NERC's 2025 Long-Term Reliability Assessment projects summer peak demand growing by 224 GW and winter peak demand by 246 GW over 2026 to 2035, with AI and other data centers driving most of that growth.¹ New resource additions continue to lag behind it. That gap is what stretches a utility interconnection timeline for an AI campus into years rather than months. Bridging gives that campus an intermediate option: energize from on-site gas generation while the connection is built, then transfer to grid supply once it's complete.

What Bridging Is
Bridging does not bypass the utility. The site is energized from on-site power generation, with the MV-UPS and its integrated battery storage as the stability block between the generating asset (i.e., gas turbine) and the data halls. Operation can begin as IT hardware is delivered, instead of on the utility's grid construction schedule.

One MV-UPS interface, two stages. On-site generation power the load from day one; utility interconnection takes over once construction finished, with no change to the interface in between.

Why the Generator Cannot Take the Load Directly
A gas turbine has a defined ramp-rate limit. AI training workloads produce fast, software-driven demand swings that traditional data center loads do not, which EPRI names as a key source of forecasting error for electric companies planning around AI-driven demand.² Connected directly, the generator meets those swings as frequency and voltage excursions.

Standby generators and low-voltage UPS systems are built to carry a load through a utility outage, not to serve as the primary supply. The MV-UPS sits between generator and load and operates grid-forming on the load side, at a rated output of 34.5 kV. In islanded mode the MV-UPS also provides gas turbine soft starting/stopping and black start.³

What a Bridging Architecture Should Deliver
A bridging architecture has to meet four requirements.

First, the turbine is a high-capex asset that takes the load's behavior at two points in the project. During bridging it is the only generation source, so the load's pulsations reach it directly. Pulsations that align with the shaft's resonant frequency cause sub-synchronous torsional interaction (SSTI) and forced power oscillations, which NERC's guideline flags for large loads co-located with generation⁴ and the grid. At utility connection the same behavior becomes an interconnection question. The guideline calls for oscillation risk to be studied before such loads are energized and encourages transmission owners to consider monitoring and mitigation provisions in interconnection agreements.⁴ The MV-UPS is designed to keep those pulsations off the generator and the grid.³

Second, the MV-UPS conditions the power drawn from the source, and integrated batteries absorb or inject the difference, so the source sees a controllable, predictable power trajectory instead of the load's steps.³ That predictability is what the utility studies at grid interconnection.

Third, a defined interface gives large load entities defined characteristics to document. NERC notes that co-located generation lets large loads switch between grid supply and on-site resources, which makes net demand harder to forecast, and asks large load entities to document expected dispatchability, ramp rates and on-site generation and storage.⁴ With the MV-UPS holding a near-constant draw at the source, on-site generation only has to supply the average power rather than the load's peaks and swings, and forecasting can be built on that average instead of a profile of repetitive swings.

Fourth, the load-side interface is designed to stay the same when the source changes from on-site generation to the utility. The design intent is that transfer becomes tuning to the utility's requirements, not rebuilding the site. NERC's guideline expects ride-through requirements to be set per interconnection, so some settings will still change, and the utility still studies the connection.⁴ 

NERC is still updating its registry criteria and reliability standards for large loads and has issued its guideline as interim direction.⁴ Brought in during planning, the MV-UPS lets engineers size its storage to the site and plan for both islanded and grid-tied operation.³

A bridging architecture is a bet on the turbine and the schedule at once, and the interface decides which way it goes. The MV-UPS holds that interface steady regardless of which side is supplying power, keeping the risk off the asset that cost the most to install. Get it right, and the campus starts earning revenue well before the utility connection is complete.

Uncertain whether a site needs an MV-UPS ahead of its utility connection?Answer this four-question check.

References

1. NERC, 2025 Long-Term Reliability Assessment, January 29, 2026.
2. Electric Power Research Institute, Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption, White Paper 3002028905, May 2024.
3. GE Vernova Power Stability Fact Sheet: MV-UPS for Data Centers, Rev1, August 10, 2026.
4. NERC Large Loads Working Group, Reliability Guideline: Risk Mitigation for Emerging Large Loads, May 2026 (Final), Executive Summary and Chapters 4 and 5.