TerraFlow and DG Matrix pair storage with compute

TerraFlow Energy and DG Matrix have agreed to deploy a vanadium redox flow battery and solid-state transformer architecture against a live high-performance computing load in the United States, putting a new data-centre power arrangement into operation rather than limiting it to laboratory testing.

The initial deployment will combine TerraFlow’s LDUPS long-duration uninterruptible power system with DG Matrix’s Interport solid-state transformer technology and Dell PowerEdge servers. The companies have not disclosed the location, MW rating, MWh capacity, customer, or commissioning date, so the announcement establishes the architecture and commercial agreement without yet defining the scale of the first installation.

The electrical problem is straightforward even if the proposed solution is not. High-performance processors can change power demand quickly as workloads move between idle and intensive operation, creating steep load ramps that pass through UPS equipment, transformers, switchgear, and the grid connection unless local power systems absorb them first.

TerraFlow’s approach places long-duration storage directly in that chain. Its LDUPS platform is based on flow-battery technology in which liquid electrolyte is stored separately from the electrochemical stack. The company publishes modular power blocks from 1.25MW to 25MW and energy durations from four to more than 48 hours for the wider platform, but those values should not be read as specifications for the undisclosed first deployment.

Flow batteries separate stored energy from power output more clearly than many packaged lithium-ion systems because increasing electrolyte volume can extend duration without increasing the electrochemical stack in the same proportion. That can suit long-duration duty, although tanks, pumps, piping, climate control, auxiliary loads, and a larger physical footprint create a different balance-of-plant requirement.

TerraFlow publishes a nominal response time below five milliseconds for the LDUPS platform and positions it as a continuously active power-management layer rather than equipment that waits for a utility failure. In a data-centre application, that creates the possibility of smoothing repeated compute-driven changes during normal operation while retaining stored energy for longer disturbances.

DG Matrix provides the power conversion and routing layer through Interport. A solid-state transformer uses power electronics rather than relying only on a conventional line-frequency magnetic transformer, allowing voltage conversion and power-flow control to be managed electronically. The platform is designed to connect the grid, on-site generation, storage, and compute loads through a programmable architecture.

Interport’s power module was increased from 200kW to 400kW in September using silicon carbide devices from STMicroelectronics while retaining broadly the same module footprint. Higher switching performance and power density matter in data-centre applications because electrical rooms and distribution equipment compete for space with the computing infrastructure that generates revenue.

The combined system is intended to absorb rapid changes in server demand before they are passed upstream. Storage can provide or accept power during a load ramp, while the solid-state transformer controls how that change is presented to the facility distribution system and the grid connection. The result, if the controls perform as intended, would be a more predictable electrical load at the boundary between the data centre and its supplier.

That differs from a conventional UPS design whose primary task is to maintain power during an interruption until standby generation starts or another source takes over. TerraFlow and DG Matrix are proposing a system that remains active during ordinary operation, using storage and programmable conversion to manage power quality and load behaviour as well as backup requirements.

The trade-offs will be determined by efficiency, capital cost, footprint, service requirements, and utilisation. Every conversion stage introduces losses, while long-duration flow storage occupies substantial physical space compared with short-duration lithium-ion systems. Those costs have to be justified by reduced grid volatility, improved resilience, faster connection, or the ability to use the battery for several functions rather than leaving it idle for most of its life.

Control integration will be central to the field trial. Server loads are driven by software and workloads rather than by the needs of the electrical system, so the power architecture has to react without disrupting compute performance. The storage controller and solid-state transformer must also respect battery state of charge, converter limits, protection settings, and any constraints imposed at the grid connection.

The first installation will therefore provide more useful evidence than the headline claim that the architecture is new. Measured response time, conversion efficiency, thermal performance, battery utilisation, and the shape of the power seen by the upstream network will determine whether the concept offers an engineering advantage over conventional UPS and distribution arrangements.

The commercial agreement is also intended to provide a framework for later deployments. Scaling beyond one site will require the companies to show that the architecture can be engineered around different utility voltages, server configurations, redundancy requirements, and data-centre operating models without turning each installation into a bespoke research project.

For now, the undisclosed project size keeps the emphasis on technical validation. TerraFlow and DG Matrix have defined the components and the operating objective; the next material milestone is a live installation producing enough operating data to show whether long-duration flow storage and solid-state conversion can make a rapidly varying compute load easier for the wider power system to serve.