{"CACHEDAT":"2026-06-05 16:46:01","SLUG":"solution-vjmdqmmd1B","MARKDOWN":"![](/api/attachments.redirect?id=5fa506c7-7192-472d-a9cf-2391f4a38047 \" =1000x700\")\n\n* **AI Training (4.5%):** This is the energy used to *build* the models (like GPT-4 or Claude). While training a single model is incredibly energy-intensive (sometimes using as much power as 1,000 homes for a month), it is a one-time cost.\n\n \\\n* **Non-AI Data Center Tasks (70%):** This represents the \"traditional\" internet—Netflix streaming, email, cloud storage, and social media. While still the majority, this slice is shrinking as a percentage because AI demand is growing so much faster.\n\nSource: [www.brookings.edu](https://www.brookings.edu)\n\n\n\\","HTML":"
\n
\n
AI Training (4.5%): This is the energy used to build the models (like GPT-4 or Claude). While training a single model is incredibly energy-intensive (sometimes using as much power as 1,000 homes for a month), it is a one-time cost.\n
\n
\n
\n
Non-AI Data Center Tasks (70%): This represents the "traditional" internet—Netflix streaming, email, cloud storage, and social media. While still the majority, this slice is shrinking as a percentage because AI demand is growing so much faster.\n