为可持续的人工智能提供动力

Balancing growth with environmental responsibility Authors ShalabhKumar Singh SenthilRamani MatthewRobinson BhaskarGhosh SanjayPodder StephanieJamison AdamBurden Global Lead–TechnologySustainabilityInnovation Global Lead–Data & AI Chief Strategy &Innovation Officer Global InnovationLead Global SustainabilityServices Lead and GlobalResources IndustryPractice Chair Principal Director–Accenture Research Managing Director–Accenture Research AI is scaling fast, but at what cost? Our new analysisintroduces the Sustainable AI Quotient (SAIQ), ametric that tracks how efficiently AI transforms money,energy and emissions into measurable performance,using tokens as a standardized unit of performance. Itenables businesses to balance financial viability, energyresilience, and environmental impact based on theirunique organizational priorities. Drawing on proprietarymodeling and expert insights, we identify fourimperatives to help leaders scale AI sustainably: smartersilicon, cleaner data centers, more strategic AI use andgovernance-as-code. Preface We believe the companiesthat ask “What arewe getting from theresources we’re puttinginto AI?”—and act on theanswer—will be the onesthat define the future. This insight prompted us to act. We reached out to 17external experts including academics, practitionersand thought leaders from computer science, AIand the energy industry. We also brought togetherexperts from across Accenture’s Sustainability,Technology Innovation, Strategy and Data & AIpractices. Our aim was to understand what leadersshould do to ensure long-term performanceand resilience. Artificial intelligence is no longer an emergingtrend—it’s reshaping everything from healthcarediagnostics to retail supply chains. As AI adoption grows, the energy, emissions andwater costs to support large-scale AI are risingjust as fast. If left unchecked, the environmentalfootprint of AI could threaten corporatesustainability goals and overstep our planet’s limits.But this isn’t just a sustainability risk—it’s a designand innovation opportunity. Solving for it could be apathway to resilience and long-term value creation. A key outcome is a new metric: Sustainable AIQuotient (SAIQ). It measures how effectivelycompanies are converting investment—dollars,energy, water—into valuable AI outcomes. This metricis more than a benchmark. It is a strategic lens fordecision-making. We wanted to ground the conversation in facts.Rather than relying on scattered estimates of AI’senergy and resource demands, we produced ourown global estimates of AI’s energy consumption,carbon emissions and water use. The goal: toidentify what leaders need to know now based onreal infrastructure data, modeled scenarios andsustainability impact metrics. To help companies maximize their return on AIinvestments—not just in financial terms but alsoenergy, water and environmental impact—wedeveloped a practical framework around fourimperatives: deploy smarter silicon, decarbonize datacenters, use AI thoughtfully and embed sustainabilityinto AI governance. Together, these actions allowbusinesses to scale AI sustainably in alignment withthe new metric, balancing growth with environmentalstewardship and regulatory readiness. We began with the question: Are we scaling AI ina way that’s sustainable? Based on current trends,the answer is no. The infrastructure fueling the AIrevolution is consuming unprecedented amounts ofelectricity and water at rates that rival entire nations. AI’s efficiencyparadox Artificial intelligence is no longer just a frontierof innovation—it’s fast becoming the backbone ofglobal business. According to Accenture’s Front-Runners Guide to Scaling AI, organizations that arekeen to reinvent themselves using gen AI expecta 13% increase in productivity, a 12% increase inrevenue growth, an 11% improvement in customerexperience and an 11% decrease in costs within18 months of deploying and scaling gen AI acrosstheir enterprise.1But its ascent comes with acost we can no longer afford to ignore. Theinfrastructure powering today’s AI revolution isconsuming vast amounts of electricity and waterwhile emitting significant carbon, creating anenvironmental footprint that risks undermining thevery progress AI is meant to deliver. Carbon emissions for AI may account for 3.4%of total global emissions—an 11-fold increasein one decade. Left unchecked, this trajectorywon’t just damage the planet. It will inflate costs,stress supply chains and expose companies toregulatory risk, carbon taxes and stakeholderbacklash. Power use in AI datacenters is expectedto grow over 10x bythe end of the decade,reaching 612 terawatt-hours (TWh), equivalentto the power demand ofCanada in 2022. It’s the irony of the era. The same technologythat represents a generational leap forward forbusinesses in terms of efficiency, productivity andcapability could undermine business goals. According to Accenture’s Destination NetZero report, only 16% of the world’s 2,000 la