Fresh research from VU Amsterdam estimates the environmental footprint of AI infrastructure in 2025 and highlights the need for greater transparency from companies regarding energy use, emissions, and water consumption.
What the Latest Research Shows
A groundbreaking study published in the journal Patterns by Alex de Vries-Gao from VU Amsterdam Institute for Environmental Studies has quantified AI's environmental impact for 2025. The numbers are significant: AI systems now consume roughly the same amount of electricity as entire cities and use water volumes comparable to global bottled water consumption.
The research is particularly valuable because it addresses a major data gap. While tech companies report overall sustainability metrics, they rarely break down the specific environmental costs of AI services. De Vries-Gao worked around this by analyzing analyst estimates, earnings calls, and hardware production data to create the most comprehensive picture we have of AI's resource consumption.
Breaking Down the Numbers
Energy consumption: AI systems globally consumed an estimated 23GW of power in 2025–surpassing Bitcoin mining's energy use from 2024. This translates to carbon emissions between 32.6 and 79.7 million tons annually, roughly equivalent to New York City's entire carbon footprint.
Water usage: The water consumption figures are even more striking. AI operations used between 312.5 and 764.6 billion liters of water this year. This includes both direct cooling needs for data centers and indirect consumption from power generation.
Why the wide ranges? The variation reflects the current lack of transparency in the industry. Companies don't typically disclose location-specific data or break down resource use by service type, making precise calculations challenging.
How AI Uses Water and Energy
Understanding the mechanics helps explain the scale. Data centers housing AI systems require constant cooling to prevent server overheating. This cooling process consumes significant amounts of water directly.
But there's an indirect component that's often larger: the power plants supplying electricity to these data centers also need water for cooling and steam generation. When you factor in both direct and indirect water use, the total footprint becomes substantial.
The energy intensity varies significantly based on:
- Data center location and local grid composition
- Cooling system efficiency
- Hardware generation and optimization
- Workload types and processing requirements
What This Means for Different Industries
Technology sector: Companies using AI extensively now have better data to inform their sustainability reporting and vendor selection processes.
Financial services: Organizations implementing AI for fraud detection, algorithmic trading, or customer service can better estimate their Scope 3 emissions.
Healthcare: AI applications in diagnostics and research can now be evaluated not just for effectiveness but also environmental impact.
Manufacturing: Predictive maintenance and quality control AI systems can be assessed within broader sustainability frameworks.
The Transparency Challenge
The research highlights a critical issue: most AI providers don't offer granular environmental impact data. This makes it difficult for businesses to:
- Accurately report their carbon footprint
- Make informed vendor comparisons
- Set realistic sustainability targets
- Comply with emerging regulations
Some companies are beginning to address this. Google recently published data suggesting a single AI text prompt uses about 5 drops of water, though experts debate whether this captures the full picture.
Emerging Solutions and Innovations
The industry isn't standing still. Several developments are worth watching:
Efficient hardware: New chip designs specifically for AI workloads can reduce energy consumption per operation.
Renewable energy: Major cloud providers are increasingly powering data centers with renewable sources.
Cooling innovations: Advanced cooling systems, including liquid cooling and free air cooling in suitable climates, can reduce water consumption.
Edge computing: Processing data closer to where it's generated can reduce the need for large centralized data centers.
How to Use This Information
- For procurement decisions: Ask AI vendors about their energy sources, data center locations, and resource consumption metrics.
- For sustainability reporting: Include AI-related emissions in your Scope 3 calculations, even if you need to use estimates.
- For strategic planning: Factor environmental impact into AI adoption decisions alongside cost and performance metrics.
- For stakeholder communication: Use this research to demonstrate informed decision-making about technology choices.
Looking Ahead
This research represents a baseline rather than a ceiling. As AI adoption accelerates, resource consumption will likely grow unless efficiency improvements keep pace. The study's conservative methodology suggests actual impacts could be higher when full lifecycle effects are considered.
Key trends to monitor:
- Regulatory requirements for environmental disclosure
- Industry standards for measuring AI's carbon footprint
- Technological advances in energy efficiency
- Geographic shifts in data center locations
Use this research to start conversations with your AI vendors about environmental impact. Even if they can't provide exact figures yet, asking these questions signals market demand for transparency–which is exactly what the industry needs to drive better practices.
The businesses that understand and plan for AI's environmental impact today will be better positioned as regulations tighten and stakeholder expectations evolve. This isn't about avoiding beneficial technology–it's about using it thoughtfully and sustainably.
