Designing for the AI Era: The New Realities of Data Center Infrastructure

AI acceleration is redefining first principles

AI training and inference have raised the stakes for how fast we can build and how reliably we can operate. The result is a step‑change in expectations for thermal performance, electrical architectures, deployment velocity, and resiliency. Our industry’s response will shape the next wave of capacity and the sustainability and community impact of facilities located near where people live and work.

At ebm‑papst, we see three themes converging: hybrid cooling as the operating norm, HVDC readiness in new builds, and resilient, local‑for‑local supply chains that keep projects on schedule despite global volatility. No matter the theme, the mandate is the same: upfront design for efficiency, acoustics, interoperability, and redundancy.

Hybrid cooling is here to stay

Rack-level liquid cooling is becoming unavoidable for modern chip thermal loads; air alone cannot remove that heat density. Yet liquid does not eliminate the need for air movement in the data center. Airflow management around enclosures, corridors, and supporting systems remains critical to equipment reliability and serviceability. Facilities maintain cool temperatures inside the data center by rejecting heat outdoors, commonly through air cooled chillers, cooling towers, or dry coolers.

What that means for design teams:

• Plan for coexistence: Expect rack-level liquid cooling and air throughout the room, coordinated by controls that understand both domains.
• Engineer for acoustics: As campuses move closer to communities, noise abatement for outdoor equipment will become a siting and permitting issue.
• Target every watt: Efficiency is capacity. Watts saved in cooling translate into watts available for computing, especially in power‑constrained markets.

Power density keeps rising without bigger footprints

Computing density continues to rise, while real estate and retrofit constraints remain tight. Operators are asking for more airflow and higher static pressure in the same mechanical footprint. That challenges the entire system, from fan and motor selection to controls and mechanical integration. The winning designs will deliver higher power density without sacrificing efficiency or acoustics.

Practical implications:

• Specify equipment that can scale output within fixed geometry while maintaining high wire‑to‑air efficiency.
• Use model‑based design and digital twins to simulate thermal interactions across the room and on the roof line before steel is cut.
• Treat controls as a first‑order lever: smarter modulation beats brute force.

Aligning electrical and thermal strategies for HVDC

While retrofits to high‑voltage DC (HVDC) are often impractical for existing sites, new builds increasingly evaluate HVDC‑ready architectures (frequently discussed around ~800 VDC). Higher voltage means lower current for the same power, which can reduce copper in bus bars and conductors and lower the cost. HVDC distribution can also simplify power quality, reducing harmonic mitigation hardware on sensitive circuits.

For thermal systems, this shift is more than an electrical detail. It affects drive selection, controls, and EMI/EMC considerations across fans, pumps, and outdoor equipment. Designing cooling assets to be compatible with HVDC ecosystems helps future‑proof facilities as the powertrain evolves.

Interoperability first: BMS and protocol pluralism

No two owners standardize their operational technology stacks in the same way. To stay deployable, cooling equipment should be BMS‑agnostic, supporting a wide set of protocols, including Modbus, BACnet, PROFINET, and others. This “interoperability-by-default” mindset reduces integration friction and shortens commissioning timelines, especially on fast‑tracked AI builds.

Couple that with on‑premises digital twins that mirror plant behavior, and operators can test control strategies in software, apply AI‑guided optimization, and roll successful tactics into production with confidence. We’ve consistently seen strong alignment between simulated and realized efficiency gains when these models are calibrated and maintained.

Supply‑chain resilience is a design decision

Localization is more than a procurement strategy; it’s a project‑schedule strategy. A local‑for‑local manufacturing model (Americas, Europe, Asia) lowers tariff exposure, reduces transit time, and provides redundant regional capacity when disruptions strike. In a market where schedules are measured in weeks, not quarters, the difference between six weeks of ocean freight and a domestic lead time can define a project’s critical path.

The most effective operators share multi‑year demand signals. Two‑to‑three‑year forecast corridors, supplemented by longer‑range directional views, enable upstream suppliers to secure critical components and expand capacity ahead of demand. That collaboration is how the industry turns macro uncertainty into dependable delivery.

The grid constraint is a catalyst for better designs

Many U.S. metros face power availability constraints that will shape site selection, phasing, and even micro‑siting. Some operators are assessing on‑site generation (including low‑ or zero‑carbon options) while advocating for utility‑led capacity expansions. Regardless of which path wins in each market, the mechanical plant must be ultra‑efficient, acoustically responsible, and ready to ramp in step with staged electrical availability.

What we’re working toward as an industry

• Hybrid cooling that’s quiet, efficient, and serviceable at scale.
• HVDC‑ready cooling assets that integrate cleanly into evolving powertrains.
• Interoperable controls that plug into any owner’s BMS ecosystem.
• Digital‑twin‑driven operations that turn models into measurable savings.
• Local‑for‑local resiliency with redundant regional manufacturing and long‑range planning.
• Community‑minded design that respects neighbors while unlocking compute growth.

This is a collective effort. The more we, as owners, designers, integrators, and suppliers, align early, the faster we can deliver capacity that is efficient, resilient, and responsible.