Artificial intelligence data centers are creating unprecedented demand spikes across North American power grids, forcing grid operators and regulators to rethink capacity planning and carbon policy simultaneously. A new academic study applies game theory to model this complex three-way interaction between regulators, ISO capacity markets, and technology investors.
The research constructs a Stackelberg-Bayesian framework in which a regulator acts as leader, setting carbon penalties and subsidies. The ISO capacity market then clears against traditional generation-expansion economics, while individual investors—facing incomplete information about competitors' costs and strategies—decide how much capacity to build and which technologies to deploy. This asymmetric setup yields a Bayesian Nash equilibrium that reveals the actual outcome when all players optimize under realistic uncertainty.
The study isolates AI's impact by modeling it as an additional load-growth multiplier applied to baseline expansion scenarios. This focused approach clarifies precisely how much extra capacity AI demand pulls into the market and which generation or storage technologies fill that gap. A key innovation is including second-life battery (SLB) storage—repurposed EV and stationary batteries—as a distinct competitor against conventional new-build batteries in the capacity market.
The researchers then simulate multiple policy scenarios: a carbon tax alone, a renewable energy subsidy, an SLB-specific subsidy, and various combinations. For each, they calculate the resulting equilibrium investment mix, total system carbon emissions, and investor profit margins.
Findings highlight trade-offs between cost-effectiveness and decarbonization. A carbon tax alone incentivizes renewable additions but may crowd out storage if prices are too low. An SLB subsidy accelerates circular-economy adoption of recycled batteries, reducing waste and manufacturing emissions, yet may lower capacity-market prices and deter new-build alternatives. The framework enables policymakers to stress-test capacity-market designs before rapid AI-driven growth locks in inefficient infrastructure.
