Battery Systems Tackle AI Data Center Power Spikes

## AI's Power Thirst Meets Its Match in Advanced Battery Systems The rapid, unpredictable power demands of artificial intelligence are pushing traditional data center infrastructure to its limits, creating a critical need for more responsive energy solutions. AI training models can cause power consumption to fluctuate by hundreds of megawatts in seconds, a "sawtooth" pattern that legacy power grids and backup systems were not designed to handle. This volatility can lead to equipment strain, premature failure, and costly downtime. Traditional data center backup power, heavily reliant on uninterruptible power supplies (UPS) and diesel generators, is ill-suited for the dynamic loads of AI. UPS systems typically offer only minutes of power, intended to bridge the gap until generators activate. However, these constant, small discharges from frequent power spikes can significantly shorten the service life of UPS batteries. Battery energy storage systems (BESS) are emerging as a key solution, capable of absorbing and discharging energy rapidly to smooth out the volatile power demands of AI workloads. Unlike traditional backups, BESS can provide sustained power, reducing reliance on diesel generators and enabling "peak shaving" to lower energy costs by drawing from the grid during off-peak hours. Ampace is targeting these challenges with its PU200 battery cabinet, which utilizes semi-solid-state lithium iron phosphate (LFP) cells. This technology allows for a high 10C discharge rate, meaning it can release its energy capacity quickly to meet sudden demand spikes. The system is designed for a 15-year lifespan, aiming to reduce the total cost of ownership compared to batteries with shorter life cycles. To further enhance power delivery, Ampace has partnered with DG Matrix to integrate its BESS with a multi-port solid-state transformer (SST) platform. This collaboration aims to create a "grid-active" system that provides real-time load smoothing and UPS-grade protection, aligning with infrastructure guidelines from major industry players like NVIDIA. The market for data center energy storage is drawing significant competition from established industrial technology companies. Vertiv's EnergyCore Grid system uses lithium iron phosphate (LFP) batteries and is designed for large-scale applications, scaling from 1 MW to over 200 MW. Similarly, Eaton offers lithium-ion battery systems, including its Samsung Gen 3, which is also aimed at reducing the total cost of ownership in data centers. The integration of BESS offers a significant return on investment by turning a data center's power system from a passive consumer into an active grid asset. By storing energy when prices are low and selling it back to the grid during peak demand, or providing grid stabilization services, data centers can create new revenue streams. This capability is crucial as data centers are projected to account for 5% of global electricity consumption by 2025. Innovations in battery technology are central to meeting the evolving demands of AI. Ampace's semi-solid-state technology, for example, is designed to reduce the risk of electrolyte leakage and thermal runaway, enhancing safety. This focus on safety and efficiency is critical as power densities in AI racks can be more than five times higher than in traditional data centers.