{"id":29202,"date":"2025-02-24T00:53:10","date_gmt":"2025-02-24T08:53:10","guid":{"rendered":"https:\/\/essential.construction\/news\/deepseek-called-a-net-positive-for-data-centers-despite-overcapacity-worries\/"},"modified":"2025-02-24T00:53:10","modified_gmt":"2025-02-24T08:53:10","slug":"deepseek-called-a-net-positive-for-data-centers-despite-overcapacity-worries","status":"publish","type":"post","link":"https:\/\/essential.construction\/news\/deepseek-called-a-net-positive-for-data-centers-despite-overcapacity-worries\/","title":{"rendered":"DeepSeek called a net positive for data centers despite overcapacity worries"},"content":{"rendered":"<p> [ad_1]<br \/>\n<\/p>\n<div>\n<div class=\"text-to-speech\">\n    <button class=\"text-to-speech__button button\"><\/p>\n<p>            <img decoding=\"async\" class=\"text-to-speech__button__icon\" src=\"https:\/\/www.constructiondive.com\/static\/images\/audio_icon.svg?482016190122\" alt=\"-\"><\/p>\n<p>        Listen to the article<br \/>\n        <span class=\"text-to-speech__button__audio-length\">10 min<\/span><br \/>\n    <\/button><\/p>\n<div class=\"text-to-speech__controls\">\n        <audio controls=\"\" class=\"js-text-to-speech\" preload=\"none\"><source src=\"http:\/\/res.cloudinary.com\/dmgi9movl\/video\/upload\/q_1\/v1740149887\/news\/text_to_speech\/deepseek-net-positive-data-centers-overcapacity_edbsiu.wav\" type=\"audio\/mp3\"><\/source><\/audio><\/p>\n<div class=\"text-to-speech__controls__text\">\n            This audio is auto-generated. Please let us know if you have <a rel=\"nofollow noopener\" href=\"https:\/\/www.constructiondive.com\/contact\/\" target=\"_blank\">feedback<\/a>.\n        <\/div>\n<\/p><\/div>\n<\/div>\n<p><span><span><span><span><span><span>The global race to develop sophisticated, versatile AI models is driving a surge in development of the physical infrastructure to support them. Microsoft, Amazon, Meta and Google parent Alphabet \u2014 four of the AI industry\u2019s biggest players \u2014 <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/finance.yahoo.com\/news\/alphabet-meta-microsoft-set-to-spend-230-billion-this-year-as-hefty-ai-bills-come-under-scrutiny-184912247.html\" target=\"_blank\"><span><span><span><span><span><span><span><span>plan to invest<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> a combined $325 billion in capital expenditures in 2025, much of it for data centers and associated hardware.\u00a0<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>The technology giants\u2019 spending plans mark a 46% increase from their combined 2024 capital investments, according to Yahoo! Finance, and track a <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.woodmac.com\/horizons\/gridlock-demand-dilemma-facing-us-power-industry\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>doubling in proposed data centers\u2019 average size<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> from 150 gigawatts to 300 gigawatts between January 2023 and October 2024. In December, <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.datacenterdynamics.com\/en\/news\/meta-announces-4-million-sq-ft-louisiana-data-center-campus\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>Meta said it would build<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> a $10 billion, 4 million-square-foot data center in Louisiana that could consume the power output equivalent of two large nuclear reactors.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>The planned spending announcements came after the January release of the highly efficient, open-source AI platform DeepSeek, which led some experts and investors to question just how much infrastructure \u2014 from <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/youtubetranscriptoptimizer.com\/blog\/05_the_short_case_for_nvda\" target=\"_blank\"><span><span><span><span><span><span><span><span>high-performance chips<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> to <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.utilitydive.com\/news\/ipp-stocks-reel-as-low-cost-deepseek-chills-ai-load-growth-thesis\/738482\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>new power plants<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> and electrical equipment \u2014 the AI boom actually requires. But experts interviewed by Facilities Dive see AI efficiency gains as a net positive for data center developers, operators and customers in the long run, despite shorter-term uncertainty that could temporarily create overcapacity in some markets.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<h3 class=\"standard-heading\"><span><span><span><span><span><span>\u201cEvery technology has gotten more efficient\u201d<\/span><\/span><\/span><\/span><\/span><\/span><\/h3>\n<p><span><span><span><span><span><span>DeepSeek\u2019s late January commercial launch <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.cnn.com\/2025\/01\/27\/tech\/deepseek-ai-explainer\/index.html\" target=\"_blank\"><span><span><span><span><span><span><span><span>took many Americans by surprise<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span>, but the rise of a more efficient and capable AI model follows a well-worn pattern of technology development, said K.R. Sridhar, founder, chairman and CEO of power solutions provider Bloom Energy.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>\u201cIn 2010, it took about 1,500 watt-hours to facilitate the flow of one gigabyte of information,\u201d Sridhar said. \u201cLast year it took about one tenth of that \u2026 [but] the total growth of traffic, and therefore the chips and energy, has been twice that amount.\u201d\u00a0<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>As generative AI model development has grown more capital intensive, the benefits of efficiency have increased as well, Morningstar DBRS Senior Vice President Victor Leung wrote in a <a rel=\"nofollow noopener\" href=\"https:\/\/dbrs.morningstar.com\/document\/446577.pdf?Expires=1740069497&amp;Policy=eyJTdGF0ZW1lbnQiOlt7IlJlc291cmNlIjoiaHR0cHM6Ly9kYnJzLm1vcm5pbmdzdGFyLmNvbS9kb2N1bWVudC80NDY1NzcucGRmIiwiQ29uZGl0aW9uIjp7IkRhdGVMZXNzVGhhbiI6eyJBV1M6RXBvY2hUaW1lIjoxNzQwMDY5NDk3fX19XX0_&amp;Signature=lp~VZjM-7ZEhZJOVrufRwDY7dlxPVN5jLNjxWfBknrQtcGBwST1cvCKfXfyjIF-fTjKsTCQpPTPEwehwyrqi5YASZChK-T40FNjxjgAy8B9Tqdb2Wqii~tjLB1QuL1gJC8jqz2Fy0aWeExyCvIJGi5jK8MQLnSwwwW4IitUHHYOYcOQVc-ybxCLVIbVEH0wd6DGtMdTOO2X4L3RUp5tp1pC1g~S0mbz-SlNwXcQqvCZvpsjrMzMO~4aqVvc-dh3baO-Mi~6Roid9VoXK4omwuczM7GuL1Wcnkn3TUHOWD8Z53Blf6cogKmCZBvmc2dxHGyL1~wHIPPQzQM2WPvu0NA__&amp;Key-Pair-Id=KNWV36WLG7L4J\" target=\"_blank\"><span><span>Jan. 28 research note<\/span><\/span><\/a>.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>\u201cDeepSeek&#8217;s approach to constructing and training its model could be viewed as manifesting this drive towards more efficient models and to be expected as they and other companies seek to reduce AI deployment costs,\u201d Leung said. Continued efficiency improvements could moderate demand for data center capacity over the longer term, Leung added.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>As evidenced by the tech giants\u2019 recent announcements, the biggest AI players are unlikely to pull back on planned investments on DeepSeek\u2019s account, Ashish Nadkarni, group vice president and general manager for worldwide infrastructure research at IDC, told Facilities Dive.\u00a0<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>For companies like Meta and Microsoft, the drive to build as much AI capacity as quickly as possible is \u201cexistential,\u201d with Apple the only major tech company to buck the trend so far, Nadkarni said. Those companies, and others like Elon Musk\u2019s xAI, are using huge amounts of proprietary data to train differentiated foundational models useful in building emerging \u201cagentic AI\u201d tools and other future innovations, he added.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<h3 class=\"standard-heading\"><span><span><span><span><span><span>Less training, more inference and a cooling rethink?<\/span><\/span><\/span><\/span><\/span><\/span><\/h3>\n<p><span><span><span><span><span><span>Despite the scale of their planned and in-progress AI investments, the big tech companies \u201cremain at the starting line,\u201d with commercial payoff uncertain and likely still years away, Nadkarni said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>Imminent advances that allow more cost-effective AI deployments could \u201cstabilize the AI industry in the long run\u201d by reducing the risk of a capital spending pullback that could lead to \u201ca burst investment bubble in AI-related sectors,\u201d Leung wrote.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>A reduction in the computing power needed to train new models could eventually tip the balance of new data center deployments toward smaller \u201cinference\u201d data centers, which serve user requests rather than model development and refinement, experts say. That would mean proportionally fewer developments on the scale of Meta\u2019s two-gigawatt Louisiana facility or Compass Datacenters\u2019 <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.facilitiesdive.com\/news\/former-sears-hq-sustainable-data-center-campus\/719595\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>planned campus<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> in suburban Chicago that could eventually host five massive data centers.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<section class=\"storylines-carousel-wrapper hide-small show-large\" id=\"desktop-carousel\"\/>\n<p><span><span><span><span><span><span>\u201cAn acceleration of [large language models] in large training data centers will lead to a natural acceleration of inference data centers \u2026 located in proximity to people, processes and machines,\u201d Sridhar said. Inference data centers typically consume tens of megawatts of power, compared with hundreds of megawatts for centralized training facilities.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>For now, data center proposals continue to get bigger, with the average proposed data center doubling in size from <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.woodmac.com\/horizons\/gridlock-demand-dilemma-facing-us-power-industry\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>about 150 megawatts to 300 megawatts<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> between early 2023 and mid-2024, according to an October 2024 report from Wood Mackenzie.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>Even if AI efficiency gains continue, new data centers likely won\u2019t change their plans to use liquid cooling systems, which are more effective at cooling high-powered server racks than traditional air cooling systems, said Angela Taylor, chief of staff at liquid cooling provider LiquidStack. Compass Datacenters says its Chicago-area campus will use non-water liquid cooling, while Microsoft recently unveiled a new <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.microsoft.com\/en-us\/microsoft-cloud\/blog\/2024\/12\/09\/sustainable-by-design-next-generation-datacenters-consume-zero-water-for-cooling\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>closed-loop water cooling system<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> that it said could avoid the need for 125 million liters of water annually at new facilities.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>Taylor also expects liquid cooling to play a pivotal role in new inference data centers in \u201cedge\u201d environments closer to users.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>\u201c[The] edge has harsh operating and serviceability requirements, such as air pollution, hot and humid climate [and] smaller footprints \u2026 liquid cooling shines in these types of <\/span><\/span><\/span><\/span><\/span><\/span><span><span><span><span><span><span>environments,\u201d she said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>However, some operators of existing air-cooled data centers may rethink capital-intensive plans to retrofit them with liquid cooling systems, which \u201care still several years away\u201d from widespread deployment, Nadkarni said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>\u201c[Liquid cooling] is a much heavier lift for retrofits,\u201d he said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<section class=\"storylines-carousel-wrapper show-small hide-large\" id=\"mobile-carousel\"\/>\n<h3 class=\"standard-heading\"><span><span><span><span><span><span>A reset for the power business<\/span><\/span><\/span><\/span><\/span><\/span><\/h3>\n<p><span><span><span><span><span><span>Amid long construction time frames and <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.utilitydive.com\/news\/grid-interconnection-queue-berkeley-lab-lbnl\/712926\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>grid interconnection bottlenecks<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span>, data center developers like Meta and xAI are racing \u201cto secure as much natural gas as possible,\u201d said Gabriel Kra, managing director at venture capital firm Prelude Ventures.\u00a0<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>Entergy Louisiana plans to build <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.entergynewsroom.com\/news\/entergy-louisiana-power-meta-s-data-center-in-richland-parish\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>three combined-cycle gas plants<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> to power the new Meta campus, while xAI deployed more than a <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.datacenterdynamics.com\/en\/news\/elon-musks-xai-data-center-adding-to-memphis-air-quality-problems-campaign-group\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>dozen off-grid gas turbines<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span> last year while waiting on a grid connection for its Memphis campus.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>Over the longer term, the growth of inference data centers \u201cwill create a huge need for local availability of electricity in already congested metro areas,\u201d further increasing the value of onsite power, Sridhar said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>But many data centers will still connect to the grid, and more efficient AI models call into question the longer-term economics of <\/span><\/span><\/span><\/span><\/span><\/span><a rel=\"nofollow noopener\" href=\"https:\/\/www.utilitydive.com\/news\/constellation-three-mile-island-nuclear-power-plant-microsoft-data-center-ppa\/727652\/\" target=\"_blank\"><span><span><span><span><span><span><span><span>premium-priced power deals announced recently<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a><span><span><span><span><span><span>, Kra said.\u00a0<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>\u201cIf the load growth from AI is suddenly curtailed by a factor of five or 10, that has big implications,\u201d Kra said. \u201cWe\u2019ve heard recent pitches from companies that are relying on expensive contracts with hyperscalers \u2026.[This] calls those business models more seriously into question.\u201d<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>On the bright side, lower-than-expected data center energy demand could pay environmental dividends by reducing the need for older coal power plants, Kra added.\u00a0<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<h3 class=\"standard-heading\"><span><span><span><span><span><span>Tailwinds for colocation and on-premise compute<\/span><\/span><\/span><\/span><\/span><\/span><\/h3>\n<p><span><span><span><span><span><span>If AI models continue to see efficiency gains, then larger enterprises \u2014 such as global automakers \u2014 may look more closely at bringing more technology development and computing power in-house, rather than relying on third-party resources, Nadkarni said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>That would necessitate capital investment on those companies\u2019 part but could reduce demand for centralized AI computing, he said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>Colocation data centers, where customers rent space to house IT equipment rather than keeping it in their own facilities, are meanwhile \u201cgrowing extremely fast,\u201d Taylor said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>Colocation facilities tend to be located on highly-connected, adequately powered sites near population centers, making them attractive to users who need computing power in short order, said Scott McCrady, chief client executive at Cologix, a colocation firm. Those located in major network hubs also offer very low-latency computing, which is critical for inference applications from which users expect rapid outputs, he said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>But the tailwinds benefiting on-premise and colocation compute might not be enough to sustain the present data center boom, experts say.\u00a0<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>To start, hyperscalers may push for shorter lease terms on large-scale facilities as they weigh uncertainty around future computing power requirements, Leung said.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>And the wider industry could be in for a temporary period of overcapacity, similar to what happened in the aftermath of the dot-com boom of the late 1990s and early 2000s, Nadkarni said. It took years for the market to absorb the glut of fiber networks and data centers built back then, he added.<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p><span><span><span><span><span><span>\u201cI have a nagging worry we\u2019re going to see the same thing this time,\u201d Nadkarni said. \u201cI don\u2019t think we can keep building compute at the same rate.\u201d<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<\/p><\/div>\n<p>[ad_2]<br \/>\n<br \/><a href=\"https:\/\/www.constructiondive.com\/news\/deepseek-net-positive-data-centers-overcapacity\/740617\/\" rel=\"nofollow noopener\" target=\"_blank\">This article was originally posted at Source link <\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>[ad_1] Listen to the article 10 min This audio is auto-generated. Please let us know if you have feedback. The &#8230; <a title=\"DeepSeek called a net positive for data centers despite overcapacity worries\" class=\"read-more\" href=\"https:\/\/essential.construction\/news\/deepseek-called-a-net-positive-for-data-centers-despite-overcapacity-worries\/\" aria-label=\"Read more about DeepSeek called a net positive for data centers despite overcapacity worries\">Read more<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[457],"tags":[],"class_list":["post-29202","post","type-post","status-publish","format-standard","hentry","category-construction-dive","generate-columns","tablet-grid-50","mobile-grid-100","grid-parent","grid-33"],"_links":{"self":[{"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/posts\/29202","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/comments?post=29202"}],"version-history":[{"count":0,"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/posts\/29202\/revisions"}],"wp:attachment":[{"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/media?parent=29202"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/categories?post=29202"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/essential.construction\/news\/wp-json\/wp\/v2\/tags?post=29202"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}