As data centers increase in scale, density, and energy intensity, airflow and thermal behavior have become primary design risks rather than secondary checks. Outdoor equipment yards, generator installations, and high-density interior spaces are increasingly constrained by site conditions, architectural features, and reliability requirements that cannot be adequately evaluated using simplified calculations or rules of thumb.

This presentation focuses on the practical application of Computational Fluid Dynamics (CFD) in real data center projects, with an emphasis on how CFD is used as a design support tool. The session will outline a typical CFD workflow used in practice, including site specific weather and wind assessment, geometry development for buildings and equipment yards, detailed equipment definition based on manufacturer data, computational domain setup, and scenario based simulations under critical operating conditions.

Both external CFD (chiller yards, generator yards, exhaust recirculation, cross-contamination, enclosure and screening effects) and internal CFD (data halls, electrical rooms, and battery rooms) will be discussed, highlighting the types of questions each analysis is intended to answer and where CFD adds value beyond conventional mechanical design methods. The presentation will illustrate common failure modes - such as hot air recirculation, pressure buildup, insufficient makeup air, and thermal stratification, and how CFD results have directly informed equipment layout, spacing, exhaust strategies, and mitigation measures early in design.

The objective of this session is to provide designers and building performance practitioners with clear context for when CFD is warranted, how it is implemented in practice, and how results are translated into actionable design decisions that improve reliability, energy performance, and long term operational resilience in data center facilities.

As AI drives unprecedented hyperscale data center growth, communities increasingly scrutinize environmental impact while operators balance sustainability with mission-critical reliability. Building Energy Modeling (BEM) has emerged as an essential tool for designing and operating sustainable, high-efficiency data centers. It provides data-driven insights that optimize energy performance across the entire facility lifecycle—from informed site selection and energy-efficient design to operational control and decarbonization strategies. This presentation explores how early-stage design decisions and BEM deliver measurable reductions in whole building life cycle carbon and water consumption, advances in green building certifications, while managing community perception—the four pillars defining social license to operate. Drawing on building engineering physics this session moves beyond aspirational narratives to focus on quantifiable differentiators. The future of AI infrastructure depends not just on what we build, but how we build it and how effectively we demonstrate commitment to communities and climate.

The proliferation of data centers and the increased use of AI has necessitated a rapid change in the practice of designing data center mechanical systems from what has historically been a steady state thermal management problem into a dynamic one. This presentation highlights two such situations: short breaks in chiller availability, and load spikes.

The critical maximum temperature limit for cooling water supplied to servers is only a few degrees higher than nominal operating setpoints. When a chiller goes offline for maintenance or experiences a power outage, uninterruptable power supplies keep the pumps moving the facility loop cooling water, but it can take up to 60 seconds for generators to come online and bring the chillers back to capacity. Sufficient cooling loop volume is critical to preventing server supply temperatures outside their range of acceptability. Modeling the dynamics of energy transfer within the cooling loop in the timescale of seconds to assess the impact of thermal energy storage tanks or the modification of piping sizes and layouts to ensure continuous server operation is critical.

Training AIs can cause a server load to increase by orders of magnitude within milliseconds while the valves that maintain consistent supply temperatures can take minutes to modulate from fully closed to open. Dynamic energy modeling on a second-by-second time scale allows examination of supply temperature variations needed to design accommodations for minimizing temperature excursion, allowing valves to modulate on a time scale far greater than that of the loads they serve.

Transient behavior in data center cooling systems spans multiple time scales, from rapid air-side thermal response to slower water-side dynamics and control-driven effects. Accurately capturing these interactions remains challenging, particularly when computational fluid dynamics (CFD) is applied to short-duration events such as equipment transitions, fault scenarios, or control changes.

This presentation examines several transient modeling strategies applicable to data center cooling analysis, including air-side transient CFD, water-side transient CFD, and loosely coupled or boundary-driven approaches. The discussion emphasizes modeling assumptions, time-scale alignment, and the practical implications of each method.

Common sources of discrepancy in transient simulations—such as steady-state initialization artifacts, representation of thermal inertia, numerical time-step selection, and simplified control logic—are highlighted to illustrate both the strengths and limitations of current approaches.

The presentation concludes with practical guidance on selecting an appropriate transient modeling strategy based on study objectives, available data, and acceptable uncertainty. The intent is to clarify what transient CFD can realistically provide in data center applications, and where engineering judgment remains essential.