Top AI labs, including OpenAI, Anthropic, and Google, are digging deep into their pockets to make significant progress in the AI landscape. In 2023, a report revealed that OpenAI spends up to $700,000 per day to keep ChatGPT running.
More recently, OpenAI CEO Sam Altman jokingly shared some insight about how much electricity ChatGPT consumes when people give a courteous “please and thank you” while interacting with the company’s latest AI models.
“I wonder how much money OpenAI has lost in electricity costs from people saying please and thank you to their models,” tomieinlove on X indicated.
While responding to the question, CEO Sam Altman jokingly indicated that ChatGPT spends tens of millions of dollars on the extra effort needed.
However, the large sum of money used to keep ChatGPT running and responding to such queries doesn’t seem to worry the executive. In fact, Altman indicated that it’s money well spent.
While a separate report revealed that Microsoft and Google’s power consumption surpasses the electricity usage of over 100 countries, a new report by Epoch AI dismissed the claims as “napkin math”, further suggesting that ChatGPT doesn’t consume as much power as previously thought.
Epoch AI claimed that the data was based on an assumption that OpenAI powers its latest AI models with outdated GPUs.The report detailed that ChatGPT running GPT-4o only consumes 0.3 watt-hours when generating a response for a query.
Money isn’t the only factor that drives AI development and progression, AI labs heavily rely on data centers for computing power. Similarly, data centers require cooling water to prevent their servers from overheating, which would lead to inevitable malfunctions and potential downtime.
In 2023, a separate report revealed that Microsoft Copilot and ChatGPT consume one water bottle for cooling when generating a response to a query. And as it seems, AI models continue to demand more cooling water as they advance.
For instance, OpenAI’s GPT-3 model consumes four times more water than previously thought, while GPT-4 consumes up to 3 water bottles to generate a mere 100 words.
