AI data centers are driving fast growth in electricity demand. This paper builds a game-like model to see how a carbon tax and different subsidies would change which new power plants and batteries get built. The authors pay special attention to second‑life batteries — packs retired from electric vehicles — and whether these cheaper but shorter‑lived units can win revenue in a capacity market and help meet AI‑driven demand.

The researchers create a three‑level “Stackelberg–Bayesian” game. At the top, a regulator (the leader) picks a policy package of a carbon tax and subsidies. In the middle, a single system operator clears a capacity market against an energy balance using a standard generation‑expansion model. At the bottom, competing investors for renewables, conventional generators, and storage (including first‑life and second‑life batteries) choose how much capacity to build and how to operate it. Investors act with incomplete information, so their equilibrium is a Bayesian Nash equilibrium. The model isolates AI impact by adding a simple load‑growth multiplier; the paper notes U.S. data‑center load rose from about 76 TWh in 2018 to 176 TWh in 2023 and could double or triple by 2028.

Second‑life batteries (SLBs) are modeled with physical detail. Packs retired from electric vehicles typically keep 70–80% of their original energy and degrade with cumulative energy throughput. They are cheaper and faster to deploy than new batteries and have lower embodied carbon, but they wear out sooner. The authors represent SLB life as 5–8 years versus 10–15 years for new battery energy storage systems (BESS), track capacity by build vintage, and include repurposing costs and possible replacement if a short life causes early retirement. The core market question the model studies is whether capacity and energy revenues — together with a possible SLB subsidy — make SLB investment privately profitable.