Topics
late
AI
Amazon
Image Credits:Nikada / Getty Images
Apps
Biotech & Health
clime
Image Credits:Nikada / Getty Images
Cloud Computing
Commerce
Crypto
Image Credits:Epoch AI
Enterprise
EVs
Fintech
Fundraising
contraption
Gaming
Government & Policy
Hardware
layoff
Media & Entertainment
Meta
Microsoft
privateness
Robotics
Security
Social
Space
Startups
TikTok
Transportation
Venture
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
datum shopping center to train and incline AI may before long contain billion of chips , be hundreds of billions of dollar bill , and involve power equivalent to a heavy city ’s electricity gridiron , if the current tendency hold .
That ’s according toa new studyfrom researchers at Georgetown , Epoch AI , and Rand , which looked at the outgrowth trajectory of AI information centers around the world from 2019 to this year . The co - authors compile and canvas a dataset of over 500 AI data center project and found that , while the computational performance of datum centers is more than doubling per annum , so are the power requirements and capital expenditures .
The findings illustrate the challenge in establish the necessary infrastructure to patronize the evolution of AI technology in the issue forth tenner .
OpenAI , which late state thatroughly 10 % of the macrocosm ’s populationis using its ChatGPT platform , has a partnershipwith SoftBank and others to advance up to $ 500 billion to establish a net of AI data centers in the U.S. ( and possibly elsewhere ) . Other tech giants , includingMicrosoft , Google , andAWS , have collectively pledged to pass hundreds of millions of dollar this twelvemonth alone inflate their data center footprints .
According to the Georgetown , Epoch , and Rand study , the hardware costs for AI data heart like xAI ’s Colossus , which has a price tag of around $ 7 billion , increased 1.9x each twelvemonth between 2019 and 2025 , while king needs climb 2x annually over the same period of time . ( Colossus draws an reckon 300 megawatts of power , as much as 250,000 households . )
The study also found that data centers have become much more energy effective in the last five year , with one key metric — computational performance per watt — increasing 1.34x each class from 2019 to 2025 . Yet these melioration wo n’t be enough to make up for growing power needs . By June 2030 , the lead AI information center may have 2 million AI chips , cost $ 200 billion , and ask 9 GW of tycoon — more or less the output of nine nuclear reactors .
It ’s not a fresh Book of Revelation that AI data center electricity demands areon pace to greatly try the power power system . datum center of attention energy intake is forecast to grow 20 % by 2030,according to a recent Wells Fargo analysis . That could fight renewable origin of power , which are dependent on variable weather , to their limits — spur a wild leek - upin non - renewable , environmentally damaging electricity sources like dodo fuels .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
AI data eye also lay otherenvironmental threat , such as eminent water white plague , and take up worthful real acres , as well as erode state revenue enhancement bases . A study by Good Jobs First , a Washington , D.C.-based nonprofit organization , estimatesthat at least 10 states misplace over $ 100 million per year in taxation gross to data centers , the result of too generous incentives .
It ’s potential that these projections may not fare to glide by , of course , or that the time musical scale are off - kelter . Some hyperscalers , likeAWS and Microsoft , have pulled back on data point center undertaking in the last several weeks . In a note to investor in mid - April , analysts at Cowenobserved that there ’s been a “ cooling ” in the datum center marketplace in early 2025 , signal the industry ’s fear of unsustainable elaboration .
