Green AI or Greenwashing?
ESG
Greenwashing happens when companies present themselves as sustainable without proving that their operations actually reduce environmental impact. In AI infrastructure, this risk is becoming acute.
The cloud is not weightless. Every AI token has a physical footprint: electricity, cooling, land, permits, local infrastructure and community trust. When a data center claims to be “green” because it buys renewable energy certificates or publishes climate commitments, but still relies on water-intensive cooling in water-stressed areas, that is not sustainability. It is resource extraction with better branding.
The issue is simple: AI servers generate enormous heat. To keep them operational, many conventional data centers use cooling systems that transfer heat away from chips and servers through chilled water loops, chillers and cooling towers. In evaporative cooling, water is not merely circulated. Part of it is evaporated into the air to remove heat. That means local water is consumed, often at precisely the wrong time: during hot periods, when both compute loads and community water stress can peak.
This creates a hidden trade-off. Operators may reduce electricity use by using water for cooling, but the cost is pushed onto the surrounding community. Local households, farmers, municipalities and ecosystems may be left competing with industrial-scale AI infrastructure for the same scarce resource. When the economic value of the data center is captured elsewhere, while the environmental burden remains local, the model becomes extractive.
Example of Greenwashing in Practice
Consider a hypothetical hyperscale data center built in a semi-arid region. The operator promotes the facility as “100% renewable” because it purchases renewable energy certificates from wind farms located hundreds of kilometers away. Its sustainability report highlights carbon neutrality and energy efficiency metrics.
However, the same facility relies on evaporative cooling towers that consume millions of liters of local freshwater each day during peak summer months. The region already faces seasonal droughts, and local authorities have imposed water restrictions on agriculture and households. Despite this, the data center continues to operate at full capacity, drawing from municipal water supplies.
Public communications emphasize clean energy usage, but omit or downplay water consumption. Community stakeholders are not fully informed about the long-term impact on local water availability. In this scenario, the data center appears sustainable on paper, yet shifts environmental pressure onto the surrounding community. This is a clear example of greenwashing: highlighting one positive metric while obscuring another critical impact.
Vitruvian rejects this model
Our position is that sustainable AI infrastructure must be engineered, not advertised. It must be measurable, verifiable and designed around the community from day one. In the Terakraft model, water is not consumed for data center cooling. Water is used as a clean source of hydropower, creating renewable electricity, which powers compute, which produces intelligence. The chain is transparent: water to power, power to compute, compute to intelligence.
This is why the relevant metric matters: WUE, Water Usage Effectiveness. Our design target is WUE 0 L/kWh IT. In practical terms, the data center does not rely on evaporating local water to cool servers. It does not solve the AI cooling problem by draining the community around it.
This is the difference between extractive AI and regenerative AI.
Extractive AI takes power, water and land, then exports the value.
Regenerative AI is built on clean energy, low-water design, local integration and community partnership.
Not all AI is equal. And not all tokens carry the same environmental cost.
BLOG & INSIGHTS
Exploring innovations, strategies, and the future.
Green AI or Greenwashing?
ESG
Greenwashing happens when companies present themselves as sustainable without proving that their operations actually reduce environmental impact. In AI infrastructure, this risk is becoming acute.
The cloud is not weightless. Every AI token has a physical footprint: electricity, cooling, land, permits, local infrastructure and community trust. When a data center claims to be “green” because it buys renewable energy certificates or publishes climate commitments, but still relies on water-intensive cooling in water-stressed areas, that is not sustainability. It is resource extraction with better branding.
The issue is simple: AI servers generate enormous heat. To keep them operational, many conventional data centers use cooling systems that transfer heat away from chips and servers through chilled water loops, chillers and cooling towers. In evaporative cooling, water is not merely circulated. Part of it is evaporated into the air to remove heat. That means local water is consumed, often at precisely the wrong time: during hot periods, when both compute loads and community water stress can peak.
This creates a hidden trade-off. Operators may reduce electricity use by using water for cooling, but the cost is pushed onto the surrounding community. Local households, farmers, municipalities and ecosystems may be left competing with industrial-scale AI infrastructure for the same scarce resource. When the economic value of the data center is captured elsewhere, while the environmental burden remains local, the model becomes extractive.
Example of Greenwashing in Practice
Consider a hypothetical hyperscale data center built in a semi-arid region. The operator promotes the facility as “100% renewable” because it purchases renewable energy certificates from wind farms located hundreds of kilometers away. Its sustainability report highlights carbon neutrality and energy efficiency metrics.
However, the same facility relies on evaporative cooling towers that consume millions of liters of local freshwater each day during peak summer months. The region already faces seasonal droughts, and local authorities have imposed water restrictions on agriculture and households. Despite this, the data center continues to operate at full capacity, drawing from municipal water supplies.
Public communications emphasize clean energy usage, but omit or downplay water consumption. Community stakeholders are not fully informed about the long-term impact on local water availability. In this scenario, the data center appears sustainable on paper, yet shifts environmental pressure onto the surrounding community. This is a clear example of greenwashing: highlighting one positive metric while obscuring another critical impact.
Vitruvian rejects this model
Our position is that sustainable AI infrastructure must be engineered, not advertised. It must be measurable, verifiable and designed around the community from day one. In the Terakraft model, water is not consumed for data center cooling. Water is used as a clean source of hydropower, creating renewable electricity, which powers compute, which produces intelligence. The chain is transparent: water to power, power to compute, compute to intelligence.
This is why the relevant metric matters: WUE, Water Usage Effectiveness. Our design target is WUE 0 L/kWh IT. In practical terms, the data center does not rely on evaporating local water to cool servers. It does not solve the AI cooling problem by draining the community around it.
This is the difference between extractive AI and regenerative AI.
Extractive AI takes power, water and land, then exports the value.
Regenerative AI is built on clean energy, low-water design, local integration and community partnership.
Not all AI is equal. And not all tokens carry the same environmental cost.
BLOG & INSIGHTS