AI’s demand has reached a separating point where its birthing two visions for the computing infrastructure’s future, where one’s derived on centralized global build-out of power-hungry data centers led by Big Tech, and a new decentralized model to deliver powerful AI from local clusters of ordinary computers. Data center energy efficiency metrics are changing the very nature of what AI-based computing capacity will look like.
On Thursday, researchers from École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland, introduced new distributed software that allows users to run AI models locally, sidestepping cloud inference by coordinating existing machines on local networks through a system called Anyway Systems.
Anyway Systems, created by Gauthier Voron, Geovani Rizk, and Professor Rachid Guerraoui at EPFL’s Distributed Computing Laboratory, takes a different approach, bundling local machines into an on-premises cluster capable of running open-source AI models without sending data to the cloud.
Global data center infrastructure strains are experiencing unprecedented demand, with new industry core assumptions about scale, energy, and control.
According to Goldman Sachs Research, the tension pits the current industrial reality, predicting global power data center power demand to hit a whopping 165% by 2030. All that is only to support advanced AI models against rising alternatives. And nothing else.
“Goldman Sachs Research forecasts data center demand to grow by about 50% to 92 GW by 2027, with a compound annual growth rate of 17% between 2025 and 2028,” said the report.
Then the 2030 predicted number will keep rising, considering that “while AI’s ultimate impact on society and the corporate bottom line will take time to determine, companies are pouring in capital to build a new global system of data centers for the modern economy,” it added.
The shift could improve privacy, sovereignty, and sustainability while challenging Big Tech’s dominance over AI model deployment to deliver data center energy management solutions and data center renewable energy.
Economical Alternative to Cloud AI
For years, AI tasks have relied on data center energy efficiency metrics to run its massive cloud infrastructure. Each prompt moves from a user’s device to remote servers and back – known as inference.
The architecture holds more control and resource consumption with a few tech giants, demanding better data center energy solutions, where data center renewable energy leads to the calls of change.
The architecture also places control over data center energy management solutions.
“A very big AI model like GPT-120B… can be downloaded and deployed on Anyway Systems in a few minutes, requiring no more than 4 machines with 1 commodity GPU each,” the team explained.
“For years people have believed that it’s not possible to have large language models and AI tools without huge resources… but this is not the case and smarter, frugal approaches are possible,” Guerraoui said.
AI already is and needs to shift ever further from data center energy efficiency metrics to small local networks. Whether people will trust technology more once it finally runs inside their own walls is really what needs to be asked.
For years, consumers have entrusted personal information to distant servers in major corporations – without full clarity into how it would be processed.
With local AI systems, this situation is reversed.
Such systems have a strong chance of restoring trust in AI – especially the perception on data center energy use – and could very well increase the anxiety threshold over the delay of its adoption while giving individuals a deeper sense of ownership. Over time, it would shift human–machine interaction.
The shift also dares to challenge the industry’s dependence on data center energy solutions, and especially for governments around dominion energy data center capacity increase.
Why Data Centers Matter
In a recent TechTank Podcast episode, co-host Darrell West emphasized that “AI is the transformative technology of our time, yet undergirding its growing use is the need for state-of-the-art data centers.” These facilities require enormous investment, with some hyperscale centers costing up to a billion dollars and demanding advanced hyperscale data center energy solutions.
“In 2023, data centers consumed about 4.4% of America’s electrical power… we’re going to see probably somewhere between 6.7 and 12% by 2028,” said Co-host Nicol Turner Lee highlighted the growing energy burden.
She added that some facilities use “as much as 500,000 gallons per day” of water for cooling conditions that intensify attention on data center energy efficiency measures, data center cooling energy efficiency, and rising demand for data center energy efficiency.
To add to the one that has to include energy data center solutions.
AI Sovereignty and Sustainable Computing
EPFL researchers argue that cloud-based inference exposes users risks around privacy, security, and sovereignty. Running models locally avoids these vulnerabilities entirely, pushing institutions to rethink long-term commitments to data center renewable energy and data center energy efficiency metrics.
“Anyway Systems shines on inference but it could also help reduce the resources needed for training,” Guerraoui noted, describing early results showing minimal accuracy loss.
EPFL’s David Atienza called the system “an interesting and appealing technology that optimizes resource usage while ensuring data security and sovereignty.”
Backed by Switzerland’s Startup Launchpad AI Track, the platform of data center energy efficiency metrics is now being tested across companies and public institutions.
Guerraoui believes the shift is inevitable, where they “will be able to do everything locally in terms of AI… and we, not Big Tech, could be the master of all the pieces.”
The next AI chapter may not be written in the cloud, but machines are already reshaping expectations around data center energy efficiency, data center cooling energy efficiency, and data center energy efficiency measures.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.\