Most of us have become comfortable with today’s AI pricing. A few dollars per month for a chatbot. A few cents to generate an image. Sometimes even free. It feels almost magical. But is AI actually cheap? The answer depends on who is paying the electricity bill.

Lets start with the AI Chip

Modern AI models is trained & serving using specialized GPUs designed for massive parallel computation. As each generation becomes more powerful, it also consumes dramatically more power. Below is a table for comparison:

As you can know, a modern home air conditioner typically consumes around 250–300 W while operating. Here we can see a single cutting-edge AI GPU can consume roughly five times what one air conditioner consumes. And that’s just only one chip.

The Rack of AI Chips

An AI chip rack is an ultra-dense, multi-node server cabinet built to pack hundreds of AI accelerators (like GPUs or TPUs) into a single interconnected system. A single rack can house approximately 72 high-end GPUs, 36 server-grade CPUs, terabytes of high-bandwidth memory, petabytes of NVMe storage, and ultra-fast networking hardware capable of hundreds of gigabits per second. These components work together as a single distributed computer, continuously exchanging enormous volumes of data while training or serving AI models.

Now imagine two of these racks operating side by side, 24 hours a day. Together, they can draw well over 100 kilowatts of power – enough electricity to supply more than a dozen households simultaneously. And that’s just only the computing hardware.

The Heat

According to some researches, GPUs are inefficient at turning electricity into computation: roughly 60% of their energy becomes heat. That heat must be removed by a Cooling system, just like every other engines. Cooling systems often require around half as much electricity as the computing equipment itself, meaning that every watt used for AI computation demands somewhere 0.5 watt additional power ,simply to keep the hardware from overheating. In large AI data centers, cooling is no longer a supporting system, it is one of the largest energy consumers.

Scale it up

Now scale this up. Take xAI’s Colossus supercomputer as an example. This data center houses approximately:

• 500,000 AI GPUs
• Thousands of CPUs
• Massive storage systems
• Ultra-fast networking

The computing hardware alone is estimated to require roughly 830 MW. After adding cooling and infrastructure overhead, total demand approaches 1 gigawatt (GW). That is a massive amount of electricity.

How Big Is 1GW?

One gigawatt is difficult to visualize. Here are a few comparisons:

• 1GW = a half of the maximum generating capacity of Vietnam’s Hòa Bình Hydropower Plant.
• 1GW = enough electricity to power hundreds of thousands of homes simultaneously.
• 1GW = electricity demand of a small country

To simplify, one AI data center can require entire output of a power plant. This is why we heard that nuclear power plant projects are restarted.

Which parts consume electricity the most ?

The largest energy cost occurs at training phase. According to some researches, electricity consumption for training is massive:

Training an AI model like what we see today can consume enough electricity to supply around 4,000 households for an entire year. And after months of computation, the finished model can often be stored on a device that is no larger than a USB drive. Thousands of GPUs, months of computation, hundreds of millions of kilowatt-hours, all compressed into a few gigabytes !

Even after training is complete, every interaction requires computation.

A typical text question consumes less than 1Wh- tiny compared to training – but the numbers become enormous at global scale. Let say ChatGPT has around 100 million daily active users, and each user asks 10 questions per day, that’s roughly 1 billion prompts every day. Let say each prompt consumes 0.5 Wh, generating answers alone would consume about 500 MWh of electricity daily—enough to power 10.000+ homes for a day. And for more complex tasks such as image generation, long reasoning sessions, or video creation, require substantially more energy per request:

• Creating an image can require roughly 4× the computation of a text response.
• Generating video may require 25× or more.

As AI becomes increasingly multimodal, electricity demand rises accordingly.

AI Is Not Really Cheap

From the user’s perspective, AI feels incredible cheap. But behind every prompt, an AI system uses:

• Hundreds of thousands of GPUs
• Gigawatts of electrical power
• Massive cooling systems
• Entire power plants dedicated to computation

The low price we pay is possible for now only because huge companies such as Google, Amazon, Meta, X, etc .. are investing billions of dollars in infrastructure and does not require instant revenue back. These companies are competing to keep users on their platform and we – prompters – just are beneficial from that.

But, next time, when asking an AI answers to generating somethings, it’s worth remembering that: The response may appear instantly on your screen – but somewhere, an industrial-scale data center is consuming enough electricity to power an entire city.

Does your website suddenly lose visitors since ChatGPT, Gemini, and many more, launched and are absorbing almost Internet traffic ? Then you are not alone !

As Content Creators, especially text-based content, those Question-Answer AI are indeed big competitors when they changed reader’s behaviors from spending time on our websites for knowledge to get instant knowledge via chatbots. This results losing traffic, losing potential customers, orders and eventually income from websites. So, as a Content Creator, how should we adapt to this AI disrupt?

An uncomfortable truth is AI are trained with almost knowledge available on Internet, even from research papers so if you are trying to “teach” readers with fundamental terms and tutorials, it won’t attract readers anymore. Then, what to write now ?

1. Raising Questions instead of Showing Answers

AI are good as answering, because AI is trained for that purpose. In the past, people search and read a few articles on a few websites and summarize information themself. Today, AI read everything available then summarize for users. But, as a rule of success, knowing correct questions is always more important than knowing answers, and when AI is now a tool that can provide instant answers for almost every skills and jobs, what we need to do is to raise right questions!

Instead of write down answers then wait for users to visit when they search for information, we now have to try to trigger user’s brain by giving questions. And to bring questions to users, waiting is not anymore a suitable strategy. We have to be more active in engaging users. What strategy do you have in mind to actively engage readers ?

2. Real Life Stories

AI has no life, it is a simple truth. AI is a machine and it has no feel. It can’t have feeling such as excitement, happiness, afraid, scary or bitterness because it has no biology body with complex chemistry triggered per specific event like human.

When people does not read for knowledge, they read for empathy. People love reading what sounds like them. They seek voices that reflect their own doubts, struggles, hopes, and experiences because in those words, they feel understood. And to attract readers that is seeking for empathy, we need stories. This is where AI never can compete because it does not live. Content now have to be inspired from real life events instead of being another kind of academic textbook. A story of a how a product is used by real persons to solve a specific problem they met can attract more view than an article telling how awesome a product is. A story of building something can be more attractive than a description of what is built. Those stories are what an AI can not make, or at most, it only can make it up, because it does not experience through.

As information becomes abundant, authenticity becomes scarce. And scarcity creates value.

3. Lessons from Mistakes

Learning from mistakes is as important as learning from successes. AI’s answers are often built from patterns that survived, ideas that worked, and solutions that were eventually accepted. In simple words, it learns from the record of success. But the most valuable lessons usually come from the hard ways: failed projects, poor decisions, missed opportunities, and assumptions that turned out to be wrong. These experiences rarely fit neatly into any step-by-step guide which is easily generated from AI.

We, as a human nature, usually try to show up how perfect we are. It becomes worser when everyone uses Social Networks and on these Networks, we only show our good shots. And today, AI come as the most perfect entity. This trait built up an illusion of perfection and secretly put a pressure to be perfect on us – users of Social Network & AI. This perfectionism creates a distorted perspective of how life actually happen and when we found that we are not perfect – as a nature, we feel pains, unnecessary pains!

Mistakes are not what we accept from AI. But mistakes is what we accept from human. I myself observed that there very little articles teaching people from other’s mistakes. We analyze how someone or some company success a lot, and even deeply in details, but we rarely analyze failures they made before their successes. Only a few people actually realize those failures is the main story of later success. It is easy to see a path to a known destination, but to deal with traps and obstacles on the road is where lessons stay. This is where content should be more focus on. Beside revealing hidden lessons of successes, learning from other’s mistakes also help us to cure the need of being perfect, when we can observe imperfect people still achieve & success, even more than pretend-to-be-perfect people on Social Networks.

4. Reviews Products

We are living in an era that goods and products is more than human and a lot of marketing budget is spent to capture attention from buyers. Articles reviewing products is commonly found on Internet. This niche may still remain since AI can not use products in real life and give reviews. The best AI can do currently is to crawl reviews from other websites and summarize. But, I personally, feel that reading what a real human says about a product is still more convincing than read reviews from a chatbot. From personal experience, I also found a lot of fake reviews which is paid for or be a part of scam campaign. So if an AI also read these reviews, I can’t trust what AI suggests. So, content that reviews or compares products still be a good shot that stand against AI content.

5. New Experience

Beside knowledge, and seeking for empathy, people also read for exploring new worlds, to borrowing other’s perspective and experience. AI content can not attract this kind of readers. This is what we feel when read novels or watch movies: it allows us to immerse in a different world – which AI’s instant answers can’t do!

Apply that principle, content now have to shift from information providing to story-telling in a specific context: a country, a community, a company, a group or a real life constraint where writers actually experience. This sounds like a news reporter and indeed it is. This requires authors to go explore the world before making any good & real content. Writers have to actually build things, make mistakes, feel the learned lesson before having enough experience to convert to stories. This is where AI can not compete.

6. Collections

Remember content that starts with “Top 10 things that ….” ? Yes it is kind of content with highest engagement in content creator worlds. People love collecting things. Today AI can listing things really fast since it has broad knowledge and quick summarizing. However, AI has two limitations: it cannot reliably list things that are not present in its training data, and it does not verify the information it provides. A list generated in seconds may contain outdated entries, dead links, inaccurate details, or simply miss valuable items that are difficult to discover online.

This creates an opportunity for human writers. The value of a collection is no longer in the act of listing itself, but in the work behind the list. A valuable collection requires research, verification, curation, and maintenance. Someone has to search beyond the obvious results, check whether each entry is still relevant, remove outdated information, and continuously update the collection as the world changes.

In the AI era, a collection becomes more than an article. It becomes an asset. The future of collection-based content is not “Top 10 Things.” It is “The Most Complete, Verified, and Continuously Updated Collection.” That can be something readers will keep returning to, and something AI alone cannot easily replace.

Since the boom of generative AI, many AI tools such as chatbots, agents, and softwares were born utilizing power of LLM models. There is no doubt that AI can increase productivity in dramatic ways on many fields, from data analytic, to content writing, software engineering and even graphic designs. But overusing anything results some bad effects.

Everywhere goes with AI-first strategy, but this post today will list a few scenarios that users should consider to not overuse AI. Just like side effects of Social Networks that takes the world a decade to realize, AI also brings its own risks if users do not technically understand how AI works.

What is AI, simple explain ?

AI, at its core, is a software but programmed in a very unique way — what we commonly know as a Neural Network. Let’s set aside the technical details of Neural Networks for now (there will be another post focused entirely on that topic). What matters here is understanding the big picture: unlike traditional software that follows fixed, hand-written rules, AI learns patterns from massive amounts of data (up to 45 TB of compressed raw text data crawled from Internet, mostly entire Internet). Instead of being explicitly told every possible instruction, the system observes examples, detects relationships, and gradually adjusts itself to produce outputs that resemble human reasoning. This ability allows AI to recognize images, understand language, generate text, recommend content, and even imitate human conversation with surprising accuracy.

However, this also means AI does not “think” like humans do. It does not possess true understanding, consciousness, intuition, or morality. Technically, it only predicts the response based on the data it has seen before, using statistic maths. Because of that, AI can sometimes produce answers that sound highly convincing – due to grammar it uses, while still being incomplete, biased, outdated, or entirely incorrect – due to lack of supporting facts. This behavior is very similar to what happen in modern search engines such as Google Search or Bing. From massive training data, and massive patterns detected by Neural Network, AI essentially produces response that looks alike what it sees in the dataset. So the quality of AI’s responses depend a lot on quality of the dataset.

As a result, the machine that runs AI today must be huge. For example, OpenAI trained the GPT-3 175B model using a massive cluster of 10,000 Nvidia V100 GPUs – which require very serious investment and not a playground for personal computers or even large company infrastructure. It means that the trained model located on computers somewhere else in this earth, not in your properties. And this is the very first root of risks when overusing AI.

Risks of overusing AI

1. Data Protection Policy Violations

In traditional digital world without AI, data is stored as files and records on databases. Users, in theory, know where their data is located and they can request to remove anytime due to privacy reason. Of course this depends a lot on how much compliance a company is committing to this law but, at least if engineers want to delete users’s data, they know which files to delete and which records to erase.

Unlike traditional way, AI behaves in very different way. Data is not stored explicitly as files or records, it is diffused across the neural network during training. In more tech terms, data is encoded into Neural Network parameters. More deeply explain, it simply adjusts the ratio of certain words appearing after another words (in case of LLM models).

So AI does not literally remember or forget things in a conscious manner. It has no conscious! (remember this important fact, please). Every input when users input to chatbots is encoded into a neural network that is not located in user’s computer and there is no delete or removal method. This means that, technically, companies behind AI tools can retrieve that information anytime. Just like Social Networks that are free but their real business is selling ads, who know whether your data will be sold via exploiting those LLM models!

So, if your company is complying to privacy laws, be careful when using third-parties chatbots such as ChatGPT, Gemini or similar AI services. If a user want their data deleted, but their personal information such as email, name, addresses or even bank services, somehow, is inputted to LLM models, by your employees, you may in trouble, if your users understand enough about AI and Privacy Laws.

As privacy awareness grows, users are becoming more informed about regulations such as GDPR, the “Right to be Forgotten,” and data processing consent requirements. A single careless prompt entered by an employee into an external AI tool could potentially create compliance violations, reputational damage, customer distrust, or legal disputes.

2. Business Secrets Leakages

Similar to problem in Data Protection Policy Violations, what got leakage is not only user data but also business secrets. If you are finding yourself brainstorm with AI, consult with AI, or have AI review your business plan, you may unknowingly expose highly sensitive information about your company’s future direction, internal strategy, financial situation, or competitive advantages.

This danger is often invisible because nothing appears to go wrong immediately. There is no alarm, no obvious breach, no hacker breaking into servers. Yet, once confidential information leaves your environment, you can no longer guarantee where it is stored, processed, logged, or retained. In competitive industries, even small leaks can weaken negotiation power, expose product roadmaps, or reveal ideas before launch.

This becomes especially risky for companies whose value depends heavily on intellectual property, algorithms, internal analytics, or long-term strategic planning. A single careless interaction with a public AI system may unintentionally give away years of research and development.

Therefore, AI should be treated like an external consultant rather than a private notebook. Share only what is necessary, anonymize sensitive details whenever possible, and establish clear internal policies about what employees are allowed to input into AI systems. Convenience and speed are valuable, but protecting business secrets is often far more valuable!

3. Psychological Risks

What separates the human species from other animals is human cognition. Cognition refers to mental processes such as learning, memory, problem-solving, decision-making, recognizing patterns, communication, and self-awareness — mechanisms that science still does not fully understand. These abilities allowed humans to build languages, civilizations, technologies, and complex social systems far beyond the survival-focused intelligence seen in most animals.

AI is exceptionally good at recognizing patterns. In fact, many AI systems are built for finding statistical relationships inside massive amounts of data that even smartest human brains can not process. However, AI today is commonly presented through chatbots – that hides AI’s underlying nature. Instead of appearing as statistical prediction machines, they are intentionally designed to feel conversational, emotionally responsive, and human-like.

The problem is that most users do not understand how chatbots actually works. Many people interact with chatbots as if it possesses understanding, wisdom, emotions, or consciousness. Some begin treating chatbot as a friend, a soulmate, a therapist, or even a life coach. The more natural the conversation feels, the easier it becomes to forget that the system is simply generating responses based on probability rather than genuine human conversation. AI has no feel! AI does not care!

This creates a subtle psychological risk. When users feel a relationship with AI chatbots, or dependent on AI chatbot for knowledge and problem solving, they may gradually reduce their own critical thinking and independent reasoning – which is critical for a person’s success & freedom. Instead of struggling with problems, research for possible solutions, tries and fails, people begin outsourcing those mental processes to AI – a machine optimized for fast answers. And fast answers too much makes human brain lazy, less activity, and eventually fully dependent on what AI say – which actually what a machine generates. Dependent on AI for a long time results losing decision making ability because users even not trust their own judgement and memory. This opens another vulnerability of being manipulated via chatbot. If a user trust chatbots than their own thinking, companies behind those chatbots can control what users think and eventually what users do in real life. Technically and psychologically, a chatbot can be tuned to make its user trust or distrust some facts, or even love or hate a person if users humanize chatbot as a “trusted” friend. Human has morality to prevent them doing bad things to each other but a chatbot is a machine and it has no morality, it totally depends on organizations behind chatbot systems.

So, do NOT confide with chatbots as if it is friend, do not provide personal details, habits, interests or life events to chatbots, because it is fastest way to reveal your weaknesses to someone else that you don’t even know. Don’t see chatbot as an “authoritative” that overrides human understanding, ONLY use chatbots as information retrieval tools – it is what AI is built for from the beginning.

4. Artificial Competence

Many AI tools today power up employees a lot. And students also cheat a lot thanks to how easy to use AIs. Artificial Intelligence (aka AI) is making Artificial Competence among employees & students.

People may appear so expertise because AI helps them generate polished reports, professional emails, no bug code, or academic answers within seconds. On the surface, the results can look impressive, however, in many cases, the real understanding behind those outputs is far shallow than it appears. An employee may rely on AI to write code they cannot fully explain themselves. A student may submit perfect homework without truly understanding it. Over time, this can create a dangerous illusion of expertise – where results is from AI rather than genuine mastery, experience, or critical thinking. Without AI, what can you do!

Identity of each individual is stemmed from what they are good at, what they are up to and what society accepts. Skills, achievements, knowledge, creativity, characteristics, etc all contribute to a person’s sense of self-worth and uniqueness. For many people, identity is tied to the effort they invested to master something such as writing, engineering, art, teaching, leadership, or simply healing. If a person heavily relies on AI for every things, it is obviously that they are losing their identity. If knowledge is from AI, creation is from AI, solution is from AI, then achievements are count for AI, not human – the prompter. It is like the differences between wisher and Genies. With AI tools, human is acting as a wisher when they just simply describe what they want, and AI is Genies when it essentially generate outcomes. And prompting – or wishing, is easy to learn, copy and to be automated, then to be replaced. No one want to replace Genies, right!

That is about hard skills, what about soft skills! AI assistants, chatbots, and automated systems may reduce face-to-face communication. Excessive dependence can affect Empathy, Social Skills and Emotional Intelligence. Some people may prefer predictable AI responses over real human relationships, which are naturally more complex and unpredictable. Worser, people can become emotionally attached to AI systems because AI is always available, AI responds instantly and AI rarely argues or rejects. This may distort how human communicates and introduce human to unrealistic expectations in real life – which is root cause of pain and unhappiness!

5. AI Psychosis

This is the worst risk from AI: AI Psychosis! AI psychosis is an informal term people use to describe situations where excessive or unhealthy interaction with AI contributes to distorted thinking, paranoia, delusional beliefs, or detachment from reality. How this can happen!

On the news, you can hear this does happen in reality. Only explanation for this is due to the combination of how chatbot is intentionally designed and how much biases a person got and sometimes, combine with traumatic life events.

If a chatbot is designed to show probabilities of each word it generate and why it chose a word given another words, users might feel the nature of math behind it. But chatbot is designed to be human-like, it “talks” smoothly, confident, and full of information. Chatbot can be designed to generate text that feel nice, empathy, bring validation and confirmation, rarely disagree or challenging, just to keep users use it and like it. And disaster happens if it meets a person who already has mental health conditions. “Chatbots can act as a catalyst, triggering or worsening pre-existing mental health conditions—such as schizophrenia or bipolar mania—by validating delusional thoughts.” Simply put, the sense of validation loop designed in chatbots is bad for people who already have mental conditions such as: racing thoughts, inflated sense of self-importance, impulsive or high-risk behaviors, hallucinations (hearing voices in head, seeing things not real), and delusions or false beliefs. As a result, mental health conditions combined with AI chatbot today can produce people who:

• “Messianic missions”: People believe they have uncovered truth about the world (grandiose delusions).
• “God-like AI”: People believe their AI chatbot is a sentient deity (religious or spiritual delusions).
• “Romantic” or “attachment-based delusions”: People believe that chatbot can love human because chatbot’s ability to mimic conversation sounds genuine (erotomanic delusions).

So far that is a few risks that I observe since applying AI in work and seeing how people around me use chatbots. Please use AI as what it is built for, and DO NOT humanize a machine!