A new whitepaper, "ABB Power Protection of AI Data Centers," provides useful technical insights for designing resilient, energy-efficient power protection systems tailored to the unique demands of artificial intelligence (AI) workloads and modern hyperscale data centers.
As AI continues to transform industries worldwide, Domagoj Talapko, Business Development Manager, ABB Electrification, discusses the unique power challenges that separate AI applications from traditional data center operations. With higher power densities and the need for greater reliability, advanced UPS systems are ensuring the availability of mission-critical infrastructure.
How are AI data centers different from traditional data centers when it comes to UPS requirements?
When it comes to power requirements, traditional data centers typically operate at power densities of 5-10 kW per rack. AI data centers can operate at 30-50 kW per rack, with some high-performance computing clusters already exceeding 100 kW per rack, with 1 MW on the horizon. They also generate sudden, unpredictable power spikes, which require fast, resilient UPS systems and flexible distribution.
Our teams work closely with data center designers to deliver solutions that can handle these challenges while optimizing energy use. We’re also collaborating with NVIDIA to develop next-generation solutions for AI’s demands.
What are the main challenges that AI poses for data center operators?
There are three technical challenges data centers face. First is reliability—AI training runs can take weeks or months, and a single power interruption can result in massive computational losses and costs.
Second is power density management—concentrating so much power in such small footprints creates thermal and electrical challenges that traditional solutions weren't designed to handle.
Third is scalability—AI infrastructure is growing exponentially, and power protection systems must be able to scale seamlessly without disrupting existing operations. Our work with customers such as Applied Digital and Kevlinx show how collaborating with ABB is delivering innovation to solve these challenges.
How can UPS systems meet AI data center demands?
Modern UPS systems for AI applications must deliver three key capabilities: higher power density in smaller footprints, superior dynamic response to handle rapid load changes, and modular scalability to grow with AI infrastructure. At ABB, our MegaFlex UPS systems are specifically designed for these requirements. For example, our recent 1.5 MW installation demonstrates how modular UPS architecture can provide N+1 redundancy while maintaining energy efficiency above 97%. The systems must also integrate seamlessly with intelligent power distribution and monitoring systems to provide real-time visibility into power quality and consumption patterns—essential for optimizing AI workload management.
Besides the conventional low-voltage approach, innovation in the domain of medium-voltage UPS, such as HiPerGuard, can address the challenges posed by AI workloads, with greater inherent robustness and resilience and efficiency of over 98%.
How can AI facilities balance robust power protection with energy efficiency?
AI data centers consume significant amounts of energy, so every efficiency improvement translates to valuable cost and carbon reductions. Modern UPS systems can achieve efficiency levels of 96-99% in eco-mode or online mode, compared to 85-92% for older technologies.
Beyond the UPS itself, we're seeing intelligent power distribution systems that minimize conversion losses, advanced cooling integration that optimizes the entire thermal management system, and predictive analytics that identify inefficiencies before they impact operations. ABB offers specific strategies for achieving maximum protection and maximum efficiency.
Looking ahead, how should organizations planning new AI data centers approach their power protection strategy?
Start with scalability and flexibility as core design principles. AI technology is evolving rapidly, and your power infrastructure must be able to adapt. Choose modular, scalable solutions that can grow incrementally rather than requiring complete overhauls. Prioritize systems with proven reliability in high-density applications—this isn't the place to experiment with unproven technology. And critically, partner with experienced providers who understand both electrical engineering and the AI application requirements. ABB provides a comprehensive framework for this planning process, including case studies from successful deployments. For further information, check out the whitepaper: "ABB Power Protection of AI Data Centers"
About the author
Domagoj Talapko is a reliability-focused electrical engineer with over 18 years of experience in power systems, infrastructure design, and mission-critical operations. He holds a PhD in Reliability Engineering and leads global business development at ABB, driving innovation in high-power UPS systems and energy continuity solutions for AI-optimized data centers.
