Joules-to-GDP: How Much Economic Value Do AI Systems Produce per Unit of Energy?

AI systems are increasingly evaluated on their ability to complete real occupational tasks end to end and produce economic value. However, existing evaluations of economically valuable AI-produced work typically measure that value separately from the end-to-end resources used to produce it. We ask: How much economic value does an AI system create per unit of energy consumed? To answer this question, we define two metrics within our framework: Dollar Value per Joule (DPJ) and Time Saved per Joule (TPJ). DPJ divides the dollar value of completed work by the measured end-to-end energy consumed to produce that output, while TPJ divides the human labor time saved by the same energy. We apply the framework to 220 GDPval tasks spanning 44 occupations, five AI systems, and four accelerators. Our analysis reveals three main findings. First, economic return per joule varies substantially across types of work: Computer and Mathematical tasks achieve the highest DPJ at 139.1 × 10⁻⁶ USD/J, Engineering tasks achieve the highest TPJ at 2.22 × 10⁻⁶ hr/J, and occupation-category DPJ and TPJ vary by factors of 8.94 and 3.75. Second, model choice reveals a tradeoff between task value and return per joule. GLM 5.2 produces the highest value (106.72 USD per representative task), while Gemma 4 31B leads DPJ at 189.6 × 10⁻⁶ USD/J and TPJ at 4.19 × 10⁻⁶ hr/J. Third, accelerator choice further affects economic return per joule. Together, these findings show that the economic return on energy depends on both the work being performed and the AI system used to perform it. By connecting economically useful work to measured energy, the framework supports resource-aware choices across workloads, models, retry policies, and accelerators.