For some, the AI revolution is already here. With around $20 a month, you can have an AI chat assistant at your beak and call, ready to answer any question that you have. Some are better at writing code that runs the applications that you use every day, and some can plan your holiday trip to Italy. There are a lot of startups that rely on these AI chatbots to be “smart” on your app. But what is the true cost of using those AI chatbots in the first place? Are they ready to replace humans?
A lot of energy
An AI model that takes your questions processes them, and answers them most convincingly takes a lot of energy. I mean a lot. Amazon and Google are planning to build nuclear power plants to fuel their AI data centers. Many big tech companies have failed or abandoned their environmental goals to meet AI demand.
It is reported that ChatGPT uses 500 thousand watts of electricity per day. ChatGPT daily usage can power 180,000 US households.
Microsoft energy usage more than doubled from 11 TWh (terra Watt Hour) in 2020 to 24 TWh in three years. Google will use 26 TWh in 2024. Not to be left behind, Facebook invested in AI technology which increased their electricity usage from 5TWh in 2019 to 11 TWh in 2022. One watt of energy is equivalent to a full meal. So each query is equivalent to drinking a glass of water.
Worth It?
To get the answer, we need to boil down how much energy it takes for the model to answer a single question, on average. According to some estimates, ChatGPT processes around 10 million queries per day. That equates to 416 thousand queries per hour. Let’s round that up to 420 thousand to make things easier to calculate. With around 20 kWh of energy used per hour, each query will use around 0.04 watts of energy on average.
Of course, an average human can answer more than one question over a glass of water. But in some cases, it will take some energy to answer a complex question.
Conclusion
Currently, AI is in the growth phase. In some situations, it is a solution looking for a problem. AI might help to better process large amounts of data more than a human can, for example, finding out what will proteins fold into or the best way to create a nuclear fusion reaction at the lowest costs.
Case in point with computers. The early days were a room size computers that used the electricity of a small town. Today, you can get supercomputer power in a base laptop for under $1,000. Web pages in the early days were crude and barely functional. Today, you can play games in a browser that would require a powerful computer when the game was released. Who knows what kind of AI we will have in 20 years?
Plug
Support this free website by visiting my Amazon affiliate links. Any purchase you make will give me a cut without any extra cost to you
| Base | Pro | |
|---|---|---|
| iPhones | iPhone 16 / iPhone 16 Plus - (Amazon) | iPhone 16 Pro / iPhone 16 Pro Max - (Amazon) |
| Watch | Apple Watch SE (Amazon) / Apple Watch Series 10 | Apple Watch Ultra 2 (Amazon) |
| AirPods | AirPods 4 (Amazon) | AirPods Pro 2 (Amazon) / AirPods Max (Amazon) |
| iPad | iPad 10 (Amazon) / iPad Mini (Amazon) | iPad Air M2 (Amazon) / iPad Pro M4 (Amazon) |
| Laptops | MacBook Air M3 (Amazon) | MacBook Pro M3 (Amazon) / MacBook Pro M3 Pro/Max (Amazon) |
| Desktop | Mac Mini (Amazon) / iMac (Amazon) | Mac Studio / Mac Pro |
| Displays | Studio Display (Amazon) | Pro Display XDR (Amazon) |
Other Ecosystem Items