Does AI Really Use Fresh Water? What Filipinos Need to Know Before Rejecting AI

Artificial intelligence can write, design, analyze, translate, organize information, and help turn ideas into working products. But every AI response depends on physical infrastructure.

There are servers processing the request, electricity powering those servers, and systems preventing the equipment from overheating. Some data centers use freshwater directly for cooling. Water may also be used indirectly in producing the electricity that keeps those facilities running.

So yes, AI has an environmental cost.

That truth should not be hidden by people who support AI. It should also not be stretched into the claim that every prompt, model, facility, and country has exactly the same environmental impact.

As the founder of FMB&CO., a company that uses AI extensively, I believe our responsibility is not to become defensive. It is to help people understand the issue, acknowledge what we still cannot measure, and improve how we use the technology.

Why does AI use water?

AI itself does not physically drink water. The demand comes from the facilities and energy systems that make computation possible.

Servers release heat while processing data. Cooling towers may remove that heat through evaporation, which consumes water. Other facilities use closed-loop systems, air cooling, dry cooling, or hybrid designs that can reduce direct freshwater demand, sometimes with trade-offs in electricity use or cost.

The electricity behind an AI request can carry another water footprint. Some power plants withdraw or consume water while generating electricity. This means a complete assessment may include both water used at the data center and water associated with its energy supply.

Water withdrawal and water consumption are not the same

Public discussions often mix two different measurements. The United States Geological Survey defines a water withdrawal as water removed from a source for use. Consumptive use refers to the portion that is evaporated, incorporated into products or crops, consumed by people or livestock, or otherwise not immediately returned to the water environment.

This distinction matters. A large withdrawal is not automatically the same as permanently consuming the entire amount, although both can affect local water systems depending on timing, quality, temperature, and scarcity.

Why one viral number cannot describe all AI

Researchers have shown that the water efficiency of AI can vary according to where and when computing takes place. A model running in a cool region with a lower-water energy mix may have a different footprint from the same workload processed during extreme heat in a water-stressed location.

Google reported that the median text prompt in its Gemini Apps production environment used approximately 0.26 milliliters of water and 0.24 watt-hours of energy under its measurement method. That number is useful because it is based on a specific production system. It is not a universal measurement for every company, model, image, video, or prompt.

A small amount multiplied across millions or billions of requests can still become significant. The point is not that the concern disappears. The point is that responsible discussion requires scale, location, technology, and methodology, not only a dramatic number detached from its original assumptions.

We should not defend innovation by denying its consequences. We should not reject an entire technology because its infrastructure still needs improvement.

The larger issue is infrastructure at scale

The International Energy Agency projects that global electricity consumption from data centers could reach around 945 terawatt-hours by 2030. AI is not the only source of data-center demand, but it is a major force behind current expansion.

This is why communities have every right to ask where the power and water will come from, what happens during drought, whether households and farms will be protected, and whether environmental information will be made public.

Those are not anti-technology questions. They are questions of governance, planning, and public accountability.

FMB&CO. uses AI extensively

FMB&CO. uses artificial intelligence in creative development, strategy, research support, digital products, operational systems, and the exploration of new ideas. We will not hide that use or reduce it to a fashionable disclaimer.

AI has helped us organize complex work, test ideas, develop concepts, and build with resources that would previously have required much larger teams. For small firms, independent creators, and community-led projects, that access can be transformative.

Because we benefit from AI, we must also acknowledge its environmental and social costs.

FMB&CO. does not operate a hyperscale data center or train a large foundation model from the ground up. Most of our use depends on services provided by other technology companies. But outsourcing infrastructure does not erase our indirect contribution to demand.

Did we always use AI efficiently?

No.

When I began using AI, I did not fully understand the differences between models, prompting methods, processing requirements, image generation, and environmental impact. I experimented. I generated multiple versions. At times, I may have used a more powerful system than the task required.

I believe many users began in the same way. The platforms gave us powerful tools before giving ordinary people a clear environmental guide.

I will not pretend that I knew everything from the beginning. But accountability should not become a performance of guilt. The meaningful question is what we choose after we learn more.

What an individual user can realistically do

• Use AI when it creates real value, rather than adding it to every small interaction.
• Write clearer instructions so fewer unnecessary retries are required.
• Reuse approved outputs instead of regenerating the same material without reason.
• Choose smaller or simpler models when the platform offers a suitable option.
• Reserve heavier image, audio, and video generation for defined creative needs.
• Fact-check important outputs so computation does not become misinformation at scale.

These choices help, but individual users should not carry the entire burden. Most people cannot see which facility handled a request, which cooling method was used, or how much electricity and water were consumed.

The companies building and operating the infrastructure hold the greatest capacity to reduce impact.

What technology companies should disclose

Providers should publish meaningful water and energy information, explain the boundaries of their measurements, and identify operations in water-stressed regions. They should distinguish direct cooling demand from indirect electricity-related water use and distinguish withdrawals from consumption.

They should also invest in smaller and more efficient models, reclaimed water, closed-loop systems, improved hardware, lower-impact energy, and infrastructure that protects communities during scarcity.

Microsoft has announced newer data-center designs intended to use closed-loop chip-level cooling without evaporating water for cooling during operation. That does not make an entire facility impact-free, but it demonstrates that water use is an engineering choice that can be improved rather than an unchangeable feature of computing.

What responsible AI means to FMB&CO.

Responsible AI begins by asking whether AI is necessary for the task. A structured database, saved reply, traditional software function, or human decision may sometimes be faster, more reliable, and less resource-intensive.

When AI is appropriate, we should match the model to the work, reduce unnecessary generation, preserve useful outputs, and maintain human review for factual, cultural, ethical, and public-facing decisions.

We will not claim that our work is carbon-neutral, water-positive, or environmentally harmless without independent evidence. We will not use a social purpose as permission to ignore environmental cost.

Our aim is to create more meaningful value from every use of technology and less disposable computation.

Francine’s perspective · Opinion

Supporting AI does not require silence.

I remain a firm believer in artificial intelligence. I believe it can democratize creativity, expand access to knowledge, support people with limited resources, preserve culture and language, and help small organizations build what once belonged only to institutions with greater capital.

I also believe that no community should be expected to sacrifice its water, electricity, or quality of life without transparency and protection.

Those beliefs are not contradictory.

Being pro-AI does not mean being anti-environment. Criticizing irresponsible infrastructure does not make someone anti-progress. We can support innovation while demanding better engineering, stronger regulation, clearer reporting, and human accountability.

When I started using AI, I did not know everything. I may not always have selected the most efficient model or written the best prompt. Learning that does not require me to abandon the technology. It requires me to use it with greater intention.

The future should not be built through denial, and it should not be stopped by fear. It should be built through education, transparency, and the willingness to improve.

With love,
Francine Marie Bautista
Founder and CEO, FMB&CO.

Frequently asked questions

Sources and further reading

This article separates provider-specific measurements from universal claims and distinguishes direct data-center water use from indirect water associated with electricity generation.

International Energy Agency: Energy demand from AIGoogle Cloud: Measuring the environmental impact of AI inferenceMaking AI Less Thirsty: water footprint and geographic variationUnited States Geological Survey: withdrawal and consumptive use of waterUnited States Department of Energy: water efficiency in data centersMicrosoft: next-generation data-center cooling design