AI Now Does an Hour of Human Work for 4 Cents

AI systems can now handle some tasks that would take a person weeks or months. But being able to do something isn't the same as doing it cheaply. So a researcher built a database comparing two numbers for each task: how much computer power the AI burned, and how long a human would take to do the same thing. The finding was blunt. For a typical task, the machine cost roughly 4 cents per hour of human work. Even for the most expensive jobs tested — a tough math proof and a full video game playthrough — the cost landed near $15 an hour.

Put that next to real wages and the gap is huge. The median US worker earns about $25 an hour. Skilled white-collar workers, like research scientists or software engineers, effectively cost employers $100 to $300 an hour once benefits and overhead are counted. When the machine costs pennies against that, the economics get uncomfortable fast. It's a bit like discovering your car runs on pocket lint instead of petrol.

There is a real catch. The author admits the figures are educated guesses, because AI companies don't publish how much computing power their models use. And what you actually pay for AI today is roughly ten times the raw hardware cost, because companies have to cover training and research. Still, even at ten times the cost, most tasks stay cheaper than the human doing them.

Why it matters beyond money: if digital workers are this cheap, the limit on replacing jobs stops being skill and starts being electricity and chips. That could mean a flood of AI workers arriving faster than anyone planned for — affecting wages, hiring, and who controls the technology.