Data center rack density is moving into territory that would have been exceptional only a few years ago.
Uptime Institute's 2025 survey found an average modal rack density of 7.5 kW when high-density facilities were excluded. Compare that with today's AI infrastructure. An NVIDIA GB200 NVL72 rack consumes approximately 120 kW at full load, while the newer GB300 NVL72 can require up to 142 kW.
That difference is more than a cooling challenge. Every increase in compute density has an electrical infrastructure behind it that must reliably deliver the power, including transformers, switchgear, busways, breakers, PDUs, and the connections between them.
We've previously looked at what high-density data centers mean for electrical safety programs. This time, the focus is on the equipment itself: what happens to the electrical distribution system when substantially more power is concentrated into each rack?
AI infrastructure is changing what a single rack can contain and how much power that rack can demand.
Systems such as NVIDIA's GB200 and GB300 NVL72 combine dozens of GPUs and CPUs into rack-scale computing platforms. Liquid cooling is an important part of what makes this density practical, removing heat directly from high-power components and allowing compute resources to be packaged more densely than traditional air cooling could practically support.
But solving one thermal constraint does not make the electrical demand disappear.
A 120 kW rack still needs 120 kW delivered through the electrical system. As facilities deploy more high-density racks, transformers, switchgear, busways, and distribution capacity must scale with them. In some designs, the power architecture itself is changing, including the industry's emerging move toward 800 VDC data center infrastructure.
That is where AI rack density stops being only a compute or cooling metric and becomes an electrical infrastructure issue.
Switchgear sits directly in the path between incoming electrical power and the equipment consuming it. As data center loads increase, the systems responsible for distributing that power become increasingly important to reliability and electrical safety.
Current flowing through a conductor creates heat. Connections such as bus joints, breaker stabs, cable terminations, and lugs introduce small amounts of electrical resistance. Properly designed and maintained equipment is built to manage its rated loads, but problems develop when resistance increases.
A loose connection, deteriorating contact surface, improper torque, corrosion, or another developing defect can create a localized point of higher resistance. Because resistive heating increases with the square of current, a connection problem that was manageable at a lower load can become much more consequential as loading rises.
This is why rising rack density changes the reliability conversation. Facilities are not simply adding more servers. They are scaling the electrical systems required to support them.
Higher-capacity electrical systems can also affect available fault current, the maximum current that could flow at a particular point during a short circuit. Switchgear is designed and rated to withstand and interrupt specified fault-current levels. Common low-voltage switchgear ratings include 65 kA and 100 kA, with some equipment rated higher.
Why does that matter? During a fault, the electrical system may need to withstand and interrupt enormous current in a very short period. Available fault current is also one input engineers consider when calculating potential arc flash incident energy. Protective-device clearing time, working distance, equipment configuration, voltage, and other variables matter too.
As data centers expand transformers and distribution capacity, those engineering assumptions need to remain current. An upstream system change can alter conditions downstream, which is why designing arc flash risk out of the data center requires looking at the broader electrical system rather than individual components in isolation.
Equipment failure becomes even more consequential when critical electrical infrastructure cannot be replaced quickly.
The transformer market is a clear example. Recent reporting on the U.S. power-equipment market shows that some transformer lead times now exceed 160 weeks as utilities, manufacturers, and data center developers compete for production capacity.
That changes the reliability equation.
If a switchgear connection begins overheating, the immediate priorities are preventing equipment damage and protecting personnel. But if the problem progresses into a failure that damages a transformer or other difficult-to-replace infrastructure, the facility may face more than an emergency maintenance event. It can become a procurement problem measured in years.
The rapid data center boom across the U.S. adds to that pressure as developers and utilities race to expand power capacity. At the same time, continued expansion could strain the skilled trades workforce needed to build and maintain increasingly complex electrical infrastructure.
When replacement equipment and qualified labor are both constrained, maintaining the health of installed equipment becomes an operational priority.
Facilities teams cannot stop AI rack density from increasing, and even well-designed electrical systems still require maintenance and visibility into changing equipment conditions.
The practical question is how to identify a developing problem before it becomes equipment damage.
As we covered in why quarterly thermal inspections aren't enough for AI-era data centers, periodic inspection provides a snapshot of equipment condition. Continuous monitoring can provide additional visibility between those scheduled inspections.
By monitoring critical connection points over time, facilities teams can identify abnormal temperature trends and investigate developing issues. Instead of relying only on conditions captured during a periodic inspection, teams gain information about how equipment is behaving as loads change.
GraceSense™ Hot Spot Monitor (HSM) and HSM 600 solutions provide one example of this approach. Fiber-optic temperature probes installed at critical connection points continuously track temperature conditions, providing additional visibility without requiring the enclosure to be opened simply to collect temperature data.
Continuous monitoring does not replace inspection, maintenance, electrical safety procedures, PPE, or engineering controls. It provides another source of condition data that can support maintenance and reliability decisions.
That visibility becomes increasingly important as facilities are also preparing their data center safety programs for evolving NFPA 70E priorities alongside changes in high-density electrical infrastructure.
The AI density curve is not only changing what happens inside the rack. It is changing what is required from every layer of electrical infrastructure feeding it.
Higher-capacity systems, changing fault-current conditions, heavily loaded connections, long equipment lead times, and limited maintenance windows all increase the consequences of a developing equipment problem going unnoticed.
Careful electrical design, inspection, and preventive maintenance remain essential. But when critical equipment operates continuously and replacement lead times can stretch into years, facilities teams also need visibility into what is happening between scheduled maintenance events.
The goal is not simply to react faster when switchgear fails. It is to identify developing conditions early enough to act before they become an outage, safety event, or long-term procurement problem.
If your facility is evaluating how increasing rack density may be affecting its electrical infrastructure, Request a Data Center Risk Review to take a closer look at your current systems, maintenance strategy, and opportunities to improve visibility into critical electrical assets.
Zero Harm. Zero Downtime.