Researchers from MIT and the MIT-IBM Watson AI Lab published EnergAIzer, a framework that estimates how much power an AI workload will consume on a given GPU or accelerator in seconds with ~8% error, versus hours or days for traditional simulation-based methods. The tool identifies repeatable patterns in AI computation kernels and applies correction terms for memory bandwidth inefficiencies, and generalizes to novel hardware without re-running full simulations. The work is motivated by projections that data centers will consume up to 12% of U.S. electricity by 2028; EnergAIzer targets both data center operators scheduling workloads and model developers who want to estimate energy cost before deployment. The paper was presented at the IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) in April 2026.