From Paradox to Progress A Net Positive AI Energy Framework 2025

What it is: defining net-positive AI energy “Net-positive AI energy” refers to a future in which the energy and resource savings enabled by AI exceed the energy and resources consumed throughout the AI system life cycle. The concept encompasses not only energy and resource savings but also broader system-level benefits, including improved energy security, enhanced grid reliability, optimized capacity and reduced operational costs,14 which are all essential to a resilient and sustainable energy future. Achieving this outcome requires recognizing the broader AI and energy nexus, the interconnection between AI’s demand for electricity, water, land and critical minerals, and the ecosystems that sustain them. While this paper acknowledges these intertwined resource impacts, its primary focus is on the energy dimension and how AI can be scaled in ways that strengthen energy security, competitiveness and sustainability. Expanding data centres increases energy consumption and emissions, and intensifies water demand for cooling; mineral extraction impacts land and biodiversity. Addressing these cascading pressures through holistic, resource-conscious design ensures that AI’s growth delivers net benefits for the economy, environment and society alike. Why it matters: the scale of the challenge To meet AI’s accelerating demand, recent estimates suggest more than $2 trillion in data centre projects are already planned or under construction globally over the next decade.15 The infrastructure required to power and connect these facilities is expanding at a similar pace. In the US alone, utilities are expected to invest $1.1 trillion over the next five years in new generation and grid capacity, primarily to serve data centres and growing AI workloads.16 This capital infusion will drive a historic infrastructure build-out, reshaping global power demand, grid planning and investment priorities. This growth could start to offset annual renewable energy gains,17 while disrupting national energy planning and straining grids. In many regions, building new renewable, nuclear and grid capacity quickly enough to meet AI-driven demand is not feasible, meaning part of this expansion may still rely on fossil fuels. Even if supply expands, infrastructure upgrades may lag, potentially constraining energy availability and hindering AI growth.18 A net-positive approach must therefore address not only emissions, but also how clean energy is allocated to ensure AI supports, rather than competes with, broader decarbonization efforts. Understanding the hidden drivers of AI’s energy footprint While several forces shape AI’s rising energy use, two main (but not exclusive) drivers are: 1 The Jevons paradox: As AI becomes more accessible and its use expands, a key challenge will be ensuring efficiency gains create real value rather than triggering even more AI use that cancels out the energy and resource savings.19 AI data centres’ energy and climate implications FIGURE 1 Use effect Use drives demand for new capacity Uses Website hosting, social networks, cloud, video platforms, traditional AI, generative AI, agentic AIData centres and digital infrastructure Wafers, AI accelerators, servers, colocation, hyperscalers Increase in investments for new data centres and capacityWidespread integration of generative AI into servicesIncrease in use per userExplosion in the number of users of AI servicesSupply effect New capacity enables new uses Source: The Shift Project. (2025). Intelligence artificielle, données, calculs: quelles infrastructures dans un monde décarboné? https://theshiftproject.org/app/uploads/2025/09/Synthese-RF-PIA-1.pdf. “Net-positive AI energy” refers to a future in which the energy and resource savings enabled by AI exceed the energy and resources consumed throughout the AI system life cycle. From Paradox to Progress: A Net-Positive AI Energy Framework 7