2 Prediction within sample, measure by R was over 0.99 for all SPP zones except GRDAN1 (0.960), meaning the model fits the training data with extremely high accuracy. Testing on the 1/3 sample set that is held out of the training provides additional confidence 2 in the prediction additional years. This out-of-bag R is also very high and consistent, above 0.95 for all zones except GRDAN1. The GRDAN1 pricing zone is small relative to all other zones and had significant variation between years in the training data unrelated to weather, which decreased the performance of the regression. However, because of the size of the zone relative to the rest of SPP, this was not a major concern for the study. OPPORTUNISTIC LOADS Scenario electricity growth from EnergyPATHWAYS provides separate projections for data centers and hydrogen electrolysis. These loads are given special attention because of their potential magnitude and a step-change in expected growth in the past several years. Projections for both have significant long-term uncertainties, and the proportion of this growth that will occur within SPP is even more uncertain. Both data centers and electrolysis have a need for clean electricity due to corporate commitments and tax credit restrictions, respectively. We refer to these as opportunistic loads because they are economically opportunistic with regards to their location based on electricity prices and available supply. This makes SPP a relatively attractive place for these loads to relocate from a cost of energy perspective, since SPP has some of the best wind resources in the world. On the other hand, siting difÏculties, water availability, transmission constraints, and labor force availability may slow down the deployment of these loads. Data Centers Data centers consumed an estimated 250 TWh in 2024 across the U.S., and the median forecast is that this load will triple by 2030, based on the sources in Table 3. The recent boom in AI is the primary reason cited for the expected data center growth. Large Language Models (LLMs) take an extraordinary amount of computation, both in their training and in their use. Between these, EER estimates the ratio of electricity consumption between training and using LLM requests to be 5:1. This is significant because training LLMs have far lower latency requirements potentially allowing far more flexibility when it comes to siting these loads. Today, major data center hot spots exist in Virgina, Texas, and the Northwest. The development of data centers in these locations was historically driven by their proximity to major population centers, cheap real estate, available workforce, and, secondarily, low-cost electricity. However, if new potential data center loads including LLM training and cryptocurrency mining do not need to be located near population centers, it increases the possibility that SPP could see a disproportionate share of future data- center load growth. While we reflect a single estimate for use in this Scenario analysis, the long-term uncertainty on such a point estimate is large. SPP FUTURE LOAD SCENARIOS | EVOLVED ENERGY RESEARCH | 15