Optimal energy portfolio investment strategies for data centers under deep market uncertainty

Artificial intelligence is driving soaring electricity demand, with data centers projected to account for 44% of U.S. load growth through 2028. Energy planning for data centers requires robust investment strategies amid volatile electricity prices and grid interconnection delays. This study introduces a regret minimization framework, inspired by decision theory and game theory, to optimize energy portfolios for data centers, with potential relevance to other energy-intensive sectors. Using hindsight benchmarks and historical back-testing, the framework employs deterministic optimization and full cross-validation to evaluate portfolio robustness through regret-based metrics. In a hyperscale data center case study, diversified portfolios combining solar photovoltaic (PV), wind energy, and gas turbines demonstrate lower regret than grid-only strategies and provide resilience to evolving regulatory requirements. Reducing grid reliance from 100% to 20% mitigates cost volatility and interconnection risks. Low-carbon technologies, like nuclear small modular reactors (SMR), become viable with reduced capital costs and limited land availability. This work offers a robust planning framework for optimal strategic investments not only for data centers but also for energy-intensive industries navigating the modern, volatile energy landscape. • Proposing a regret-minimization framework for robust data center energy investment. • Reducing grid dependency and enforcing emission targets minimizes financial regret. • Diversified behind-the-meter portfolios are a powerful hedge against price volatility. • Identifying viability pathways for nuclear SMRs under emission and land constraints. • Grid-only energy strategies lead to high financial risk and poor portfolio robustness.