Is AI Bad for the Environment?

Artificial intelligence has become one of the most powerful technological trends of the 21st century, reshaping industries from healthcare to logistics. But as its use spreads, so does concern about its environmental footprint. The question isn’t simply “Is AI bad?” — it’s more nuanced: What environmental costs come with its growth? What benefits might it bring? And can these costs be managed or even turned into net gains?

AI and Energy Consumption

AI systems run on large server farms called data centers. These facilities require enormous amounts of electricity to power the processors that train and run models. A growing body of research shows that overall data centers already account for about 1–2 percent of global electricity demand, and this figure has been rising as AI workloads increase year over year. Global demand for data center electricity could more than double by 2030, with AI-specific processing driving a disproportionate share of that growth. Estimates suggest AI-related data center electricity use may climb from around tens of terawatt-hours today to hundreds by the end of the decade if current trends continue. These increases are tied not only to model training but also to serving millions of users worldwide.

Even though individual AI actions (like generating a prompt) consume a very small amount of energy — comparable to a lightbulb running for a few minutes — the scale matters. Millions of such interactions add up across users globally, and more advanced models require more powerful hardware. Plus, research shows that training large AI models, especially state-of-the-art ones, generates significant greenhouse gas emissions over their lifecycle.

Water Use and Local Resource Strain

Another part of the environmental impact — often less visible — is water consumption. Most traditional data centers use water in cooling systems to dissipate the heat generated by thousands of servers. In regions facing water scarcity, this can stress local freshwater supplies. Researchers project that global AI demands could result in billions of cubic meters of water withdrawals annually by the mid-to-late 2020s, potentially exceeding the total water use of medium-sized countries.

Experts also emphasize that water impacts go beyond just cooling: manufacturing AI hardware (chips, servers) uses water in production processes, and poorly treated wastewater from these operations can harm ecosystems.

‍

Hardware, Materials, and E-Waste

AI systems depend on specialized semiconductors like GPUs and TPUs, which require rare minerals such as cobalt, lithium, and rare earth elements. Mining and refining these materials often harm ecosystems, pollute water systems, and generate significant ecological damage. Frequent hardware upgrades and disposal of outdated equipment also contribute to growing volumes of electronic waste, which can release toxic substances if improperly managed.

Carbon Emissions and Broader Climate Effects

As more AI infrastructure is deployed, its associated carbon footprint becomes more important. Some projections estimate increases in carbon emissions from AI workloads over the next decade unless offset by cleaner energy sources. Data center emissions — including electricity generation and indirect emissions from cooling infrastructure — could continue to rise if global energy systems do not rapidly shift toward renewables.

Counterpoints: Efficiency Gains and Broader Benefits

However, the picture isn’t purely negative. AI can also help mitigate environmental impacts in other sectors. For example, AI is used to optimize energy grids, improve industrial efficiency, reduce agricultural water use, and model climate risks more accurately. At COP30 climate talks, both sides of the debate acknowledged AI’s potential to help climate efforts while also presenting its own footprint challenges.

Moreover, some efficiency studies show that AI hardware and software are becoming more power efficient, resulting in lower per-query energy consumption over time. Improved AI infrastructure can yield better environmental performance than older systems if deployed responsibly.

The Case of the AI Hub in Sines, Portugal: Good or Bad?

One of the largest recent AI infrastructure initiatives in Europe is unfolding in Sines, Portugal. Microsoft announced plans to invest $10 billion into developing AI computing infrastructure there in partnership with developers Start Campus, NVIDIA, and others. This project will bring tens of thousands of GPUs and aims to position Portugal as a key European hub for AI.

Critics might view any large data center expansion as adding to global energy and water demand. But the Sines hub has features that differentiate it from conventional facilities:

Renewable Power and Efficiency: The first phase of the Sines campus — a 26 MW facility known as SIN01 — is powered entirely by renewable energy and uses innovative cooling techniques, including seawater rather than freshwater sources. This reduces reliance on local potable water and lowers emissions.

Environmental Recognition: The project has been recognized with European industry awards for sustainable data center design, reflecting efforts to minimize its environmental footprint.

Strategic Infrastructure Advantages: Sines’ location on the Atlantic coast also makes it a desirable hub for subsea internet cables and green energy generation, potentially lowering energy costs and environmental impacts compared with inland regions that depend on fossil-based grids.

How to Think About the Trade-Offs

No major technology is without environmental cost, and AI is no exception. The key issues are:

Scale and Growth vs Efficiency:
AI’s total resource use grows as models become more powerful and widespread. However, hardware and algorithmic improvements are steadily improving per-use efficiency — meaning the impact per task can decline even as total impact rises.

Context Matters:
AI’s footprint is intertwined with broader electricity and water systems. Where data centers are plugged into renewable grids and cooled using non-freshwater sources (like seawater), the local impact is much lower than in water-stressed inland regions.

Policy and Transparency:
Experts argue that better reporting on energy and water use by tech firms — and stronger sustainability standards — could help align AI’s growth with climate goals. Without this transparency, it’s hard to manage its environmental costs effectively.

Summary

Short answer: AI has real environmental impacts — through energy consumption, water use, resource extraction, and emissions — that must be acknowledged and managed. At the same time, the technology also offers potential tools to help solve environmental problems.

Specific projects like Sines demonstrate that it’s possible to build large-scale AI infrastructure with sustainability in mind, but only if energy sources, cooling systems, and resource planning are consciously designed for minimal ecological disruption.

The future of AI and the environment depends less on whether AI exists and more on how societies choose to govern, measure, and innovate in building and powering these systems.

‍