Given that electricity generation investments are expected to operate for 40 or more years, the decisions we make today can have long-term impacts on the electricity system and the ability and cost of meeting long-term environmental goals. This research investigates socially optimal near-term electricity investment decisions under uncertainty in future technology costs and policy by formulating a computable general equilibrium (CGE) model of the U.S. as a two-stage stochastic dynamic program. The unique feature of the study is a stochastic formulation of technological learning. Most studies that include technological learning utilize deterministic learning curves in which a given amount of investment, production or capacity leads to a given cost reduction. In a stochastic framework, investment in a technology in the current period depends on uncertain learning that will result and lower future costs of the technology. Results under stochastic technological learning suggest that additional near-term investment relative to what is optimal under no learning can be justified at technological learning rates as low as 10–15%, and at the 20–25% rates commonly found in literature for advanced non-carbon technologies, significant additional near-term investment can be justified. We also find it can be socially optimal to invest more in non-carbon technology when the rate of learning is uncertain compared to the case where the learning rate is certain. Increasing marginal costs produce an asymmetric loss function that under uncertainty leads to more near-term non-carbon investment in attempt to avoid the situation of high non-carbon costs and an external economic environment that creates high demand for non-carbon technology.