Sign up today to get the best of our expert insight in your inbox.
What the data center boom means for grids
The artificial intelligence boom will generate huge demand for reliable, round-the-clock electricity at scale – and grids must adapt
5 minute read
The context
Electricity demand in developed countries has been largely stable for decades, but the energy transition will drive a massive increase in electricity demand to 2050. Having stayed relatively flat for 20 years, the Energy Information Administration projects power demand in the US will rise from a record 4,082 billion kWh in 2024 to 4,179 billion kilowatt hours (kWh) in 2025, and Wood Mackenzie expects it to continue to increase by 2-3% per year through 2030. By 2050, it is predicted to hit 5,178 billion kWh.
At the same time, as more distributed energy resources come online, grids must adapt to serve as multidirectional distribution hosts for customers who will increasingly supply electricity to the grid as well as taking power from it. As the leader in grid transformation, S&C Electric Company is well-placed to assess the implications for the power sector of this step change in the size and nature of resource required, and outline how the grid must adapt to address it.
A significant new source of demand
Cloud computing and industrial scale crypto mining have made data centers a notable new source of electricity demand, but with the rapid commercialization of generative artificial intelligence (AI), this demand will be turbocharged over the next decade. However, while big tech firms are used to a ‘move fast and break things’ ethos, the power sector works in timeframes of five-to-ten years, making access to electricity a major obstacle to growth. Data centers will also be competing with other new sources of demand for electricity, particularly transport, heating and cooling but also industrial processes and potentially green hydrogen.
The scale of the challenge
As a new and rapidly evolving commercial technology, estimates vary widely on the potential impact of AI in the US, although few doubt that it will be hugely significant. Having established the American AI Initiative via Executive Order in his first term, one of President Trump’s first actions on returning to power was an executive order aimed at removing barriers to US leadership in AI.
With most of the big global tech firms based in the United States, a significant percentage of the world’s data is stored in the country. Wood Mackenzie estimates that the US already hosts up to 40% of total global data center capacity, a market worth over US$68 billion in 2024. As AI adds significantly to demand, data center capacity is almost certain to escalate significantly. New data centers requiring almost 24GW were announced in the first half of 2024, more than triple the announcements for the same period in 2023, and overall 2025 investment is set to be double that seen in 2023.
The big four tech firms (Amazon, Meta, Google and Microsoft) forecast annual capex increases of 16% per year through 2025, while analysts’ forecasts for electricity demand growth from data centers range from 10-20% per year through 2030. At the top end of these demand growth estimates, 35 gigawatts (GW) of demand would be added. Overall, there seems little doubt that data centers will drive a large proportion of electricity demand growth through to the end of the decade.
The shape of the problem
Siting decisions for large data centers tend to focus on existing hubs within Independent System Operator (ISO) regions, further straining already overloaded local grids. Growing concentrations of large load within ISOs in locations such as Northern Virginia in PJM, Dallas-Fort Worth in ERCOT and the northern Midwest in SPP/MISO are likely to result in a highly asymmetrical power market. Using a proprietary Security Constrained Economic Dispatch (SCED) model to forecast real-time (RT) electricity prices, Wood Mackenzie analysts found that additional announced large loads could result in RT price increases of anywhere from US$1 in NYISO to almost US$70 in PJM. To complicate matters, data centers are also getting bigger, with hyperscale facilities (over one gigawatt (GW) or one million square feet in size) becoming increasingly common.
A further issue is that data centers need access to reliable, resilient, round-the-clock sources of electricity; creating potential issues if facilities rely on power from renewables – which many tech firms currently prefer from an ESG standpoint. As S&C points out, the challenges data centers create for the grid therefore relate not just to surging demand, but to the added uncertainty created by adding a concentration of large loads to a system, particularly in terms of forecasting peak and hourly loads. To operate efficiently, data centers require significant heating and cooling – American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) guidelines recommend that commercial scale facilities be kept within a temperature range of 18-27o C.
Most data centers operate around the clock. However, facilities dedicated to tasks such as crypto mining that are not time critical can scale their activity levels up and down based on intraday electricity prices, leading to significant variability in demand. Another emerging issue is that of large data centers automatically disconnecting during normal grid faults. This can lead to operating issues similar to those caused by the inherent intermittency of inverter-based resources (IBRs) such as wind and solar.
A growing issue
Power issues are consistently the most common cause of serious and severe data center outages – and they are likely to get worse. In its latest Long-Term Reliability Assessment, the North American Electric Reliability Corporation (NERC) found that lower than needed dispatchable capacity points to ‘critical reliability challenges’ facing the industry. It concluded that around half of the United States is at risk of shortfalls that could cause reduced power supply and outages in the next decade. According to the US Energy Information Administration, states with the biggest issues in terms of both frequency and duration include Maine, West Virginia and Vermont. Meanwhile, in a separate study looking specifically at data centers, Gartner predicted that 40% of existing facilities globally will be operationally constrained by access to sufficient power by 2027.
What happens if a data center goes down?
Outages are extremely costly for data centers. Research by the Uptime Institute found that more than half of significant incidents cost more than $100,000, and 16% costing more than $1 million. Generally, they lead to downtime, loss of revenue and additional costs, as well as significant stress and the need for additional resource to address the issue. As data centers are increasingly used for AI applications, the uncertainty and disruption caused could become even more serious due to the opaque nature of AI algorithms.
Read the second article in this series, in which we call on S&C’s expertise to look at how utilities and large-load customers can adapt their systems to cope as data centers place increasing demands on the grid.
In partnership with S&C Electric Company
S&C Electric Company provides a range of products and services to address electricity distribution issues and has been working with data center operators for many years.
Find out more about how their solutions can help.