An MIT study published in iScience finds that data centers could lower U.S. energy costs by 2–7% if they shift more than 20% of their consumption to non-peak hours. Using the Gen X power grid model across Texas, the Mid-Atlantic, and western U.S. regions — which together host ~82% of U.S. data centers — researchers found flexible scheduling spreads fixed grid costs over higher average usage. However, environmental outcomes vary by region: Texas, with 54% wind power, could see 40% fewer CO2 emissions under flexible data center use, while the Mid-Atlantic could see a 3% increase as flexible scheduling inadvertently supports coal plants. The study also distinguishes between AI training workloads (more flexible) and inference workloads (demand-driven). Researchers suggest a 'connect and manage' policy — offering faster grid hookups in exchange for time-of-use flexibility — as a practical regulatory lever to incentivize adoption.
