City, Study Warns
When the engines of innovation emit as much as whole metropolises, climate policy must account for unseen infrastructure as well as visible skylines.
Geneva, December 2025
A new study indicates that data centers running artificial intelligence workloads have generated a volume of carbon dioxide in 2025 comparable to the annual emissions of a megacity like New York, a revelation that has drawn urgent attention from environmental scientists, tech industry leaders and policy makers. The findings underscore the tension between rapid technological growth and the imperative of climate stewardship, suggesting that the environmental footprint of computational infrastructure now rivals that of some of the most carbon-intensive urban populations in the world. While the exact comparison naturally involves estimates and modelling, the central message is clear: the emissions linked to AI-driven processing are no longer negligible and require systematic response.
The study analysed energy consumption patterns, cooling requirements and server utilisation rates across hundreds of major AI-centric data centers, covering facilities used by cloud providers, research institutions and commercial technology firms. It found that the cumulative emissions associated with powering, cooling and maintaining these centers in 2025 approximate the magnitude of carbon emissions produced annually by the combined activities of households, transportation and industry within New York City. The comparison serves as a stark illustration of how modern information infrastructure — which underpins everything from digital assistants to large language models and autonomous systems — has environmental consequences that extend far beyond the visible footprint of office towers or city transport networks.
Experts point out that the exponential increase in demand for AI services — from personalised consumer applications to large-scale scientific computing — has driven data centers to operate at higher intensities and for longer cycles. Unlike conventional computing tasks, AI workloads are often characterised by sustained, high-power requirements, placing heavier burdens on electrical grids and cooling systems. This dynamic amplifies the environmental impact of data centers relative to more sporadic computing operations of the past, embedding a new form of energy consumption deeply within the ecosystem of modern technology.
The implications of these findings traverse sectors. On the one hand, they provoke reflection within the technology industry about the sustainability of current development trajectories. Companies investing in AI are increasingly pressed to reconcile innovation with environmental responsibility, exploring strategies such as greater reliance on renewable energy sources, optimisation of algorithms for energy efficiency and deployment of advanced cooling technologies that reduce power draw. On the other hand, regulators and environmental advocates view the data as a call to integrate digital infrastructure into broader climate policy frameworks, rather than treating the tech sector’s emissions as peripheral to core industrial concerns.
The study also highlights geographic and economic dimensions of the issue. Data centers are often sited in regions where energy costs and infrastructure capacity make expansion feasible, which can create uneven patterns of emissions distribution. Some countries and subnational regions have begun to implement carbon pricing schemes or energy-intensity standards that target digital infrastructure, but there is no globally harmonised approach yet. The magnitude of AI-linked emissions — on par with a major city — adds weight to arguments for international cooperation on energy standards, transparency in reporting and incentives for cleaner operations.
Critics caution, however, that focusing solely on emissions comparisons risks obscuring broader systemic questions. They argue that the environmental impact of AI must be considered alongside its potential contributions to climate mitigation, such as optimising energy grids, improving materials science or enhancing environmental modelling. In their view, the challenge lies not only in reducing the carbon footprint of AI itself but in aligning its applications with solutions that address climate change at scale.
Public awareness of the issue remains uneven. While energy consumption by digital services is a topic of occasional debate, few users of AI or cloud-based tools consider the environmental cost embedded in each query, recommendation or automated process. As the study makes clear, the environmental footprint of technology is not abstract; it manifests in measurable emissions with real climate consequences. Translating those implications into public policy and corporate accountability remains a task for governments, industry consortia and civil society alike.
In confronting the intersection of technological progress and environmental impact, the conversation must expand beyond efficiency increments and into systemic climate strategies. The comparison with a city like New York — itself a complex urban system with intertwined emissions sources — underscores that digital infrastructures are now agents in the global climate equation. Addressing their emissions will require both technological innovation and collective will.
Phoenix24. The visible and the hidden, in context. / Phoenix24. Lo visible y lo oculto, en contexto.