UPS that doesn't degrade under AI workloads

Your questions, answered.

• How does a flywheel-based UPS handle the load profile of AI-dense data centers differently than chemical batteries?

  AI training workloads create sharp, repetitive power cycles that accelerate chemical degradation in lithium-ion cells — compressing replacement intervals to 3–5 years under heavy cycling. A flywheel stores energy kinetically, not chemically. It absorbs the same sharp cycles without degradation, and its 10C discharge rating provides the burst power that high-density GPU racks demand during peak draw.

• Does Torus maintain grid synchronization during a voltage event, or does it disconnect and transfer?

  Torus provides voltage correction while maintaining grid synchronization — a make-before-break approach. The system corrects voltage disturbances without disconnecting from the grid, which is a requirement under emerging regulations like Texas SB6 for large data center loads.

• How does Torus UPS fit into an existing N+1 or 2N redundancy architecture?

  Torus integrates into existing redundancy frameworks. The system connects at the facility level or at individual circuits, depending on the architecture, and complements existing generators, transfer switches, rack-level UPS units, and distribution infrastructure.

• Why are traditional UPS batteries degrading faster in AI-dense data centers?

  AI workloads draw power in sharp, repetitive cycles at power densities far beyond what conventional UPS batteries were designed for. Each sharp cycle accelerates chemical degradation. The result is capacity that drops well before scheduled replacement, creating unplanned capital expense and maintenance windows.

• What is voltage correction, and why are regulators requiring it for large data centers?

  Voltage correction is the ability to maintain continuous power delivery during a grid voltage disturbance — without disconnecting and transferring to stored energy. Texas SB6 and similar emerging frameworks require large data centers to support grid stability during stress events rather than dropping off the grid when voltage sags. This requires infrastructure that injects or absorbs reactive power while remaining grid-connected.