Solmar Insights
U.S. utilities are being forced to change how they plan for large industrial electricity customers as new AI data center campuses, some with proposed loads reaching 1,000 MW or more, outgrow the capabilities of traditional grid study methods. The rise of AI training facilities is exposing the limits of old modeling approaches and prompting a rapid shift to more sophisticated electromagnetic transient (EMT) studies to maintain grid reliability.
Key figures
AI campus proposed loads reaching 1,000 MW
Large demand swings occur within seconds
Traditional interconnection studies prove insufficient
AI campuses reshape load profiles
The traditional industrial load, such as those from refineries and manufacturing, has long been considered stable and predictable, enabling utilities to assess grid impact using conventional interconnection studies. These studies typically relied on static load representations and assumed steady power consumption. However, the emergence of AI campuses has disrupted this assumption, as their IT hardware can trigger large, unpredictable fluctuations in electricity demand.
Unlike older data centers, modern AI training campuses experience demand that can change by several hundred megawatts in mere seconds. These swings occur as compute-intensive tasks start, pause, and end, often with little forewarning. The unique operating characteristics of AI loads are causing utilities to reconsider both the way they assess these projects and the kinds of models they use.
The facilities also incorporate large amounts of power-electronic equipment, which reacts more quickly to grid disturbances than traditional industrial machines. Electronic interfaces, battery storage systems, on-site generation, and advanced plant controllers further increase the complexity of these load profiles, presenting new challenges for grid operators concerned with power reliability.
Limits of legacy grid studies
Utilities have traditionally relied on power-flow and positive-sequence dynamic studies to assess the impact of large loads. These methods focus on steady-state and average behavior, assuming that most changes in demand are gradual and well-contained within system tolerances. The erratic nature of AI campus electric loads, however, breaks those assumptions.
Fast changes in demand, triggered by AI workloads, can ripple across the grid and cause system stability challenges that simple models are not equipped to capture. At the same time, power-electronic interfaces in these facilities can react to voltage or frequency issues much more rapidly than conventional equipment, leading to large and sudden transitions to backup sources or alternate modes of operation.
Utilities are now finding that conventional interconnection studies are insufficient for identifying all the potential risks these mega-loads introduce. The increased complexity and the much faster timescales of electrical interactions open new avenues for instabilities that traditional models cannot fully predict.
Rise of electromagnetic transient analysis
Because of the new challenges introduced by AI facilities, utilities across North America are turning to electromagnetic transient (EMT) simulation and analysis. EMT studies offer a much finer timescale and resolution, allowing planners to simulate and observe how a facility’s operations will interact with grid dynamics during rapid transitions, system faults, or voltage disturbances.
Among the key questions being addressed with EMT approaches are a facility’s ride-through capability, whether it can maintain stable operation during grid disturbances without cascading problems across the network. Swings of hundreds of megawatts as AI jobs start and stop could, if uncoordinated, pose significant risks not just to individual sites but also to system-wide reliability.
Additionally, EMT analysis is necessary to study ramp-rate performance: can these facilities control how quickly their demand rises or falls? Rapid, uncontrolled swings could stress network stability, especially in areas where the grid is weak. Many utilities are now pushing for detailed documentation of plant-level controls and battery or on-site generation resources to prove that demand changes are managed within acceptable margins.
Power quality and stability risks
The instability risk is not limited to major fault events or single-point failures. Fast-changing compute loads can trigger oscillatory behavior across a network, especially on lower-strength lines, exciting low-frequency and subsynchronous oscillations that may not have previously posed problems on the grid. This raises the bar for demonstrating stable operation through high-detail simulation rather than assumptions based on past behavior.
Power quality is also under scrutiny. Repeated, rapid fluctuations in significant loads can generate flicker, harmonics, and other disturbances, potentially affecting neighboring facilities and sensitive grid equipment. The technical requirements and consequences of managing these disturbances are greater than those typically encountered with earlier classes of industrial customers.
These challenges emphasize that simply scaling up old models or relying on traditional study types is inadequate. Grid planners must adopt not just new tools but also a new mindset, focusing on time-domain simulations and plant-specific modeling to underpin any new large load interconnection.
Implications for interconnection and development
The rise of exceptionally large, fast-changing AI training campuses is redefining the interconnection process for both utilities and developers. Applications proposing multi-hundred-megawatt AI facilities drive deeper scrutiny, force longer study cycles, and raise the documentation burden on project sponsors.
For buyers of data center capacity, land, interconnection rights, or industrial power equipment, this means slower queue movement and evolving technical requirements. Modeling fidelity and plant-level controls become essential; projects without detailed analysis may be delayed or denied. Developers must factor in these additional timelines and requirements when structuring transactions or securing site control.
What this means for buyers
Data center power and interconnection capacity in the U.S. are impacted as AI campus load requests reach 1,000 MW. Applicants now face expanded study requirements, including electromagnetic transient (EMT) analysis, which raises both cost and lead time. Buyers seeking to secure large-scale data center infrastructure this quarter must prepare for longer interconnection processes and more detailed technical documentation than in past cycles.
Reporting via the original publisher
