ChatGPT is free for you. But for OpenAI, every conversation costs money. A lot of money. We’re talking millions of dollars a day just to keep the servers running.<\/p>\n
How can a computer program cost that much? The answer comes down to two words: computing power<\/strong>.<\/p>\n You know graphics cards \u2014 GPUs. They’re what run video games on your PC. Except AI needs specialized graphics cards that cost between $25,000 and $40,000 each<\/strong>. And you need thousands of them.<\/p>\n Training GPT-4 cost around $100 million. Just the training. Before anyone asked it a single question.<\/p>\n Imagine building a $100 million car. And then every kilometer you drive costs fuel at $50 per liter. That’s the economics of AI.<\/p>\n There are two types of costs. Training<\/strong> is the cost of creating the model. You do it once (it takes months and millions). Inference<\/strong> is the cost of running the model when someone asks it a question. That one’s ongoing.<\/p>\n Every time you type a message to ChatGPT, servers somewhere in the world run billions of calculations to generate your response. That takes electricity. A lot of electricity. It’s estimated that a ChatGPT query uses about 10 times more energy than a Google search.<\/p>\n AI data centers consume as much electricity as a small city. Microsoft, Google, and Amazon are building power plants just to feed their AI servers. Some are buying nuclear energy.<\/p>\n In Quebec, we have a massive advantage: hydroelectricity. Clean, abundant, and cheap compared to the rest of the world. That’s one reason why more and more AI companies are looking at Quebec to set up their data centers.<\/p>\n Yes and no. Cost per query drops every year thanks to technical improvements. Models become more efficient. Hardware improves. In two years, the cost per million tokens (processed words) went from $30 to under $1.<\/p>\n But at the same time, usage is exploding. More people use AI, more often, for more complex tasks. So total costs keep climbing even as unit costs drop.<\/p>\n It’s like mobile data: the price per gigabyte dropped, but your bill went up because you consume 100 times more than before.<\/p>\n ChatGPT’s free model probably won’t stay free forever. Companies will find ways to monetize \u2014 subscriptions, ads, data. That’s why open source<\/strong> matters so much: models you can run on your own hardware, without depending on a company that can change its pricing tomorrow.<\/p>\n At Laeka<\/a>, we work with efficient open source models that run without requiring a data center. And Sherpa<\/a> is free and will stay that way \u2014 because it’s a nonprofit, not a startup chasing investors.<\/p>\n","protected":false},"excerpt":{"rendered":" ChatGPT is free for you. But for OpenAI, every conversation costs money. A lot of money. We’re talking millions of…<\/p>\n","protected":false},"author":1,"featured_media":33,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_kad_post_transparent":"","_kad_post_title":"","_kad_post_layout":"","_kad_post_sidebar_id":"","_kad_post_content_style":"","_kad_post_vertical_padding":"","_kad_post_feature":"","_kad_post_feature_position":"","_kad_post_header":false,"_kad_post_footer":false,"_kad_post_classname":"","footnotes":""},"categories":[190],"tags":[],"class_list":["post-807","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-understanding-ai"],"_links":{"self":[{"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/posts\/807","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/comments?post=807"}],"version-history":[{"count":0,"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/posts\/807\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/media\/33"}],"wp:attachment":[{"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/media?parent=807"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/categories?post=807"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/laeka.org\/blog\/wp-json\/wp\/v2\/tags?post=807"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}Thousands of Graphics Cards Running Hot<\/h2>\n
Training vs Usage<\/h2>\n
Why Electricity Is the Real Battlefield<\/h2>\n
Will Costs Go Down?<\/h2>\n
What This Means for You<\/h2>\n