{"id":20942,"date":"2026-10-09T12:10:45","date_gmt":"2026-10-09T12:10:45","guid":{"rendered":"https:\/\/news.theck1.no\/?p=20942"},"modified":"2026-10-09T12:10:45","modified_gmt":"2026-10-09T12:10:45","slug":"software-could-be-the-easiest-fix-for-hyperscalers-ai-power-squeeze-researchers-say","status":"publish","type":"post","link":"https:\/\/news.theck1.no\/?p=20942","title":{"rendered":"Software could be the easiest fix for hyperscalers’ AI power squeeze, researchers say"},"content":{"rendered":"

AI-rewritten:<\/strong> This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.<\/p>\n

Tom’s Hardware • Chris Stokel-Walker • October 8, 2026<\/strong><\/div>\n\n

Data center energy demand is rising rapidly, with projections indicating that by 2030, AI will consume approximately 945 terawatt-hours of electricity, a volume comparable to Japan’s total usage. While the industry has traditionally focused on improving hardware efficiency and securing power supplies, researchers suggest that optimizing software could offer a more effective solution. According to a 2025 survey by the Uptime Institute, average power usage effectiveness (PUE) has remained stagnant for six years. PUE measures how efficiently cooling and power systems operate but does not account for whether the software running on servers is performing useful work. Since servers account for about 60% of electricity demand in modern data centers, experts argue that improving efficiency through software changes is a sensible approach.<\/p>\n

<\/p>\n

Jae-Won Chung, a PhD candidate at the University of Michigan and researcher with ML.Energy, states that software can meaningfully help reduce power consumption. His team’s tests show that running inference in FP8, a lower-precision format, used one-third less energy than bfloat16 versions on specific tasks. Additionally, his Perseus training optimizer identified parts of large-model training jobs requiring less work and slowed them down to match busier components, cutting training energy by up to 30% without reducing throughput or changing hardware. Similar efforts are underway by major GPU firms like Nvidia, which fine-tunes power profiles to save energy while retaining performance.<\/p>\n

Beyond simple optimization, researchers emphasize shifting when and where electricity is drawn. Sophie Hall, a doctoral student at ETH Zurich, argues that the timing and location of usage interact critically with the electrical grid. She suggests delaying non-urgent batch jobs until local demand falls or routing them to regions with spare capacity and lower-carbon electricity. However, challenges remain, including data sovereignty rules preventing cross-border workload movement and concerns about moving large datasets. Ultimately, while software cannot replace better chips or new infrastructure, it acts as a control layer to maximize the utility of existing assets.<\/p>\n

\n

Source:<\/strong> Tom’s Hardware • Chris Stokel-Walker • October 8, 2026<\/p>\n

Read the original article at Tom’s Hardware →<\/a><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

AI-rewritten: This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story. Tom’s Hardware • Chris Stokel-Walker • October 8, 2026 Data center energy demand is rising rapidly, with projections<\/p>\n

READ MORE<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7],"tags":[],"class_list":["post-20942","post","type-post","status-publish","format-standard","hentry","category-it-hardware"],"_links":{"self":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/posts\/20942","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=20942"}],"version-history":[{"count":0,"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/posts\/20942\/revisions"}],"wp:attachment":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=20942"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=20942"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=20942"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}