Data centers have proliferated on the commercial power landscape, and their load requests — from several hundred megawatts to 1 or 2 gigawatts — are exponentially larger than anything previously seen, compounding challenges in the U.S. power industry.
Utilities already were facing a demand boom from myriad elements before traditional data centers began surging across the United States in recent years. New artificial intelligence (AI) data centers and their power demand five to 10 times greater than traditional data centers are compounding the data center power demand boom.
The Inflation Reduction Act (IRA) has helped stoke manufacturing demand, and electrification is accelerating across the country. Population shifts also are causing a significant rise in demand. Continued shutdowns of coal-fired power plants in certain U.S. areas is lessening capacity. Add in the massive load requests from the data center sector, and utilities have their hands full trying to meet the demand and accurately forecast future load.
Black & Veatch’s 2024 Electric Report — based on expert analyses of survey responses from nearly 700 U.S. power sector stakeholders — puts it into sharper focus, beginning with the unease in the power sector when it comes to accommodating data centers.
Forty-five percent of respondents say they have no confidence or are not very confident in their forecasting for data center loads, perhaps due to the large amount of uncertainty that can come with data center power requests and expectations (Figure 7). Industry professionals say the load requests often are wide-ranging and have an uncertain timeline — other than fast.
Such lack of certainty was evident when respondents were asked what information from data centers made planning challenging. Nearly one-third (30 percent) pointed to accurate load ramps, while another 18 percent said accurate substation interconnect dates would be helpful (Figure 8). The numbers here paint a picture of contrasting accuracy, with one group of respondents asking for full load capacity on day one with peak demand only being a fraction of that, while increasing demand comes years later and creates grid operation challenges as well as forecasting challenges.
Data centers used to have a specific location in mind — along with the size of the proposed data center, based on the clients wanting data space — when they approached utilities. But with the uncertainty of AI data center demand and their much larger power requirements, existing power supply is strained, and the equation has flipped.
Data centers are reporting that if they find the power or build the power generation themselves, then they’ll just build on the open market and plan for flexibility and expansion, due to the seemingly insatiable appetite for traditional and AI data center space. This does not bode well for utilities that must go before their local public utility commissions (PUC) and justify new power generation to meet the demands.
When asked what data centers can do to improve the speed of their project getting completed, 40 percent unsurprisingly pointed to load clarity (Figure 9). An additional three in 10 cited some form of advance payment, which would help solidify proposals and somewhat pacify the budget question in that its already paid for. Nearly one in five (19 percent) mentioned that client-furnished equipment is helpful — a method data centers are using to advance their projects. Specifically, many are building their own substations or contracting them to an engineering, procurement and construction (EPC) provider, a strategy that can cut the project construction timeline by one-half or even more.
But this tactic isn’t a given, as some utilities do not allow clients to build their own substations — or if they do, they don’t allow the data center to operate that energy asset. On that topic, one-fifth — 21 percent — of respondents said they do not allow clients to build substations, citing their desire for component standardization and their concerns about client capabilities. Eleven percent permit the construction of substations but not the operation, while 27 percent said they allow both the building and operation of the substation by the client. Working with an EPC familiar with their standards could expedite development.
In looking at the biggest challenges utilities face in getting data centers operational, nearly four in 10 (39 percent) cited available power while 30 percent pointed to necessary substation upgrades. Interestingly, almost one in four (24 percent) stated equipment availability was a major concern (Figure 10). In many cases, the wait times for circuit breakers, switchgears, transformers and gas turbines could be three years or more, something entities building substations or their own power generation must consider.
Clearly, data center appetite and demand for power exceeds what utilities can currently provide, both from a capacity and transmission standpoint. In most cases, what the data center industry is asking for doesn’t exist — it needs to be built.
Utilities and data centers also operate at entirely different speeds. Data centers often are backed by huge venture capitalists and investment funding to meet the ever-increasing traditional data center demand and the currently insatiable AI needs. Data center owners, developers and operators with available capital investment dollars make decisions very quickly and expect their partners to move swiftly on their projects. Conversely, utilities must seek the permission of their regulatory public utility commission (PUC) to build any new generation, forced to plan in terms of years and are continually scrutinized by the PUC, activists and other local influencers.
The bottom line: the two industries have fundamentally different DNAs and must find a consistent, meaningful way to communicate and coordinate, especially as the escalation in demand is in early stages. As AI and digital commerce continue to grow, the demand for co-location and data centers will only intensify. Some utilities have set up data center committees or have special team members to deal with the distinctive nature of data centers, but not enough utilities are set up with this purposeful structure.
