Running AI at Home Costs 12 Cents an Hour — Cheaper Than ChatGPT?

A Reddit user did something most of us never bother to: they measured what it actually costs to run a free AI model at home. Using a powerful gaming PC — an AMD Radeon 7900 XTX graphics card and a Ryzen 9800X3D processor — they calculated their electricity use running Qwen, a free open-source AI anyone can download. Where they live, power costs about 25 cents per kilowatt-hour (the unit on your electric bill). The result: roughly 12 cents for an hour of chatting with their own AI.

So is that cheaper than just paying for ChatGPT or a similar paid service? The honest answer is: it depends, and the 12 cents is only part of the story. The calculation leaves out the hardware. That graphics card alone costs around $900, and if you use it for a couple of hours a day over three years, the machine adds another 15 to 20 cents per hour of use. You also spend weekends setting things up, updating software, and troubleshooting when it breaks.

Paid services charge differently — usually per "token," which is roughly a word fragment — or a flat $20 a month. Those prices include the powerful computers, the electricity, and the engineers. You also get the smartest available models, which local machines still struggle to match on hard questions like complex coding or deep research. Nobody has to install anything.

The real lesson is that "free" AI has a meter running. Electricity prices swing wildly depending on where you live, from about 11 to 35 cents per kilowatt-hour, so the same setup could cost three times more in one city than another. For casual users, a $20 subscription is almost certainly simpler and often cheaper. For daily power users who care about privacy — nothing leaves their house — running AI locally can genuinely pay off.