I have read a fair amount on the current state of AI, the future of AI, and the impact of AI on the modern workforce. And yet, for all that, I have not seen as much about the environmental impact of AI including AI data centers. Also rarely mentioned is the community impact/ benefit of a data center. One of the best pieces I read came from Parnassus Investments.
Parnassus makes a relatively simple but powerful argument: AI is not weightless software. It has a very physical footprint, and that footprint creates material investment risk. Investors who focus only on AI growth potential while ignoring environmental externalities are missing part of the equation.
Their concerns center around four major risks.
First, electricity demand. The paper notes that global data centers consumed approximately 415 terawatt-hours of electricity in 2024 — roughly 1.5% of global electricity use — and this demand could nearly double by 2030. In the U.S., data centers already account for over 4% of electricity demand, with a projected 133% increase by 2030. The concentration of these facilities in places like Texas, California, and Virginia creates intense local grid pressure.
Second, water consumption, which I think is the most underappreciated issue. AI data centers require enormous cooling capacity. Parnassus cites research suggesting a large data center may consume enough water daily to supply a town of 10,000–50,000 people. Hyperscale centers may consume between 60 and 124 billion liters annually. This becomes dangerous when facilities are built in water-stressed regions such as Phoenix. By 2050, roughly one in four existing data centers could face water scarcity constraints.
Third, community cost shifting. This is where I found the paper strongest. AI companies often externalize infrastructure costs. Utilities must upgrade transmission systems, but local households and small businesses frequently pay the bill. Parnassus cites the PJM Interconnection electricity market, where data center demand contributed to an estimated $9.3 billion increase in forward auction pricing, translating into higher residential electricity bills.
Fourth, upstream semiconductor manufacturing. Companies such as Taiwan Semiconductor Manufacturing Company require massive volumes of ultrapure water and hazardous chemicals. The paper notes that producing 1,000 gallons of ultrapure water may require 1,500 gallons of municipal water, creating hidden environmental strain before AI even reaches the data center stage.
What I appreciate about Parnassus is that they are not anti-AI. They focus on solutions.
They highlight several best practices:
• locating data centers near renewable energy sources (“bring your own power”)
• closed-loop water recycling systems
• liquid cooling and direct-to-chip cooling technology
• requiring developers to fund grid and water infrastructure upgrades
• community investment commitments, job training, and fair taxation
• proactive disclosure of environmental risk metrics
• science-based emissions targets from companies such as NVIDIA Corporation, Advanced Micro Devices, and Alphabet Inc..
Here is where I think the conversation is heading.
The future of AI data centers may become the next major environmental justice issue.
Historically, communities accepted factories, ports, railroads, and oil refineries because they created local economic value. But AI data centers are different.
A hyperscale AI facility may consume enormous electricity and water while employing relatively few permanent workers. Unlike a manufacturing plant employing thousands, an AI data center might create hundreds of construction jobs but only dozens of permanent operational jobs.
This creates a fundamental fairness question:
If local communities are absorbing the environmental burden, what exactly are they getting in return?
I think the next decade will force a new social contract around AI infrastructure.
1. Communities should receive infrastructure compensation
If a data center requires grid expansion or water system upgrades, the developer should fund it — not ratepayers. This requires political backbone which has been sadly lacking in many expansion projects (ie, developers building new homes in a city but not paying for infrastructure upgrades including adding freeway lanes to accommodate the increased traffic).
2. Communities should receive tax participation
Local governments should capture enough tax revenue to improve schools, roads, transit, and environmental resilience.
3. Communities should receive workforce development
AI companies should fund technical training pipelines through local community colleges and vocational schools.
4. Water use should be regulated regionally
Building hyperscale cooling systems in drought-prone areas like Arizona or Nevada may eventually require strict permitting frameworks.
5. Environmental disclosure should become mandatory
Investors deserve to know the full energy, water, and carbon cost of AI deployment.
6. AI infrastructure should resemble utility regulation
At some point, AI compute may become so systemically important that governments regulate major data center networks similarly to electric utilities.
The Parting Glass
We talk constantly about the brilliance of AI models, but very little about the coal plant, gas turbine, reservoir, or aquifer quietly supporting every prompt we type into systems like ChatGPT, Co-Pilot, or Gemini.
The digital world increasingly has a very physical cost.
Artificial intelligence may be virtual, but its costs are profoundly physical. If communities are expected to supply the water, electricity, and land that power the AI revolution, they deserve more than higher utility bills in return.
Risk does not disappear simply because disclosure requirements are weakened. AI infrastructure may become another example of that principle.