The Cost of Bloated Code: How Inefficient AI Architecture Accentuates the Global Water Crisis

We often speak about artificial intelligence as if it exists in a whole different space: inside the cloud, behind a screen. We ask it to summarize, translate, generate, calculate, recommend, and create. Then, magically an answer appears.

But how that answer is actually generated?

Well, there are processors, cooling systems, gigantic data center, and a whole infrastructure which all consumes valuable resources like electricity and water. In the year 2024 solely, data centers consumed about 460 TWh of electricity globally, and the International Energy Agency projects that this could more than double by 2030. With the growth of AI, the environmental cost of computation can no longer remain invisible.

This is where I believe we need to ask: Are we building AI efficiently enough to justify the resources we are asking the planet to provide?

The problem is not AI itself. AI itself can help design more efficient energy systems, support climate research, and discover new scientific solutions. The problem is what happens when intelligence is built without enough consideration for efficiency. 

A major part of this problem is called bloated code.

Bloated code isn’t simply a long code. It’s a software carrying unnecessary complexity: redundant operations, inefficient algorithms, poor memory management, badly optimized data pipelines, or systems that use far more computational power than the task actually requires.

In conventional software, inefficiency may mean slower execution or a larger application. At AI scale, the consequences are much larger.

Imagine millions of users repeatedly asking relatively simple questions and every request being routed through an oversized model with unnecessary excessive computation, inefficient infrastructure, or an architecture that has not been designed around energy and resource efficiency. A single wasted computation may seem meaningless. But Constant dropping wears away the stone.

Water is one of the most overlooked parts of this conversation. Data centers generate heat and therefore require cooling. Depending on their design, location, electricity source, and cooling technology, data centers can consume a significant quantity of water. The environmental footprint is not limited to the electricity used for hardware production; it’s also associated to water consumption by cooling systems. A 2025 study estimated the real-world environmental impact of developing a series of language models, the consumption of water was 2.769 million liters of water, equivalent to about 24.5 years of water usage by a person in the United States [2]

Yet we should know: not every AI interaction has the same environmental footprint. This matters because the answer cannot be to fear every AI prompt, but stop treating computational efficiency as an optional engineering detail.

A well-designed AI architecture should ask: Does this task actually require the largest available model? Can caching prevent repeated computation? Can requests be batched efficiently? Can unnecessary context be removed? Can hardware utilization be improved? Can the workload be moved toward locations and periods with lower-carbon electricity or better water conditions?

These questions belong to software engineers, infrastructure architects, researchers, product designers, and technology companies. But responsibility does not end there, as users, we also shape demand. There is nothing inherently wrong with asking AI to help write, learn, research, translate, brainstorm, or solve a difficult problem. I myself use AI as part of my own technical and research workflow, But using a powerful computational system for every trivial decision simply because it is available is another matter.

We do not need to generate ten versions of something when one is enough. We do not need to repeatedly ask a model to perform work that could be completed once. We do not need to outsource every small thought to a machine merely because the machine is fast.

Efficiency begins with purpose.

These are not arguments against progress. They are arguments for better progress.

We should not create an artificial choice between technological innovation and environmental responsibility. The two should develop together.

The most dangerous idea may not be that AI consumes resources. We already know that. The dangerous idea is believing that resources are infinite because the computation feels invisible.

Every model has a physical home, and Every machine requires energy.

As engineers, we are not only responsible for making systems work. We are responsible for thinking about what our systems cost the world.

As users, we are not powerless either. We can learn to use AI intentionally rather than excessively. We can ask whether a task truly needs it, choose simpler tools when appropriate, and become more aware of the hidden infrastructure behind the convenience of an instant answer.

The future should not be a world in which we choose between intelligence and the environment.

It should be a world where intelligence is designed with the environment in mind.

I believe the next generation of AI should not only ask, “Can we build it?”

It should also ask:

“Can we build it better?”

“Can we build it with less?”

Because data may be digital, but its consequences are physical.

And when the code becomes too large, the models too heavy, and the architecture too careless, the cost does not disappear into the cloud.

It reaches the ground beneath us.

And ultimately, it can reach the water we all depend upon.

The goal shouldn’t be to stop AI. It should be making intelligence sustainable enough to deserve the future we are building with it.

References

[1] International Energy Agency (IEA), Energy and AI, 2025.https://www.iea.org/reports/energy-and-ai

[2] Morrison, J. et al., Holistically Evaluating the Environmental Impact of Creating Language Models, 2025.https://arxiv.org/abs/2503.05804

---

         The Writer's Profile

Verona Obeid

Electronics and communications engineer

Damascus university, Syria

Author's Bio:

Verona Obeid, Syria - Electronics and Communication Engineer and Front-End Developer with a deep interest in how people connect, communicate, and learn. I focus on exploring and teaching modern communication technologies, with a mission to make tech more accessible and empowering for women and communities.