Artificial intelligence is rewriting the rules for DC operations in the United States. Mid-size and large facilities, long focused on achieving near-perfect uptime, now face unprecedented stress on power and cooling systems. As AI workloads proliferate, they demand not only more raw energy but also more sophisticated approaches to keeping servers stable, efficient, and cool. Meeting these demands is not simply a matter of adding redundancy. It requires a holistic, integrated strategy that considers everything from rack layout to thermal management to long-term operational discipline.

The push for uptime beyond 99.99 percent has always been a guiding principle in the industry, but AI is making that goal harder to reach. Today’s processors and accelerators draw power at levels unimaginable a decade ago. Where server CPUs once drew under 100 watts a decade ago, today’s models routinely exceed 200 watts, and many high-performance servers average 500 watts or more. When clustered together, AI systems often consume five times the power of traditional server environments, yet they occupy the same footprint on the data center floor. This combination of density and intensity magnifies risks and requires operators to think differently about infrastructure.

Power: The First Battleground

It’s not enough to install large UPS systems. AI servers create rapid fluctuations in demand, making power quality as important as capacity. Even brief voltage spikes or irregularities can interrupt machine learning training, a process that may run continuously for weeks. To prevent such setbacks, power infrastructure must deliver both precision and flexibility. Every component, from the UPS to the distribution units, must be engineered to handle sudden shifts while feeding real-time data into monitoring systems. Visibility down to the health of fans, drives, and cabling helps identify problems before they cause outages. Still, equipment alone does not solve the problem. Many failures stem from human factors: mislabeled racks, poorly executed maintenance, and overlooked diagnostics. These are not headline-grabbing mistakes, but they are common and costly. The only way to avoid them is through meticulous planning, strict operational protocols, and thorough staff training. Resilience comes not from reacting to failure but from anticipating it.

Cooling presents an equally pressing challenge. The thermodynamics are simple: more energy in means more heat out. Traditional air-cooling methods, once sufficient for enterprise servers, are now straining under the load of AI hardware. Air simply cannot carry away the heat produced by processors packed into dense racks running at extraordinary power levels. As a result, the industry is shifting toward liquid cooling. Direct-to-chip cooling routes liquid through pipes connected to processors and memory modules, efficiently transferring heat. Two-phase systems go further by allowing the liquid to evaporate, absorbing more energy. Even so, liquid cooling alone is rarely enough; most facilities rely on hybrid setups that combine liquid cooling with supplemental airflow to handle residual heat.

Some operators are also experimenting with immersion cooling, where entire servers are submerged in non-conductive fluids. This method offers unparalleled thermal efficiency but requires radical redesigns of equipment layout, floor planning, and maintenance procedures. Handling a heavy, liquid-coated circuit board is no small task, and immersion systems call for new infrastructure and staff expertise. Despite these hurdles, the efficiency gains are hard to ignore. Studies suggest liquid cooling can reduce energy consumption by as much as 90 percent compared to air cooling, while improving Power Usage Effectiveness and lowering operating costs. For data centers with sustainability goals, it represents a viable long-term strategy—if they are prepared for the investment in infrastructure and skills.

The overarching challenge for DCs is finding the right balance. The cost of downtime, particularly in AI-driven industries such as healthcare, finance, and autonomous systems, is simply too high to accept preventable risk. At the same time, overdesigning a facility with excessive backup systems and unneeded capacity creates inefficiencies that inflate total cost of ownership. The solution lies in an integrated infrastructure approach, where power, cooling, monitoring, and maintenance are planned as interdependent layers of the same strategy. This holistic model ensures that resilience is baked in from the ground up, while also keeping scalability and cost control in view.

Ultimately, resilience in the age of AI is not about overbuilding or reacting to crises. It is about readiness. By adopting integrated strategies, investing in smarter power and cooling systems, and cultivating disciplined operations, U.S. data centers can meet today’s surging demand while preparing for the even greater challenges ahead.

# # #

About the Author

Carsten Ludwig is the Data Center Market Manager at R&M.