New grid model shows 83.6% capacity boost needed for AI data centers and electrified manufacturing
Data centers and electrified oil refining could drive 22.2% of annual demand by 2033.
A team led by Ilias Mitrai at the University of Texas at Austin (affiliations inferred) has released a new preprint on arXiv that models the massive strain data centers and electrified manufacturing will place on power grids. The paper, posted on May 27, 2026 (arXiv:2605.29053), develops a multi-period capacity expansion model that optimizes investments in generation, storage, and transmission while satisfying hourly dispatch constraints. Using a synthetic grid resembling the ERCOT (Texas) system, the authors simulate a seven-year horizon where data centers account for 17.5% and electrified oil refining for 4.7% of total annual electricity demand by the end of the period.
The optimal investment plan calls for an 83.6% increase in generation capacity. The model favors solar and storage due to short construction times, backed by flexible thermal generators for reliability. Sensitivity analysis reveals that construction timelines for grid assets heavily influence the timing and mix of investments—a critical insight for utilities facing unprecedented load growth from AI compute clusters and industrial electrification. The framework is generalizable to other regions and demand scenarios, offering a tool for policymakers and grid operators to plan for the era of massive electricity demand.
- Data centers projected to consume 17.5% of annual electricity demand in the synthetic ERCOT grid model.
- Optimal investment strategy increases total generation capacity by 83.6% over seven years.
- Solar and storage dominate new capacity due to fast construction, but thermal flexibility remains essential.
Why It Matters
Grid planners now have a rigorous model to anticipate the explosive demand from AI data centers and electrified industry.