ChatGPT Energy And Water Use Exposed

The Startling Resource Demands of Modern AI

Artificial intelligence (AI) is capable of extraordinary feats, but this power comes at a significant cost: energy. A vast amount of it. Research highlighted by Businessenergyuk sheds light on the sheer scale of resources consumed by models like ChatGPT, and the figures are staggering.

Every time you prompt Midjourney or Chat GPT to generate an image, an explanation, or an email, the servers of the host company engage in thousands of complex calculations. This computational effort translates directly into substantial energy consumption. To manage the heat generated and prevent server overheating, extensive water cooling systems are employed. These systems absorb thermal energy and release it through evaporation.

Data centers are at the heart of this consumption, requiring immense energy for two primary phases: the initial training of AI models and the subsequent processing of continuous user requests.

Related: Woman Starts Affair With ChatGPT: Divorced From Husband, Claims She's Never Been Happier

The Dual Drain: AI Training and Operational Energy

“Both phases are very energy intensive, and we don’t really know what the energy ratio is,” explains data scientist Alex de Vries. “Historically, with Google, the balance was 60% inference [answer generation], 40% training. But then, with Chat GPT, this kind of broke down because training ChatGPT required relatively little energy compared to running the model. The large language models that ChatGPT runs on are notorious for using huge datasets and having billions of parameters. And of course, the larger these models get, the more energy they simply require."

ChatGPTs Staggering Electricity Consumption

How much electricity does ChatGPT actually use? Assuming its 400 million weekly active users send an average of five queries to the chatbot daily, the numbers are eye-opening.

Daily Power Draw

ChatGPT consumes approximately 39.98 million kWh each day. To put this into perspective, that's enough electricity to:

• Charge eight million smartphones.
• Run the Empire State Building for 540 days.

Weekly Energy Footprint

On a weekly basis, this AI model uses around 279.86 million kWh. This is:

• Nearly four times the electricity consumed on Super Bowl Sunday (including stadium power, home TVs, and related travel).
• Sufficient to power the Las Vegas Sphere for almost three years.

Monthly Energy Usage

ChatGPT's monthly electricity consumption is approximately 1.199 billion kWh. This could:

• Power all the billboards in Times Square for over 20 years.
• Charge 332 electric cars.

Annual Electricity Demand

Annually, ChatGPT consumes an estimated 14.46 billion kWh. This figure surpasses the yearly electricity consumption of 117 individual countries.

The Hidden Water Cost of AI

The energy consumption is only part of the story. The water used by AI companies to cool their servers is freshwater – the same vital resource needed for drinking, cooking, sanitation, and firefighting. Worryingly, in the U.S., about one in five data centers draws water from regions already experiencing water stress.

Daily Water Intake

ChatGPT uses approximately 117 million liters of water every day. This is equivalent to:

• Filling 978,000 bathtubs – enough for every resident of San Jose, California, to have one.
• The entire population of Taiwan flushing their toilets at once.

Weekly Water Consumption

Each week, ChatGPT uses approximately 822 million liters of water. This is:

• More than Disneyland in Orlando uses in two weeks.
• Enough water to fill more than 864 million Stanley Quencher cups.

Monthly Water Footprint

ChatGPT's monthly water consumption is around 4 billion liters. This quantity could:

• Fill 1,780 Olympic-sized swimming pools.
• Provide everyone on the planet with two glasses of water.

Annual Water Usage

Annually, ChatGPT consumes approximately 42 billion liters of water. This is enough to:

• Fill Central Park's reservoir fourteen times over.
• Keep the Bellagio fountains running for 1190 years.