Modern artificial intelligence (AI) computing environments present power management challenges. Graphics processing unit (GPU) clusters can drastically fluctuate from idle to full load in milliseconds, stressing the infrastructure from batteries to utility connections. Two new unique uninterruptible power supply (UPS) control features enable precise technical solutions for these challenges.
Battery Shield: Protecting energy storage
Standard UPS operations may engage batteries during fast load transitions, resulting in frequent micro-discharges that degrade battery life. The new Battery Shield feature eliminates this problem by managing power steps internally within the UPS.
The technology works by dynamically controlling the DC link voltage, providing sufficient stored energy for transferring upstream fast load steps without engaging the battery system. Laboratory testing confirms the UPS can now absorb 0-100% power steps and mitigate power steps above 100% directly and keep batteries isolated from unnecessary cycling.
This matters because:
- Backup energy remains fully available for actual outages
- Battery lifetime is preserved
- Battery replacement needs are reduced
- System reliability improves
- Multiple battery chemistries can be supported (Lithium ion, valve regulated lead acid or VRLA, and nickel zinc or NiZn)
Input Power Smoothing: Protecting utility and on-site power-generating equipment
The Input Power Smoothing (IPS) feature protects upstream infrastructure by absorbing sharp AI load peaks, facilitating stable and predictable input power flow, and transforming the UPS battery system into an active power buffer. Instead of allowing AI workload fluctuations to propagate upstream to the UPS, the system dynamically absorbs or injects power to the battery to maintain stable input demands.
Technical validation shows:
- Utility and on-site power-generating equipment are protected from sudden power variations
- Reduced thermal and electrical stress on switchgear, transformers, and distribution panels upstream the UPS
- The system adapts to different load profiles
- Generator stability improves significantly
- Grid operator requirements are more easily met
The algorithm continuously calculates average output power and manages fluctuations within a user-defined range.
State of charge (SOC) management
Battery SOC is managed to maintain IPS operating reliably while preserving the battery’s primary role as a backup energy source. The IPS algorithm dynamically regulates charging and discharging cycles within a controlled SOC window. When the generator is active, the algorithm allows for more smoothing on the input to increase the input power stability.
Laboratory validation with real-world AI conditions
Extensive testing with the Vertiv™ AI load simulator verified consistent performance results across multiple scenarios:
- AI load profiles with different load steps
- Various duty cycles and frequencies
- Different battery configurations
- Multiple fluctuation range percentages
Results consistently showed stable input current despite extreme output variations, confirming the system's ability to protect both batteries and upstream infrastructure. These advances represent a significant step forward in power protection for AI environments. Rather than treating batteries as passive backup devices, these controls implement smart control strategies that optimize how the UPS interacts with both the batteries and upstream infrastructure.
The future of AI power protection
As AI workloads continue to grow more demanding, the ability to manage dynamic power patterns becomes increasingly critical for data center reliability. These advanced control innovations demonstrate that existing UPS can be optimized to handle AI's power challenges. These technologies are available now and have been validated across multiple real-world scenarios, offering data center operators a proven solution for stabilizing their AI power infrastructure.
