US AI supercomputer Nexus will compute faster than 8 billion humans combined
The U.S. research community is set to gain a major AI-powered boost.
Georgia Tech and its partners have secured $20 million from the National Science Foundation to build Nexus, one of the nation’s fastest supercomputers, built to accelerate scientific discovery using artificial intelligence.
Once completed in spring 2026, Nexus will deliver over 400 quadrillion operations per second, the equivalent of everyone in the world continuously performing 50 million calculations every second.
AI muscle for all
Designed specifically for artificial intelligence and high-performance computing (HPC) workloads, the system will help tackle challenges ranging from drug discovery and clean energy to climate modeling and robotics innovation.
“Georgia Tech is proud to be one of the nation’s leading sources of the AI talent and technologies that are powering a revolution in our economy,” said Ángel Cabrera, president of Georgia Tech.
“It’s fitting we’ve been selected to host this new supercomputer, which will support a new wave of AI-centered innovation across the nation. We’re grateful to the NSF, and we are excited to get to work.”
Unlike traditional supercomputers, Nexus emphasizes accessibility. Researchers from institutions across the country, not just major tech hubs, will be able to apply for access through the NSF.
The system is also engineered with user-friendly interfaces to make advanced AI tools more approachable for scientists in diverse fields.
“The Nexus system’s novel approach combining support for persistent scientific services with more traditional high-performance computing will enable new science and AI workflows that will accelerate the time to scientific discovery,” said Katie Antypas, National Science Foundation director of the Office of Advanced Cyberinfrastructure.
“We look forward to adding Nexus to NSF’s portfolio of advanced computing capabilities for the research community.”
Nexus will combine massive computing power with advanced memory and storage capabilities, including 330 trillion bytes of memory and 10 quadrillion bytes of flash storage — equivalent to about 10 billion reams of paper.
Stacked, that’s a column reaching 500,000 km high — enough to stretch from Earth to the moon and a third of the way back.
Supercharged brains, faster science
That scale allows it to handle complex problems and massive datasets crucial for cutting-edge research. It will also feature ultra-fast data transfer systems, ensuring researchers spend less time waiting for information to move between systems and more time focused on discovery.
The project is part of a national collaboration. Georgia Tech is building Nexus in partnership with the National Center for Supercomputing Applications (NCSA) at the University of Illinois Urbana-Champaign. Their systems will be linked via a new high-speed network, creating a shared national research infrastructure aimed at democratizing access to AI tools.
“This supercomputer will help level the playing field,” said Suresh Marru, principal investigator of the Nexus project and director of Georgia Tech’s new Center for AI in Science and Engineering (ARTISAN). “It’s designed to make powerful AI tools easier to use and available to more researchers in more places.”
Georgia Tech will begin building Nexus this year, reserving up to 10 percent of its capacity for on-campus research while managing system access through an NSF review process.
Once operational, Nexus is expected to not only speed up progress on today’s toughest scientific problems but also open doors to discoveries that researchers haven’t yet imagined.
Nexus builds on Georgia Tech’s previous experience with its precursor project, HIVE, and the university’s state-of-the-art CODA data center.
“With Nexus, Georgia Tech joins the league of academic supercomputing centers. This is the culmination of years of planning,” said Srinivas Aluru, Regents’ Professor and senior associate dean in the College of Computing.
Source: Interesting Engineering
Hair-thin chip transfers 100 million books in 7 minutes with just a spark of power
US AI supercomputer Nexus will compute faster than 8 billion humans combined