Behind the AI Curtain: Burning Water, Draining Grids, and Losing Our Minds

Over the last decade or so, since the invention of smartphones, “cloud” utilization slowly but steadily increased to large scale. Enter the “Artificial Intelligence” to take it further in terms of daily usage. Cloud doesn’t mean the data comes from sky nor it floats in the air freely!

Since the launch of ChatGPT, the focus has largely shifted to the magic of Large Language Models (LLMs). Given a prompt in the Google AI mode or ChatGPT, within seconds, you get back a detailed explanation to an article, it writes a marketing strategy, debugs your logs to find the programming errors or summarizes 1000’s of lines of health documents. It’s all frictionless!!

Most of us do not worry about what is happening behind the scenes. Focus is solely on getting the work done in shorter duration for better efficiency. But have you ever asked a question to yourself if the AI revolution is really a simple software update or is there any magic involved? While you were encouraged by mega companies to leverage the magic trick of chatbots, behind the curtain, massive physical infrastructure was quietly drinking our fresh water, consuming power and most importantly rewiring our brains.

What is the true cost of your AI assistant?

Impact to the physical and mental world when you do something as simple as ask for a travel itinerary for a nice 4-day Orlando visit with the family:

• The Sweating Giant (Water Scarcity): Surely, all of us would have experienced burning hot laptops on the lap when too many tabs are opened. AI is like millions of those laptops running at full speed inside a hidden warehouse. To survive and not melt, they sweat, evaporating fresh water every day.
• The Power-Hungry Factory (Electricity): Most of us compare Google search and a ChatGPT prompt and assume they take the same effort. But they don’t! If Google search is like turning on a small LED light, asking an AI to write your code being an assistant is like turning on floodlights at a stadium.
• The Mental Crutch (Cognitive Health): I still remember my grandparents and parents used to teach me a route to get back home by drawing lines and explaining the surroundings. Once the GPS was on our phones, we completely lost our internal sense of direction, following the lines on the map. AI is no different and is doing the same to your brain. Just like how US outsourced technological requirements to developing countries, we are outsourcing our intelligence
• The Photocopy (Originality): Over time, a photocopy of a photocopy gets blurry and loses its sharp edges. AI needs training data, and it has been created by millions of users over decades. Once everyone stops thinking and relies solely on available data, there are no creators anymore, and you lose the originality.

1. The Sweating Giant: The Water Paradox

If you look deeper into the infrastructure, the servers run at blistering temperatures. Just like how we need AC’s to control temperatures when many computers are running at your IT office, data centers pump millions of gallons of fresh water through cooling towers.

To keep them from catching fire, data centers pump millions of gallons of fresh water through cooling towers where it evaporates into the atmosphere. Research from UC Riverside found that training GPT-3 alone consumed 700,000 liters of freshwater.

To put that into everyday terms: every time you have a standard conversation with ChatGPT (roughly 10 to 50 prompts), a data center somewhere literally consumes a small bottle of fresh drinking water.

The Big Concern: AI is still in early stages but the consumption is already seeing catastrophic results due to the blind consumption. In the Dalles, Oregon, Google built a massive data center that consumed more than a quarter of the entire city’s water supply even during the drought. Google did hide the water usage as a corporate secret, which later lead to massive public backlash and farmers filed lawsuits. Residents were asking to reduce the consumption where possible, but the machines kept drinking as in required.

• Direct-to-Chip (D2C) Cooling: To stop wasting drinking water, companies are moving away from the massive AC’s. Instead, they run tubes of liquid coolant directly over the hot computer chips. Many companies including Meta and Microsoft (Immersion cooling) are spending billions to completely redesign their data centers for this liquid cooling. On the other hand, Nvidia’s latest and greatest Blackwell chips run so hot that they cannot be cooled by air anymore; they require Direct-to-Chip liquid cooling to function properly.
• The Water-Energy Nexus: The magic problem tech companies are trying to solve : If they try to recycle the water instead of evaporating it, they need massive industrial refrigerators called chillers to cool that recycled water. By now, you should have guessed that it will require massive amounts of electricity. Yes, you either burn water, or you burn power.

2. The Nuclear Factory: The Grid Panic

Powering the Internet is a known game over the past decade or so. But with AI, the math is completely different. Going back to the Google search vs ChatGPT comparison, simple Google Search takes about 0.3 watt-hours of electricity. Compare that to single ChatGPT query; takes about nearly 10 times that amount, equaling to 2.9 watt-hours.

A recent joint study by Hugging Face and Carnegie Mellon University found that a single AI image generation takes as much energy as fully charging your smartphone. Now, think about this deeply — you are generating a funny cat face to send as emoji to your friend by using fully charged smartphone battery!

Is there no solution to this problem? Partially yes and many companies made flashy announcements about using renewable energy resources like solar and wind, but Big Tech realized that traditional power grid cannot handle hunger of AI. Believe it or not, the International Energy Agency (IEA) projects that AI data centers will soon reach the energy demand equivalent of the entire country of Japan by 2030.

The Mighty Change: Citizens are finally waking up and fighting back against these hungry beasts. Look at Michigan: Oracle and OpenAI recently tried to push through a massive data center in a small town (Saline, Michigan) that would consume staggering amounts of electricity. The local boards and residents rejected the project due to the insane environmental and power grid impact. Even though developers try to force these through, the backlash has been so severe that municipalities are freezing the approvals. However, in this case Oracle was able to move forward with the construction by detailing out how Data Center doesn’t really impact water usage and energy usage, but rather provides jobs to the locality.

• The Nuclear Renaissance: To solve this, AI is trying to go nuclear. Microsoft recently signed a 20-year deal with Constellation Energy to revive the infamous Three Mile Island nuclear plant exclusively to power its AI data centers. Amazon partnered with Talen energy to plug Susquehanna nuclear power plant to it’s data center. While nuclear is clean, building new Small Modular Reactors (SMRs) takes years.
• The Alternative Menu: So, what about renewables? Why not just use solar or wind? To run AI on solar, you would need skyscraper-sized batteries to store power for the night. Tech giants are desperately exploring extreme alternatives to feed this beast — from OpenAI’s Sam Altman investing in Nuclear Fusion (creating a “star in a jar” via Helion Energy), to futuristic dreams of Space-Based Solar panels that never experience nighttime. If you want to dive deeper into how tech giants are trying to build a stomach that can digest the power of a star, check out my deep-dive article here

3. The Autopilot Trap: The Death of Critical Thinking

Have you noticed your colleagues write or refine complex emails? Or optimizing a poorly performing SQL query? Or create an agent for simple but repetitive tasks? The friction is exactly how the human intelligence grows — your brain is building neural pathways.

And there is a strong reason why you would use LLM, when you could do the same but repetitive tasks with a prompt rather than sitting and struggling for an hour — The productivity matters! But when you put the mind on autopilot mode, human engage in “Cognitive Offloading”. To explain simply; though we can calculate numbers using our brains, we got used to offloading the task to the calculators.

The Big Mistake: When we offload the logic to a machine, we stop checking the work. A famous veteran New York lawyer used ChatGPT to write legal document for federal case. AI completely hallucinated and added fake past court cases, using fake quotes and fake judges as well. When your brain is not worried to verify the facts, the brain runs on autopilot mode leading to major mistakes. Many of us are worried about what if machines fail; ideally it’s human who fail not the machines.

• The Google Effect (Memory): Years ago, we offloaded our memory to the internet. Not many remembers the phone numbers and historical dates other than few key ones because the information is available online and we can search for them.
• The ChatGPT Effect (Logic): Well, if Internet can take you one step forward, AI can take you ten steps further. We are no longer offloading memory; we are offloading logic. You can see it in everyday life: have you noticed how every corporate email suddenly sounds exactly the same? Routine phrases like “In today’s rapidly evolving world,” “unlocking potential,” and “navigating complexities” make us feel like we are talking to Robots.
• The IT Nightmare: Look at modern software development. In 2023, researches found a terrifying real-world threat: “AI package Hallucination.” ChatGPT was confidently recommending old code libraries to solve the programming problems. Out of 400 questions, 100 responses included old references which do not exist anymore. Hackers were happy to play around and create fake names using the malware. Prompt Engineers (Young developers) blindly copy-pasted the code without fact checking.

4. The AI Cold War: Why is the World Okay With This?

With so many disadvantages; if AI is drinking our fresh water, consuming power, and more importantly playing with human brain power, why are Big Tech companies, the US and China okay with it? Why are we accepting the trade-off to create a next-generation “prompt engineers”?

Well, this is arms race, not just a tech advancement!

• The Race for AGI: Internet revolution created some of the largest companies in USA and China especially. Global superpowers are not worried about a spike in water bill; they are looking at global dominance. First nation to achieve the Artificial General Intelligence (AGI)- An AI that can out-think human brain will control the future of the global economy, cybersecurity and even combat field strategies.
• Collateral Damage: Few lazy “prompt engineers” are viewed as acceptable collateral damage when you are looking at revolutionary change. Losing the AI race can be a bigger threat.

5. The “Dial-Up” Phase: How We Fix the Machine

• The Early Internet (Brute Force): Remember, how any URL search on the browser used to keep spinning before opening up the website using 3G? Webpages used to load first and then the images would take forever to display within the webpage. Early stages will either be inefficient or learning experience. Look at where we are today, with a powerful machine in hands which can control the world. Fiber optics, 5G, edge computing, satellite internet, and many more…!
• The Early AI (Where we are now): Although it has been more than a couple years since LLMs became widely popular post ChatGPT launch, we are still in “dial-up” phase. And rightly so, we are all using brute force to solve basic problems. And we do not care about the massive compute required to find out a simple distinct query in Big Query.
• The Tech Solution: Are we going to consume energy from a nuclear plant to generate few comic images at this rate? Well, future always advances with or without distractions. Very soon, we all will have highly efficient AI models that run on your laptop or phone without needing the cloud to respond. Hopefully, they will use less power, less water and offer data privacy.

6. The Pioneers: Architects, Not Users

How do the current and future generations survive this? How do we sharpen the human brain?

History always guides us in right direction. When the Internet era started, Amazon expanded to multiple nations in few years and became one of the pioneers of the internet revolution with daily usage hitting record numbers and don’t even try to think how much they make every minute! Learn what is AI, not just how to use AI. Contribute to AI, don’t just leverage AI.

• Don’t Be a Prompt Engineer: A prompt engineer is just a user. If your only skill is asking a machine to do your job, you will eventually be replaced by a machine that knows how to prompt itself. Sounds funny, but true.
• Be an Architect: Anyone can copy and paste the code; an architect understands the foundation. You could use a Claude agent to write a Java Microservice, but when Cloud Run throws an Out-Of-Memory (OOM) crash, in production, it is the architect who can quickly relate the involved systems to understand the issue in depth. Tech industry isn’t in a need for people who can just find the issue using AI, they are searching for the architects who can train the AI.

Summary: The Magic Trick is Over

Right now, everyone thinks Google AI mode is just a magic trick. It summarizes and answers our basic questions like “ who scored most number of goals in FIFA world cup” and “summarize the document” or “write an email to my manager”

• It isn’t just answering a prompt; it is evaporating millions of gallons of our drinking water.
• It isn’t just generating an image; it is draining our power grids.
• It isn’t just fixing our grammar; it is putting our critical thinking on autopilot.

You don’t have to be a software engineer to see the danger. Think about everyday life; If you use AI to write a reply to your friend, because you didn’t struggle to find the words, you skipped the emotional friction required to actually mean it. Yes, the machine solves your simple question, but you are losing the critical thinking.

The winners of the next decade won’t be the people who outsource their mind to the hidden force, but the ones who thrive to use AI to do their work faster with their smart brain