The U.S. is currently experiencing a boom in data center construction, with most new centers concentrated in just a few states. At these data centers, equipment is running hotter and drawing more power than ever before. This increased power usage has major implications not only for local electrical grids, but also for the safety of those working with highly energized equipment at these facilities.
Three forces are converging inside modern data centers. Rack densities have climbed from the single-digit kW range a decade ago to 40–80 kW for AI workloads. Some new builds are targeting even higher power consumption. Uptime pressure has tightened to the point that every minute of unplanned downtime carries significant cost. And the equipment doing the work is more compact and more energized than ever. Together, these shifts are changing how data centers approach electrical safety and maintenance, especially for the technicians who service this equipment.
A new report by Grace Technologies investigates where data center growth is happening, the amount and cost of electricity they’re using, and what this means for electrical safety in the data center industry.
Grace Technologies’ new report, Electrical Safety in Data Centers: What’s Changed and Why It Matters links this rising density to a noticeable evolution in safety practices:
There are between 4,000 and 5,400 data centers in the US, depending on the source. Electric Choice estimates that these facilities now consume roughly 200 TWh of electricity per year, about 4-5% of total U.S. electricity consumption.
Wherever load concentrates, the work of servicing energized equipment concentrates with it. That has direct implications for the electrical safety practices inside these facilities, which we return to below.
When we organize these data centers by state and the amount of power they consume, three patterns emerge:
Now that we’ve covered the load on the power grid in each state, the next step is to look at how expensive electricity is in places with lots of data centers.
Electricity prices vary by more than four-to-one between the cheapest and most expensive states:
When we compare this with our data from before, we can identify three distinct groups:
Much of the data center boom is driven by AI and other GPU-heavy tasks. Industry analyses show that this is leading to higher rack densities, the amount of power used by a single server rack.
At lower rack densities, the consequences of an electrical incident during maintenance were severe but contained. At 40 to 80 kW per rack, the math shifts. Higher fault current means more incident energy released during an arc flash event, which translates to heavier required PPE, larger arc flash boundaries, and longer recovery times when something goes wrong. The exposure profile of energized work is changing along with the equipment.
This is part of what is pushing the industry toward closed-door diagnostics. The 2024 edition of NFPA 70E reinforced that elimination and substitution sit above administrative controls and PPE in the hierarchy of controls. In practice, that means designing workflows so technicians do not have to open an energized panel in the first place. Permanently installed test points, thru-door voltage indicators, and remote verification methods let workers confirm conditions without removing barriers between themselves and live equipment.
Predictive monitoring extends the same logic across time. Continuous thermal and electrical monitoring tools watch for the early signals of degradation, including loose connections, overheating components, and abnormal load patterns. The goal is to catch issues before they require emergency intervention. For high-density AI workloads, where a single rack can draw more power than entire legacy aisles, the cost of missing an early warning is significantly higher than it was a decade ago.
These safety trends have various implications for the data center industry:
This study combines three categories of evidence to connect macro growth in data center development with state-level concentration, electricity-cost dispersion, and the operational implications for safety and maintenance. It provides a consistent, state-comparable view of where load is concentrating and what the power-cost environment looks like in those markets.
We used the data center line item from the U.S. Census Bureau’s “Value of Private Construction Put in Place” series, which provides a monthly national time series. Values were cleaned into a date-indexed series and used to describe the scale and persistence of the construction ramp from the mid-2010s through 2025. This series supports the macro claim that data center building has become structurally larger and more continuous than in prior cycles.
State-level facility counts were aggregated from a Data Center Map dataset and paired with state-level annual consumption estimates in TWh from Electric Choice. Where multiple published facility-count sources differ, the analysis treats facility counts as a footprint proxy rather than a definitive enumeration, and it uses TWh estimates to represent operational scale more directly. Derived metrics such as TWh per facility were computed to identify states where load appears concentrated into fewer, larger, or more power-dense sites.
Electricity price data was sourced from the US Energy Information Administration state-level “All Sectors” price series. November 2025 was used as the primary snapshot to represent the current power-cost environment, with November 2024 included to provide year-over-year context. Prices were cleaned to a consistent numeric format in cents per kWh and joined to the state facility/load table by state name.
We joined facility counts, estimated annual consumption in TWh, and electricity prices into a single state-level dataset. We then computed simple derived fields to support interpretation and visualization, including:
The results were interpreted using comparative rankings and a “money chart” scatterplot framework: x-axis electricity price, y-axis estimated annual TWh, bubble size facility count. This provides a compact way to distinguish high-load hubs, high-price strategic markets, and lower-price emerging markets, and to tie those segments back to the operational argument that higher density increases the payoff from safer access practices and predictive monitoring.
As rack densities scale, traditional safety protocols are no longer enough to protect your team and uptime. High-density power demands a shift to engineered solutions like continuous monitoring and closed-door diagnostics.
To explore these challenges further, watch the full webinar recording on demand—High Density, High Stakes: Electrical Safety in the Modern Data Center—featuring Jim Phillips, P.E., BScEE, IEEE Life Sr. Member, MIET, P.E. of Brainfiller and Grace Technologies CEO Drew Allen.
Zero Harm. Zero Downtime.