Rapid advancements are being made in the field of Generative AI, which is currently soaring to new peaks, notably following the unveiling of OpenAI’s latest AI model, GPT-5. This innovative technology showcases state-of-the-art capabilities not only in programming and writing but also in healthcare applications.
Although some users may claim that earlier models performed better after the company suddenly stopped supporting them, it’s important to consider the extensive resources required to create these new models. This includes significant computational power, primarily from advanced GPU units designed for AI, cooling water usage, and large amounts of data for training purposes.
Based on research conducted at the University of Rhode Island’s AI lab and as reported by The Guardian and Tom’s Hardware, GPT-5 uses more electricity than its predecessor, GPT-4.
In my role as an analyst, I found it intriguing that when OpenAI introduced GPT-5, they failed to reveal the exact energy consumption of the model during operation, which includes generating responses to prompts. This lack of disclosure necessitates reliance on assumptions and approximations in the data provided by the learning institution.
According to reports, GPT-5 uses as much as 18 watts of power per query. OpenAI states that ChatGPT receives approximately 2.5 billion requests every day, which means the energy consumption of this model is a staggering 45 gigawatt-hours. This amount of electricity could satisfy the daily needs of about 1.5 million households in the United States.
According to their study, the researchers presumed that GPT-5 requires approximately 40 watt-hours of electricity for producing a response of around 1,000 tokens, which is considered medium-length.
Modern nuclear power plants can produce between 1 and 1.6 gigawatts of electricity per hour per reactor. On the other hand, OpenAI’s data centers running GPT-5 consume approximately 18 watt-hours per query, which is roughly equivalent to the power output of two to three nuclear reactors. This much power could potentially light up a small nation.
As per Professor Rakesh Kumar from the University of Illinois, his research focus is on the energy efficiency of computational and artificial intelligence models.
A more advanced model such as GPT-5 requires significantly more energy for both training and usage, and is designed for prolonged processing tasks. Without a doubt, its power consumption will exceed that of GPT-4.
It’s important to mention that OpenAI opted to reintroduce the earlier models of GPT-5, such as GPT-40, following criticism from users. However, these older models can only be accessed with a $20/month ChatGPT Plus subscription.
During the introduction of ChatGPT-5, OpenAI promoted it as the most intelligent AI model to date, likening it to a team of PhD-level specialists. However, some users have voiced their discontent with the update, asserting that the new version has negatively impacted and diminished ChatGPT’s user experience. They’ve pointed out issues such as glitches, bugs, and even periods of unresponsiveness.
Sam Altman, the CEO of OpenAI, has emphasized upcoming advancements aimed at enhancing the model’s user interaction. These improvements include raising the weekly message limit to 3,000 messages when using GPT-5 Thinking.
The baseline: AI is a power-hungry tool
Over the last several months, numerous reports have emerged, providing estimates on the power usage required to operate these advanced AI models. For example, a recent report claimed that ChatGPT uses less power than initially believed, referring to earlier reports’ calculations as rough approximations or “back-of-the-envelope math.” The study conducted by Epoch AI also revealed that generating a response for a query using ChatGPT-4o consumes just 0.3 watt-hours of energy.
This news follows another report indicating that the energy consumption of Microsoft and Google exceeds the electrical needs of more than 100 countries, as both tech companies have intensified their focus on artificial intelligence.
Beyond issues related to privacy and security, the amount of water used by AI systems for cooling is a point of much debate. A shocking report published in 2023 disclosed that AI models like Microsoft Copilot and ChatGPT consume roughly one bottle’s worth of water for cooling each time they generate a response to a question.
This year, Sam Altman, the CEO of OpenAI, playfully mentioned that the company invests a significant amount of money, approximating to tens of millions, on generating polite responses such as “please” and “thank you”. Later, he disclosed that ChatGPT consumes 0.34 watt-hours for each response it generates. This is equivalent to the energy consumption of an oven in slightly over a second or a high-efficiency lightbulb in a couple of minutes.
A separate analysis has uncovered that the requirements of AI are currently at their peak. This report indicates that the OpenAI’s GPT-3 model uses four times as much water as was initially estimated, while creating just 100 words with the GPT-4 model may necessitate up to three full water bottles.
Sam Altman provided an alternative estimate stating that ChatGPT consumes approximately 0.000085 gallons of water for each response it generates in response to a user query. Although his data might be accurate based on relevant measurements, it’s unclear which specific model he was referring to in this context.
Read More
- ENA PREDICTION. ENA cryptocurrency
- Minecraft lets you get the Lava Chicken song in-game — but it’s absurdly rare
- Gold Rate Forecast
- Lewis Capaldi Details “Mental Episode” That Led to Him “Convulsing”
- Wrestler Marcus “Buff” Bagwell Undergoes Leg Amputation
- PS5’s ChinaJoy Booth Needs to Be Seen to Be Believed
- Yungblud Vows to Perform Ozzy Osbourne Song “Every Night”
- Elden Ring Nightreign’s Patch 1.02 update next week is adding a feature we’ve all been waiting for since launch — and another I’ve been begging for, too
- Cyberpunk 2077’s Patch 2.3 is Here and It’s Another Excellent Overhaul
- Microsoft is on track to become the second $4 trillion company by market cap, following NVIDIA — and mass layoffs
2025-08-15 14:10