GE Vernova has unveiled technology that could fundamentally reshape how AI data centers manage power, potentially tripling the company's revenue from each gigawatt of capacity it serves. The Massachusetts-based energy company revealed a medium-voltage uninterruptible power supply (MV-UPS) on August 24, designed to keep data centers running on stable, continuous power around the clock. This innovation addresses a critical but often overlooked problem: even brief power interruptions can derail AI workloads, and sudden spikes in electricity demand strain both local grids and on-site power generation systems.

Why Does Data Center Power Stability Matter So Much?

AI data centers consume enormous amounts of electricity, but the challenge goes beyond sheer volume. The real problem is variability. When AI workloads shift in size, electricity demand can spike or dip unpredictably, sending those changes back to the grid or to on-site power sources like gas turbines. Traditional uninterruptible power supplies operate at low voltage, meaning they protect only small sections of a facility separately, leaving the larger electrical system vulnerable to these sudden swings.

GE Vernova's MV-UPS moves protection to the medium-voltage level, creating what the company calls "a stability block between the facility and its power supply." This allows one coordinated system to protect much larger portions of a data center while preventing sudden demand changes from rippling back to the grid or generation sources. The result is smoother power flow for AI workloads of all sizes and less strain on regional electrical infrastructure.

How Could This Technology Transform GE Vernova's Business?

The financial implications are substantial. Currently, GE Vernova generates roughly $200 million to $300 million in total electrification content, including substations and grid software, for every gigawatt of data center capacity it serves. During the company's second-quarter 2026 earnings call, CEO Scott Strazik indicated that adding the MV-UPS, along with solid-state transformers and other emerging technologies, could potentially double or triple that per-gigawatt revenue.

What makes this possible is that GE Vernova can now offer an end-to-end power solution. The company already provides gas turbines for power generation, transformers and switchgear for grid connection, and energy-management systems. The MV-UPS fills the final gap, giving GE Vernova control over the entire power cycle from grid connection to power delivery within a data center. This integrated approach is rare in the industry; few competitors can offer such comprehensive, on-site power infrastructure.

What Are the Key Components of GE Vernova's Expanded Power Stack?

• Gas Turbines: Generate power on-site, reducing dependence on grid connections and allowing data centers to operate independently.
• Medium-Voltage Uninterruptible Power Supply: Stabilizes electricity flow across large facility sections, preventing demand spikes from affecting the grid or power sources.
• Solid-State Transformers: Enable more compact and efficient power conversion closer to computing racks, allowing precise control of power distribution as AI tasks speed up or slow down.
• Substations and Switchgear: Connect facilities to the power grid and manage electrical distribution throughout the infrastructure.
• Energy-Management Systems: Coordinate power supply with facility demand changes in real time.

The solid-state transformer (SST) deserves particular attention. GE Vernova completed a 5-megawatt SST prototype in July and is scheduled to deliver its first unit to an unnamed hyperscaler customer later in 2026. The SST bridges power generation and computing infrastructure, such as racks based on Nvidia's Blackwell architecture, enabling automatic power adjustments as workloads fluctuate.

When Will This Technology Actually Be Available?

The MV-UPS is expected to ship in mid-2027 and be implemented in its first project later that year. This timeline matters because data center operators are eager for solutions that can reduce grid strain and improve power reliability. GE Vernova's second-quarter 2026 results underscore the urgency: the company reported $24.2 billion in orders and a backlog of roughly $176 billion, with its Electrification segment alone accounting for $40.6 billion of that backlog. The company's gas-equipment backlog grew by 16 gigawatts to 116 gigawatts, with expectations to reach at least 125 gigawatts by year's end.

How Does This Fit Into the Broader AI Infrastructure Boom?

GE Vernova's power innovations arrive at a critical moment. Nvidia reported in its second-quarter fiscal 2027 earnings that data center revenue reached $89.02 billion, up 117 percent year-over-year, with hyperscale customers alone contributing $48.71 billion. The company is also expanding beyond graphics processing units (GPUs) into central processing units (CPUs), networking, and storage processing, positioning itself as a reference architecture provider for AI factories. Nvidia's procurement commitments jumped to $279 billion from $119 billion the previous quarter, mainly for memory components, signaling that AI infrastructure demand will remain elevated for years.

This sustained demand for computing power directly translates to sustained demand for reliable, efficient power infrastructure. Data center operators cannot deploy more AI accelerators without solving the power problem first. GE Vernova's integrated approach addresses that constraint by offering hyperscalers and data center operators a way to bypass years of waiting for grid connections and instead deploy self-contained, on-site power systems that can scale with their computing needs.

What About the Energy Efficiency Challenge?

While GE Vernova focuses on power delivery and stability, researchers at Boise State University are tackling the energy consumption problem from a different angle. Assistant Professor Omiya Hassan received nearly $500,000 from the National Science Foundation to develop a three-year project called "3D Integrated Parallel Fabrics enabling Layered Opto-electronic Processors for Near-memory AI Computing" (3DPFLOPS). The project aims to reduce the energy AI systems consume by minimizing data movement between memory and processors.

"Electronics are ideal for the precise calculations needed during AI training and complex computations, while photonics excel at moving vast amounts of data quickly during inference," explained Omiya Hassan, assistant professor in the College of Engineering's Department of Electrical and Computer Engineering at Boise State University.

Omiya Hassan, Assistant Professor, Department of Electrical and Computer Engineering, Boise State University

Hassan's team is combining light-based computing (photonics) with 3D-stacked chip architecture to shorten the distance data must travel. This "near-memory computing" approach could reduce energy consumption for AI inference, the process of running trained models to generate responses. If successful, the technology could lower energy costs per query, speed up AI response times, and reduce the environmental footprint of data centers.

The research also has implications beyond data centers. The same efficiency gains could allow smartphones, wearables, and other devices to run more AI features locally while preserving battery life and reducing dependence on distant, power-hungry servers.

How to Understand the Data Center Power Challenge

• The Volume Problem: AI data centers require enormous amounts of electricity, with demand growing faster than traditional power infrastructure can supply, forcing operators to seek alternative solutions like on-site gas turbines and renewable energy partnerships.
• The Stability Problem: Even brief power interruptions can crash AI workloads, and sudden demand spikes strain both local grids and on-site generation, requiring specialized equipment like uninterruptible power supplies to maintain continuous, high-quality power.
• The Efficiency Problem: Moving data between memory and processors consumes significant energy, so researchers are exploring photonics and 3D chip architecture to reduce this "data movement tax" and lower overall system power consumption.
• The Integration Problem: Most power infrastructure components are sold separately, forcing data center operators to coordinate multiple vendors; integrated solutions like GE Vernova's end-to-end stack simplify deployment and improve efficiency.

The convergence of these challenges is driving innovation across the industry. GE Vernova's MV-UPS and solid-state transformer address the stability and integration problems by offering hyperscalers a comprehensive power solution. Meanwhile, research projects like 3DPFLOPS tackle the efficiency problem by rethinking chip architecture itself. Together, these developments suggest that solving AI's power crisis will require innovation not just in power generation and delivery, but also in how chips are designed and how data moves through computing systems.