Applications: Data Centers
Pages 233–242 (10 pages)
Text on page 241
BESS Design Configurations for AI Data Centers BESS enables differentiated energy storage solutions for AI training and inference, each with distinct operational requirements
BESS Design Parameter
Training Campuses
Inference Campuses
Power Tolerance
Tolerates brief interruptions; requires high instantaneous power delivery
Requires 99.999% uptime; zero-tolerance for interruptions;
Scale & Burst Capacity
BESS can supply multi-MW to GW-scale bursts and manage ramping when large training jobs start or pause; 4060% load swings common
Sustained high load; 24 hour backup for always-on fleets
Response Time
Millisecond ramping management; handles computational bursts within seconds
Sub-50ms cutover; BESS is more tightly integrated with UPS and grid services to ride through faults and provide fast frequency response.
Primary Function
Co-located BESS can support “power-firstˮ sitting near renewables or nuclear, with batteries smoothing variability and supporting microgrid-style operation.; support flexible interconnection
BESS plus smart control can arbitrage time-of-use prices for always-on inference fleets, reducing operating cost without compromising SLAs
Source: Terakraft, Flexgen
Extracted from the PDF; charts and images may read imperfectly.
Page 241 · 9/10