{"id":3473,"date":"2025-09-04T19:45:54","date_gmt":"2025-09-04T17:45:54","guid":{"rendered":"https:\/\/gregarednak.si\/?p=3473"},"modified":"2025-09-04T19:45:56","modified_gmt":"2025-09-04T17:45:56","slug":"ai-proti-googlu-koliko-energije-porabi-en-ai-odgovor","status":"publish","type":"post","link":"https:\/\/gregarednak.si\/en\/ai-proti-googlu-koliko-energije-porabi-en-ai-odgovor\/","title":{"rendered":"AI vs. Google: How much energy does one AI response consume?"},"content":{"rendered":"<div class=\"wp-block-aioseo-table-of-contents\"><ul><li><a class=\"aioseo-toc-item\" href=\"#aioseo-1-zakaj-je-ta-primerjava-sploh-pomembna\">Why is this comparison even important?<\/a><\/li><li><a class=\"aioseo-toc-item\" href=\"#aioseo-3-kako-nastane-poraba-pri-klasicnem-iskanju\">How spending is generated in classic search<\/a><\/li><li><a class=\"aioseo-toc-item\" href=\"#aioseo-4-zakaj-je-generativni-ai-lahko-drazji\">Why generative AI can be more expensive<\/a><\/li><li><a class=\"aioseo-toc-item\" href=\"#aioseo-5-postena-primerjava-po-scenarijih\">Fair comparison across scenarios<\/a><\/li><li><a class=\"aioseo-toc-item\" href=\"#aioseo-6-kaj-najbolj-vpliva-na-porabo-v-praksi\">What has the greatest impact on consumption in practice?<\/a><\/li><li><a class=\"aioseo-toc-item\" href=\"#aioseo-9-kako-zmanjsati-svoj-odtis-checklist\">How to reduce your footprint<\/a><\/li><li><a class=\"aioseo-toc-item\" href=\"#aioseo-10-pogosta-vprasanja-in-miti\">Frequently asked questions and myths<\/a><\/li><li><a class=\"aioseo-toc-item\" href=\"#aioseo-11-zakljucek\">Conclusion<\/a><\/li><\/ul><\/div>\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-1-zakaj-je-ta-primerjava-sploh-pomembna\">Why is this comparison even important?<\/h2>\n\n\n\n<p>Generative AI is derived from search <em>\u201cfind pages\u201d<\/em> moves to <em>\u201cwrite me an answer\u201d<\/em>For readers, this means faster insight, for business, better processes, and for the environment, a new question: how much is <strong>actual energy consumption<\/strong> to such an answer compared to classic Google search? Understanding this relationship helps in making mature decisions: when AI really brings value and when it is better to stick with search or simpler models.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-3-kako-nastane-poraba-pri-klasicnem-iskanju\">How spending is generated in classic search<\/h2>\n\n\n\n<p>When searching, the browser sends a short request to the data centers, where a match is initiated with the already prepared <strong>indices<\/strong> and <strong>cache<\/strong>. Much of the work is done in advance (indexing the web), so it is <strong>inferential<\/strong> part of an individual query is relatively short and energy-efficient. Typically, we are talking about a fraction of a watt-hour (about ~0.3 Wh per query), with small variance \u2014 most queries are similarly \u201clightweight.\u201d<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-4-zakaj-je-generativni-ai-lahko-drazji\">Why generative AI can be more expensive<\/h2>\n\n\n\n<p>LLM does not return links, but <strong>generates text<\/strong>This requires:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Multiple calculations per token<\/strong> (each new word is the result of matrices and attention to context).<\/li>\n\n\n\n<li><strong>More context<\/strong> (longer prompts \u2192 more tokens \u2192 more cycles).<\/li>\n\n\n\n<li><strong>Multimodality<\/strong> (image, sound, video) and above all <strong>\u201creasoning\u201d<\/strong> tasks where the model makes inferences in multiple steps.<\/li>\n\n\n\n<li><strong>Less caching<\/strong> results (answers are unique), so there are fewer duplicate hits than with a search.<\/li>\n<\/ul>\n\n\n\n<p>Together, this means that spending can rise quickly \u2014 sometimes it stays comparable to search, sometimes it\u2019s <strong>an order of magnitude higher<\/strong>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-5-postena-primerjava-po-scenarijih\">Fair comparison across scenarios<\/h2>\n\n\n\n<p>Below are <strong>framework ranges<\/strong> for energy consumption per response\/query. The figures are for guidance (they do not necessarily include all infrastructure elements), but show the differences between tasks.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Scenario<\/th><th>Typical consumption (Wh per response)<\/th><th>Note<\/th><\/tr><\/thead><tbody><tr><td>Classic search (Google)<\/td><td>~0,3<\/td><td>Short, standardized process via indexes\/caches.<\/td><\/tr><tr><td>AI: short text prompt<\/td><td>~0.24\u20130.3<\/td><td>Today&#039;s optimized systems can be compared to search. <a href=\"https:\/\/cloud.google.com\/blog\/products\/infrastructure\/measuring-the-environmental-impact-of-ai-inference\/?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noreferrer noopener\">Google Cloud<\/a><\/td><\/tr><tr><td>AI: \u201creasoning\u201d\/longer answer<\/td><td>~5 and up<\/td><td>The number of tokens and multi-step reasoning raise the consumption. <a href=\"https:\/\/iea.blob.core.windows.net\/assets\/34eac603-ecf1-464f-b813-2ecceb8f81c2\/EnergyandAI.pdf?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noreferrer noopener\">iea.blob.core.windows.net<\/a><\/td><\/tr><tr><td>AI: image \u2192 generation\/analysis<\/td><td>~1\u20132<\/td><td>Images are significantly more expensive than text. <a href=\"https:\/\/iea.blob.core.windows.net\/assets\/34eac603-ecf1-464f-b813-2ecceb8f81c2\/EnergyandAI.pdf?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noreferrer noopener\">iea.blob.core.windows.net<\/a><\/td><\/tr><tr><td>AI: short video \u2192 generation<\/td><td>~100+<\/td><td>Example: ~115 Wh for ~6 seconds of video. <a href=\"https:\/\/iea.blob.core.windows.net\/assets\/34eac603-ecf1-464f-b813-2ecceb8f81c2\/EnergyandAI.pdf?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noreferrer noopener\">iea.blob.core.windows.net<\/a><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p><strong>What to remember:<\/strong> at <strong>short text messages<\/strong> tasks AI can <strong>in search rank<\/strong>, at <strong>complex<\/strong> and <strong>multimodal<\/strong> but easily jumps on <strong>10\u00d7 or more<\/strong>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-6-kaj-najbolj-vpliva-na-porabo-v-praksi\">What has the greatest impact on consumption in practice?<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Length of the challenge and response:<\/strong> more tokens = more calculations.<\/li>\n\n\n\n<li><strong>Model selection:<\/strong> Smaller\/optimized models are significantly more economical than huge \u201creasoning\u201d models.<\/li>\n\n\n\n<li><strong>How to use:<\/strong> batching multiple questions, context reuse (cache), avoiding unnecessary image\/video generation.<\/li>\n\n\n\n<li><strong>Infrastructure:<\/strong> Newer accelerators, better orchestration, and low-carbon flow (where\/when available) reduce the footprint.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-9-kako-zmanjsati-svoj-odtis-checklist\">How to reduce your footprint<\/h2>\n\n\n\n<p><strong>For users<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>To start, ask <strong>short and to the point<\/strong>; asks for <strong>concise<\/strong> answer.<\/li>\n\n\n\n<li>Generate images\/videos <strong>only when they bring added value<\/strong>.<\/li>\n<\/ul>\n\n\n\n<p><strong>For developers\/products<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Default<\/strong> on <strong>smaller models<\/strong>; larger and \u201creasoning\u201d include conditionally (feature flags).<\/li>\n\n\n\n<li>Use <strong>caching<\/strong> partial results and <strong>batching<\/strong> requests.<\/li>\n\n\n\n<li>Introduce <strong>length restrictions<\/strong> (max tokens), trim context, use <strong>retrieval<\/strong> instead of blindly loading large instructions.<\/li>\n\n\n\n<li>On-the-go <strong>consumption measurement<\/strong> (telemetry) and parameter adjustment (temperature, max tokens).<\/li>\n<\/ul>\n\n\n\n<p><strong>For businesses\/IT<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Route tasks to time\/areas with <strong>more low-carbon energy<\/strong>.<\/li>\n\n\n\n<li>Think about <strong>quantization\/distillation<\/strong> internal models and rules for economical use (e.g. \u201cno-image-by-default\u201d).<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-10-pogosta-vprasanja-in-miti\">Frequently asked questions and myths<\/h2>\n\n\n\n<p><strong>Is AI always 10x more greedy than Google?<\/strong><br>No. With short text prompts, it can be done today <strong>comparable to search<\/strong>; the difference explodes when it comes to complexity (reasoning, images, video).<\/p>\n\n\n\n<p><strong>Do multiple smaller calls consume less than one debt?<\/strong><br>Not necessarily. If each call involves a large model and a long context, it can be <strong>one well-designed longer call<\/strong> more efficient.<\/p>\n\n\n\n<p><strong>Do \u201cgreen\u201d plugs solve everything?<\/strong><br>They help, but <strong>efficiency at source<\/strong> (model, call, architecture) remains key. Spending less is always better than \u201cjustifying\u201d later.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"aioseo-11-zakljucek\">Conclusion<\/h2>\n\n\n\n<p>AI is not inherently an \u201cenergy blunder\u201d nor \u201cfree magic\u201d. <strong>Task context<\/strong> is the one that stretches the spectrum from <strong>comparable spending with search<\/strong> (short text) to <strong>an order of magnitude higher<\/strong> (reasoning, images, video). If you want good results with a small footprint, optimize <strong>model, call and path<\/strong> \u2014 and consider whether a generative answer is really the best choice for a given task.<\/p>\n\n\n\n<p><strong>Sources (for key figures):<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Google Cloud: <em>Measuring the environmental impact of AI inference<\/em> (median ~0.24 Wh per text prompt). <a href=\"https:\/\/cloud.google.com\/blog\/products\/infrastructure\/measuring-the-environmental-impact-of-ai-inference\/?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noreferrer noopener\">Google Cloud<\/a><\/li>\n\n\n\n<li>IEA: <em>Energy and AI<\/em> (comparative ranges for text\/image\/video, e.g. ~115 Wh for a short video). <a href=\"https:\/\/iea.blob.core.windows.net\/assets\/34eac603-ecf1-464f-b813-2ecceb8f81c2\/EnergyandAI.pdf?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noreferrer noopener\">iea.blob.core.windows.net<\/a><\/li>\n<\/ul>","protected":false},"excerpt":{"rendered":"<p>Zakaj je ta primerjava sploh pomembna Generativni AI se iz iskalnega \u201cnajdi strani\u201d premika v \u201cnapi\u0161i mi odgovor\u201d. Za bralce to pomeni hitrej\u0161i vpogled, za biznis bolj\u0161e procese, za okolje pa novo vpra\u0161anje: kolik\u0161na je dejanska poraba energije na tak odgovor v primerjavi s klasi\u010dnim Googlovim iskanjem? Razumevanje tega razmerja pomaga pri zrelih odlo\u010ditvah: kdaj [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":3475,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[273,272],"tags":[],"class_list":["post-3473","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-trajnost"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/posts\/3473","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/comments?post=3473"}],"version-history":[{"count":1,"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/posts\/3473\/revisions"}],"predecessor-version":[{"id":3474,"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/posts\/3473\/revisions\/3474"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/media\/3475"}],"wp:attachment":[{"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/media?parent=3473"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/categories?post=3473"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/gregarednak.si\/en\/wp-json\/wp\/v2\/tags?post=3473"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}