While OpenAI celebrates the advanced capabilities of its newest AI model, GPT-5 is secretly guzzling electricity at an alarming rate. Independent university research shows it devours up to 40 watt-hours per medium-length response—about 1,000 tokens. That's roughly nine times more power-hungry than its predecessor. Talk about an energy hog.

The average GPT-5 query consumes around 18 watt-hours, dwarfing GPT-4's modest 2.1 watt-hours for similar tasks. And sometimes it gets worse. Complex queries can spike to that 40 watt-hour ceiling. OpenAI, predictably, hasn't bothered to share official energy figures. Typical.

This consumption leap isn't just a minor technical footnote. With an estimated 2.5 billion daily queries, GPT-5 could collectively gulp down 45 gigawatt-hours daily. That's equivalent to powering 1.5 million American homes. Or running multiple nuclear reactors. Just to answer questions about pasta recipes and help debug code. The EU AI Act requires strict monitoring of energy consumption for high-risk AI systems.

GPT-5's daily energy binge could power entire cities, all so you can perfect your carbonara recipe.

The culprit? GPT-5's massive size and complexity. All those fancy new capabilities—solving tough problems, generating longer outputs—come with a serious electrical price tag. More parameters mean more computation. More computation means more juice.

Let's put this in perspective. Traditional AI queries and searches typically use less than 1 watt-hour. Early ChatGPT responses? A fraction of that. Now we're talking about 40. Progress, huh?

This energy binge raises obvious environmental concerns. AI systems already strain power grids, and GPT-5 dramatically accelerates this trend. The model may employ a mixture-of-experts architecture that selectively activates only parts of the system, but it's clearly not enough to curb its voracious appetite. The proliferation of these power-hungry AI technologies is expected to cause serious grid stability issues in the near future. Data centers housing these models need constant cooling and redundant power systems. All while consuming electricity that could power entire small nations.

The irony? As these models get "smarter," they're becoming environmental liabilities. Nine times the power for what—slightly better haikus? Sure, GPT-5 can solve complex problems, but at what cost? Maybe the next breakthrough shouldn't be more capabilities, but finding a way to deliver them without threatening the power grid.