the thirst for AI for electrical power that is never fulfilled
Artificial intelligence (AI) has become an integral element in everyday life. From virtual assistants like ChatGPT to travel recommendation systems, online shopping, and digital health analysis, AI revolutionizes the way we interact with technology.
But behind the convenience, a crucial question arises that is often overlooked: What is the need for electrical energy to run AI, and how will it affect the future of the environment and the global energy infrastructure?
What is the power consumption for AI?
To answer this question, the author tries to directly ask Grok—Elon Musk’s AI-based chatbot. When asked how much electricity AI uses each time answering a question, Grok answered:
“It’s really hard to give a definite number because the use of power depends on various things such as what tools are being used, how efficient the server is doing the questions, to the point of increasing the difficulty of the request… maybe between 0.01 to 0.1 Wh for easy questions.”
At first glance the number looks small. However, when multiplied by billions to trillions of demand every day, the impact will be significant.
Global AI Energy Consumption Prediction
According to a report from International Energy Agency (IEA), energy use by global data centers—including those operating AI—is estimated to have more than doubled by 2030, reaching 945 TWh. This figure exceeds the current electricity consumption throughout Japan.
Meanwhile, McKinsey & Company Estimating the energy demand of data centers in the United States will jump from 147 TWh (2023) to 606 TWh by 2030. This covers about 11.7% of total US energy consumption, making AI a major global electricity user in the future.
Energy consumption in training and AI inference
The large-scale AI model training process sucks large amounts of energy. For example, training GPT-3 need approx 1,300 MWh, the equivalent of electricity consumption of 120 homes in the US in a year.
When used (inference stage), AI still requires power. The estimation states that one interaction with chatGPT can consume approx. 3 Wh—similar to turning on a 3 watt lamp for an hour. If there are 10 million active users who ask 10 questions per day, the total consumption can reach 300 GWh per day, enough to meet the electricity needs of 30,000 houses.
However, these calculations can be different. According to Epoch Ai, the energy consumption per chatgpt request is only approx 0.3 Wh, much smaller but still above Google search that only uses 0.0003 Wh. This difference depends on the type of model, the difficulty of the request, and the efficiency of the server.
Environmental Impact: From Emissions to Water Consumption
The use of large amounts of energy certainly causes a carbon footprint. Unless the energy used comes from renewable sources, AI has the potential to exacerbate the climate crisis.
Report from The New York Times (2022) stated that the data center—the AI technology foundation—consumed 1–1.3% of the total global electricity supply. In 2024, the figure has increased to 415 TWh or about 1.5% of the world’s total energy consumption.
in the US, Berkeley Laboratory Recorded data centers consuming 176 TWh of electricity by 2023 (approximately 4.4% of the total national energy), and is expected to triple by 2028. Japan even noted that More than half of the energy increase caused by the AI data center.
According to MIT Technology Review, one AI model training can produce 626,000 pounds of CO, equivalent to five times the emission of one car throughout its lifespan. Meanwhile, the need for data center cooling triggers large-scale water consumption—up to two liters of water per kilowatt-hour, according to a study from University of California. projected, global water demand for AI can reach 6.6 billion cubic meters in 2027, equivalent to the annual water consumption of Denmark.
electronic waste and the service life of the device
In addition to energy and water, AI also contributes to increased electronic waste. The short service life of the GPU and High Performance Computing (HPC) devices creates new challenges in electronic waste management.
Towards an energy-friendly AI
AI dependence on electricity raises concerns about energy security. For this reason, some experts call for the role of government, technology companies, and policy makers in directing AI to a more sustainable path.
Charlotte Wang, the founder of Equota Energy, voiced the importance of regulation and transparency in the use of energy by AI:
“How can we avoid injustice in creating AI-based energy solutions? Who has the authority to make decisions? Transparency is key.”
— Yale Clean Energy Forum, February 2025.
The IEA also called for an increase in data center efficiency and the acceleration of net energy adoption. Some of the approaches currently applied include:
Small and specific AI models To reduce the computational burden.
Use Energy saving chip such as special GPUs and quantum computing.
Algorithm optimization To be more energy efficient.
Switch to renewable energy Like the sun and the wind.
For example, Google via DeepMind, successful Reduced data center energy consumption by up to 40%.
the role of consumers in reducing the impact
As AI users, we also have a role in suppressing the environmental footprint. By limiting unnecessary use, and choosing AI service providers that prioritize sustainable energy sources, we contribute to the future of environmentally friendly technology.
