TerraFlow and DG Matrix Pair Flow Battery With Solid-State Transformer for Live AI Compute

TerraFlow Energy and DG Matrix announced on September 28 a commercial agreement to deploy a VRFB solid-state transformer system designed to power a live high-performance computing cluster in the United States. The initial installation will combine TerraFlow’s long-duration uninterruptible power system with DG Matrix’s programmable Interport solid-state transformer and Dell PowerEdge servers, creating a test of how long-duration storage and software-defined power conversion can buffer rapidly changing AI computing loads.

The companies have not disclosed the deployment location, total megawatt load, battery energy capacity, project cost or commercial-operation date. That makes the announcement a technology-deployment milestone rather than evidence that the architecture has already proven performance at hyperscale. The central claim is narrower: the partners intend to put the combined system against live compute loads and use it to manage power flow between the grid, on-site generation, storage and servers in real time.

VRFB solid-state transformer system targets fast-changing compute loads

In their September 28 announcement, TerraFlow and DG Matrix said the architecture is intended to absorb rapid changes in computing demand before those changes propagate upstream into electrical infrastructure. TerraFlow’s LDUPS platform combines uninterruptible-power functionality with hours of energy storage, while DG Matrix’s Interport platform performs programmable power conversion and routing among sources, storage and compute loads.

The approach addresses a problem that is becoming more important as AI data centers adopt denser computing hardware. Power demand inside an AI facility can change quickly as large GPU clusters move between workload states. Those variations do not automatically translate into a grid reliability event, but sufficiently large and synchronized load changes can complicate planning, protection and operational coordination.

The North American Electric Reliability Corporation is already developing dedicated standards for computational loads. NERC’s Project 2026-02 identifies reduced visibility, system-instability risk during disturbances and planning uncertainty among the issues associated with rapidly growing data-center and other computational loads.

Long-duration flow storage adds a different role than conventional UPS batteries

TerraFlow describes its LDUPS architecture as a long-duration system rather than a conventional short-ride-through UPS. The company’s LDUPS platform supports vanadium or organic electrolyte chemistries and is designed in modular 1.25 MW building blocks with scalable energy duration. In the announced deployment, the vanadium redox flow battery is expected to provide both fast response and multi-hour energy capacity.

That combination is important because data-center power systems normally separate several functions. UPS equipment handles very short interruptions and power-quality events, backup generators or other on-site generation provide longer-duration support, and grid-connected batteries may perform peak shaving or other energy-management tasks. TerraFlow and DG Matrix are proposing to coordinate several of those functions through a common power architecture.

Microgrid Media has previously examined solid-state transformer testing at UC San Diego, where a California Energy Commission-funded demonstration is designed to connect a 2 MW AI computing load through medium-voltage power electronics and flexible-load controls. The TerraFlow-DG Matrix project is distinct, but both efforts reflect growing interest in replacing static data-center electrical architectures with systems that can actively control how computing loads interact with the grid.

DG Matrix is also increasing the power density of its Interport platform

DG Matrix said on September 25 that its Interport platform had increased output from 200 kW to 400 kW using silicon-carbide power electronics while maintaining essentially the same power-module footprint. That product update does not establish the size of the TerraFlow deployment, but it provides context for the platform being integrated with the flow-battery system.

For the initial project, the companies have not disclosed how many Interport units will be deployed or the voltage architecture feeding the Dell servers. They also have not published a target for round-trip efficiency, load-ramp smoothing, outage duration, grid-services revenue or the amount of utility capacity the system could defer.

Those omissions matter because claims about making data centers more flexible need operating evidence. Microgrid Media’s coverage of NERC’s computational-load reliability work has highlighted the need for better modeling, commissioning, telemetry and coordination as large computing facilities connect to the bulk power system.

The commercial test will matter more than the first-of-a-kind label

TerraFlow and DG Matrix describe the deployment as the first integrated U.S. architecture combining a vanadium redox flow battery with a solid-state transformer to directly power a live high-performance computing cluster. That characterization is a company claim and should be treated as such until independently established across the market.

The more consequential question is whether the system can demonstrate measurable performance under real AI workloads. Useful results would include how quickly storage responds to workload transients, how much upstream load variability is reduced, how efficiently power moves through the architecture, and whether the battery can provide meaningful energy duration without compromising UPS-level response.

Data-center developers are already exploring batteries, on-site generation and grid-forming systems as alternatives to waiting solely on conventional utility expansion. Microgrid Media has reported on integrated battery and behind-the-meter power systems in Texas that are similarly designed around faster time-to-power for hyperscale computing.

The VRFB solid-state transformer system announced September 28 moves that trend toward a more tightly integrated electrical architecture. The next milestone will be disclosure of the installation’s scale, location and operating data showing whether the combined flow battery and solid-state transformer can actually reduce the volatility presented to the grid while maintaining the power quality demanded by live AI compute.