AI, Energy Use, and the Question We Should Really Be Asking

Artificial Intelligence is rapidly becoming part of everyday life. From writing emails and generating images to analysing documents, coding software, and answering questions in seconds, AI is transforming how businesses and individuals interact with information.

But alongside the excitement comes a growing concern:

How much energy does AI actually consume, and is it significantly worse than a “normal” internet search?

The answer is yes, AI generally uses more energy than a traditional search engine query. But the reality is far more nuanced than many headlines suggest.

A traditional internet search is relatively lightweight. Search engines primarily retrieve indexed information that has already been processed and organised. The system identifies relevant pages, ranks them, and returns links in fractions of a second. While this still requires vast global data centre infrastructure, the computational effort per search is comparatively small.

AI systems work differently.

When you ask an AI a question, particularly a large language model, the system is not simply retrieving stored answers. It is actively processing billions of parameters, predicting language patterns, generating responses word by word, and often handling significantly larger quantities of data in real time. If the request involves image generation, video creation, coding analysis, or complex reasoning, the computational demand increases dramatically again.

This additional computation requires electricity. Large AI models run on high-performance processors, often GPUs, designed for massively parallel mathematical calculations. These chips consume substantial power and generate significant heat, requiring equally energy-intensive cooling systems within data centres.

The scale becomes important when millions or even billions of AI requests are processed every day.

Some estimates suggest a single AI query may consume several times more electricity than a conventional search engine query. However, the comparison is difficult because not all searches are equal. Asking for a weather forecast is very different from generating a 4K AI video or analysing thousands of pages of technical documents.

However, one important point is often missed in the discussion around AI energy use:

Efficiency is not just about the electricity consumed by the computer system, it is also about the efficiency of the human using it.

A single AI query may return exactly the information the user needs in one interaction. Without AI, the same task might involve ten separate internet searches, opening multiple web pages, reading conflicting articles, watching videos, comparing sources, and spending thirty minutes or an hour gathering the same understanding.

The energy comparison should therefore not simply be “one AI search versus one internet search”.

In many cases it is “one AI interaction versus an extended period of internet activity and human time”.

That human efficiency matters.

Businesses increasingly use AI to accelerate work that would otherwise consume hours of staff time. AI is now commonly used for data analysis, text generation, searches, reports, articles, coding support, image creation, spreadsheet formulas, dataset reviews, email drafting, and document formatting.

I increasingly use AI in my own work in exactly this way.

For example, I may draft the technical points and structure of a LinkedIn article myself, then use AI to expand it into a clearer narrative style, improve readability, correct spelling and grammar, and reshape the text into a consistent format that reflects the writing style I have gradually “trained” the AI to produce.

So is it my article or an AI article?

In reality, it is still my article.

The ideas, technical understanding, themes, conclusions, and direction come from me. AI simply helps turn rough technical notes into polished communication more efficiently. It becomes a productivity tool, similar to spellcheck, calculators, CAD software, spreadsheets, or photo editing tools.

The same applies to many other business activities. I use AI to help find Excel formulas, review datasets, generate supporting images, restructure emails, summarise information, and process technical data more quickly. These are tasks that would otherwise consume significant amounts of time, time that can now be redirected into higher-value engineering, consulting, analysis, or client work.

Of course, none of this removes the need for human judgement.

Whether information comes from AI or a traditional search engine, results still require a sense check. Facts should be verified, assumptions challenged, and primary sources reviewed where accuracy matters. AI can confidently present incorrect or incomplete information just as internet searches can surface unreliable websites or biased opinions.

The responsibility still sits with the user.

Training a modern AI model can also require enormous amounts of electricity over weeks or months across thousands of processors. This is where much of the environmental impact occurs. Once trained, individual user interactions are far less energy intensive, although still generally higher than traditional search.

At the same time, AI may also reduce emissions elsewhere.

Businesses are already using AI to optimise logistics routes, reduce heating and cooling demand in buildings, improve industrial efficiency, identify energy waste, automate repetitive office tasks, accelerate scientific research, and improve renewable energy forecasting. In many cases, the carbon savings enabled by AI could outweigh the emissions associated with running the systems themselves.

The challenge is that society is still in the early stages of understanding where AI creates genuine value and where it simply adds computational overhead.

Do we need AI-generated summaries for every search result?
Do we need AI-generated images for casual entertainment?
Should every software product include AI features by default?

These are questions of proportionality, not just technology.

The energy sector itself faces a similar balancing act. Electricity demand from data centres is growing rapidly, and AI is becoming a visible contributor to that growth. This is increasing pressure for faster deployment of renewable energy generation, grid upgrades, energy storage, and more efficient computing hardware.

Interestingly, the AI industry is already pushing processor manufacturers toward greater efficiency because electricity has become a major operational cost. Just as LED lighting dramatically improved over time, AI hardware will almost certainly become far more energy efficient in the coming years.

There is also a broader historical perspective worth considering.

Every major technological shift has increased electricity demand initially. Railways, refrigeration, telecommunications, computers, and the internet all expanded energy consumption. Yet many also delivered enormous productivity gains that reshaped economies and improved living standards.

AI may follow the same path.

The real question is not whether AI uses energy. It clearly does.

The more important question is whether the value created by AI justifies the energy consumed, and whether society can power this next wave of digital infrastructure using lower-carbon electricity sources.

AI is here to stay.

Like all new tools, it takes time to understand how best to use them, where they genuinely add value, and where their limitations are. The technology can assist, accelerate, organise, and automate, but ultimately the question of who remains in charge is still answered very simply:

The user.