grid studies. For developers, the commercial models are evolving. Some adopt an energy-as-a-service model with fixed-price capacity and variable fuel pass-throughs; others establish tolling-style agreements with uptime guarantees and power quality SLAs tailored to AI cluster constraints. Having a grid-scale battery on site can generate additional revenue streams once interconnection is complete, including frequency regulation and other ancillary services, but during the transition period, most operators prioritize stability over market participation.
A second, more subtle point is reusability. The modular design of the equipment creates a portfolio effect. Engines can be lifted out and moved as campuses “graduate” to reliable grid service. Batteries can be reconfigured or recontracted as standalone storage. That portability reduces the risk of stranded capital if a load’s long-term profile changes.
Reliability and power‑quality implications for AI data centers
AI clusters enforce strict electrical tolerances. Their power supplies can trip on relatively small voltage dips, and their combined size means even a single campus can match a mid-sized power plant. In the Virginia incident, about 1.5 GW of data center load dropped almost all at once, challenging grid balancing. Local microgrids reduce the risk that external disturbances spread into the racks and cause transfers to standby. When a transfer is needed, the microgrid’s engines are already running and synchronized with the site’s bus, with the battery absorbing the transient. This architecture is designed to meet the UL, ISO, and tier standards that operators set for uptime, while aligning the site with new grid-integration regulations from reliability authorities.
Environmental and permitting trade-offs
Using natural gas as the primary power source over a long period has clear environmental implications, and it’s important to state them directly. Compared to diesel backup fleets operating for limited hours, a main power gas plant will emit more because it runs much longer. Compared to grid power, the overall impact depends on the local generation mix and the duty cycle. Regulatory boundaries are in place. New stationary spark-ignition engines must meet federal standards under 40 CFR Part 60 for NOx, CO, and VOCs; operators need air permits, and in Texas, a specific standard permit defines acceptable configurations and emissions for natural-gas-fired engine EGUs. On a pollutant-by-pollutant basis, natural gas engines generally emit substantially less particulate matter and sulfur oxides than diesel, and—at similar duty—lower NOx than uncontrolled diesels. However, Tier 4-Final diesel after-treatment can reduce this gap. Methane supply-chain emissions are another part of the picture; several developers pair these projects with carbon-offset programs or renewable natural gas to lower carbon intensity, but the effectiveness of such measures remains a topic of active policy debate.
Batteries alter the profile in two ways. Operationally, storage cuts peaks and improves engine loading, boosting fuel efficiency and lowering emissions per MWh. Functionally, a battery enables short-duration outages and black-starts without running engines. These benefits are real, but they do not remove the key fact that bridge power still depends on fossil fuels during the transition. Policymakers, communities, and developers will continue to examine siting and environmental justice concerns, particularly in areas where campuses are located near already burdened airsheds.
Risks developers and offtakers must manage
Commercial risk centers on four main areas. First, fuel: projects rely on natural gas markets, so volatile basis differentials can make cost predictions difficult unless hedged. Second, permitting: while engine-based plants have clearer permitting paths than boilers or turbines in Texas, delays and public opposition still pose risks to completion. Third, interconnection: bridge periods can be delayed if studies or network upgrades are needed; contracts should specify what happens if the “bridge” takes longer than expected. Fourth, technology: AI hardware roadmaps are shortening update cycles and increasing rack densities. Power distribution and cooling systems supporting 30–60 kW/rack today may need upgrades sooner as operators aim for 100 kW/rack and higher.
There is also a system-level factor to consider. If many sites adopt self-supply for extended periods, coordination with grid operators becomes even more important, not less. Reliability organizations are creating new frameworks for large load interconnection, flexible demand, and emergency disconnection rules; developers designing grid-supportive microgrids will probably find it easier to secure long-term operating boundaries.
What “after the bridge” looks like
When a campus switches to firm grid service, operators have three main options. Engines can be demobilized and redeployed; they can be kept for N+1 or N+2 backup, shifting runtime from continuous duty to rare emergencies; or they can be dispatched as a peak-shaving asset under a demand-response or interruptible-service arrangement. Batteries usually stay, either as behind-the-meter peak management and ride-through assets or as market-facing storage if export is allowed. Co-location with a storage owner-operator adds flexibility: the site can continue using storage for power quality and resilience while also monetizing grid services, depending on tariffs and interconnection limits. Meanwhile, an increasing number of jurisdictions are considering “priority interconnection pathways” for large loads supported by on-site generation or flexibility, indicating that bridge-plus-grid hybrids may become common rather than rare.
A broader trend, not a one-off
The Texas pilot is not isolated. In late August 2025, a developer in Pennsylvania proposed a 944 MW behind-the-meter gas plant paired with storage to supply a data center campus, with a backup grid tie and an operational goal later this decade. Equipment vendors now publish design guides and white papers specifically for “bridge power” in mission-critical applications, and microgrid integrators are customizing controls to AI’s “spiky” load shapes. These signals indicate a stable submarket in which on-site generation and storage projects support each other until the network advances.
Conclusion
Bridge power offers a practical solution to a mismatch in infrastructure development: the digital economy’s rapid pace of building infrastructure and installing servers considerably outstrips the grid’s slower process of adding wires and increasing capacity. Hybrid microgrids, made up of natural gas generators and battery storage, enable developers to launch data centers on predictable timelines despite multi-year interconnection delays, while still allowing for future grid integration and asset repurposing as needs change. This method improves local reliability and power quality for sensitive AI clusters and—when carefully designed—can lower system risks by avoiding large, sudden shifts between grid supply and on-site power. It also places environmental and regulatory responsibilities on the site owner. Engines used as primary power sources emit pollutants; permits and compliance are crucial; methane supply chains are significant. For professionals, the main question is whether the speed-to-market advantages outweigh the costs during the bridge period and whether the transition plan is feasible.
In 2026 and beyond, expect to see more projects that resemble Texas: colocated with storage, modular, contractually flexible, and designed for eventual integration with the grid. In a constrained system, the quickest route to reliable power is often the one you can build yourself—so long as you also plan for the day you will not need it.