What this threat is
Training large AI models and running them at inference scale requires substantial compute, and compute requires electricity. Data centers that house AI training and serving infrastructure have become major electricity consumers in a short period, and the trajectory is upward. The hardware supply chain adds another layer: producing the chips that power AI training involves energy-intensive manufacturing, rare materials, and significant water use for cooling at the facility level. These direct energy demands are the most visible part of AI's environmental footprint, and they're concentrated enough that communities near large data center buildouts are already grappling with their effects on local electricity grids and water supplies.
The "rebound effect" is a less visible but potentially larger problem. Efficiency improvements from AI don't necessarily reduce total consumption; they often enable more consumption. If AI makes manufacturing processes more efficient, those efficiency gains tend to get reinvested in higher output rather than lower energy use. If AI reduces the cost of logistics, it can enable more logistics rather than fewer. This dynamic plays out across sectors: AI-optimized agriculture might enable higher yields but also enable expansion into land that wasn't previously farmed, AI-optimized travel planning might make trips more efficient but also make more trips appealing, AI-powered recommendation systems might increase consumption in aggregate even if each individual decision it facilitates is more resource-efficient than what it replaced. The efficiency gains are real, but they don't automatically translate into lower overall environmental impact.
The indirect effect of AI optimizing for economic output is the most speculative but potentially most significant dimension. AI systems deployed to maximize production, revenue, extraction efficiency, or logistics throughput will, by default, optimize for those objectives and not for environmental externalities that aren't explicitly included in the objective. An AI that makes an oil extraction operation more efficient is making that operation produce more oil. An AI that optimizes a shipping network is making that network move more goods. These applications of AI aren't inherently malicious, but they represent the application of powerful optimization capability to activities that are already major drivers of climate change, without any built-in constraint on the environmental consequences of the optimization. Whether this constitutes "AI accelerating climate change" depends on how you attribute causation, but the practical effect of removing friction from carbon-intensive activities is to enable more of them.
The picture isn't one-sided. AI is also being applied to climate solutions in ways that could be genuinely important. Accelerating materials science research for better batteries, optimizing renewable energy grid management, improving the accuracy of climate models, detecting methane emissions from satellites, identifying optimal locations for renewable energy infrastructure, and reducing waste in supply chains are all areas where AI applications could have positive climate consequences. The question for this threat isn't whether AI has any positive climate applications, it clearly does, but whether the aggregate effect of AI deployment at scale is net positive or net negative, and whether the governance frameworks being put in place are adequate to ensure the former.
Why it matters
Climate change is already a civilizational threat by itself, with well-documented consequences for food security, water availability, human health, displacement, conflict, and economic stability. If AI significantly worsens climate outcomes, those consequences become more severe, and they interact with the social and political instability that climate disruption produces. A world dealing with more frequent and severe climate impacts is also a world with less institutional capacity to govern AI, less economic surplus to invest in safety research, and more political instability of the kind that makes coordinated international action harder. The two crises aren't independent of each other; they compete for the same institutional bandwidth and compound each other's effects.
The justice dimension deserves explicit attention. The populations that bear the largest costs from climate change tend to be in lower-income countries and lower-income communities within wealthier countries. The populations that capture most of the economic benefit from AI tend to be in wealthier countries and within those countries, in higher-income groups. If AI accelerates climate change, the distribution of costs and benefits becomes even more skewed: the people who benefit least from AI development bear the largest share of its climate consequences. This isn't an abstract fairness concern; it has direct implications for the political sustainability of the international cooperation that both climate action and AI governance require.
The resource constraint problem affects the long-term trajectory of AI development itself. Current AI scaling relies on continued increases in compute, which requires continued expansion of data center infrastructure, chip manufacturing, and energy supply. If AI's energy demands come to compete with climate goals in ways that are politically or practically unsustainable, that creates pressure to curtail AI development, or alternatively, to curtail climate action. Neither outcome is good, and the conflict between them is easier to avoid now, by building sustainability constraints into AI development from the start, than later, when the infrastructure is already built and the political dynamics are harder to shift.
Where things stand today
The evidence on AI's energy use points consistently upward. Major technology companies have acknowledged that their energy consumption is growing significantly as a result of AI, and several have reported that this growth is making it harder to meet the climate commitments they made in earlier years. The direction of the evidence is clear even without reliable aggregate figures, which are difficult to verify because data center energy use isn't systematically disclosed. Water consumption for cooling is similarly trending upward, a concern that's particularly acute in water-stressed regions where some large AI infrastructure is being built.
Corporate sustainability commitments in the AI sector are real but incomplete. Major AI companies have made commitments to use renewable energy for their data centers, and some have achieved meaningful progress on that goal. But the renewable energy they're procuring is coming from a grid that isn't yet fully decarbonized, and their demand growth is putting upward pressure on overall electricity consumption that can be hard to decarbonize fast enough. Research on whether AI's climate applications genuinely offset its direct energy footprint shows promising results in specific domains (materials discovery, weather modeling, grid optimization) but isn't yet sufficient to claim a confident net-positive balance across all AI applications at the current scale of deployment.
The EU AI Act includes provisions that touch on AI's environmental impact. High-risk AI systems are required to document their resource use, including energy consumption, which creates at least a baseline of transparency that was previously absent. The Act also instructs the European AI Office to develop codes of practice for GPAI models that will include sustainability dimensions. These are early-stage governance interventions; they don't yet constitute a comprehensive framework for ensuring that AI development happens within planetary boundaries, but they establish a precedent for treating energy and environmental impact as legitimate regulatory concerns rather than purely private business decisions.
How Better Societies helps
Compliance: EU AI Act energy transparency requirements and the sustainability provisions being developed for high-risk systems create real documentation and disclosure obligations. Our compliance programs help organizations understand what's required, implement the resource-use tracking and documentation systems the regulation demands, and position their AI deployments to meet evolving environmental standards before they become enforcement priorities.
Summit: The intersection of AI and climate requires coordinated thinking across AI developers, energy policymakers, climate scientists, and environmental advocates who rarely sit in the same room. The Better Societies Summit creates that space, with specific programming on the governance frameworks, disclosure standards, and investment incentives needed to ensure AI development and climate goals can coexist rather than conflict.
Accelerator: Founders building AI applications with genuine environmental benefit, whether in renewable energy optimization, climate modeling, sustainable materials, or supply chain decarbonization, and founders building tools that make AI infrastructure more energy-efficient or transparent about its resource use, are doing work that matters. The Better Societies Accelerator supports founders who are working to make the AI-climate relationship net positive rather than net negative.