The Lux FP64 Supercomputer at Oak Ridge
Lux is the machine at the center of Marc Chaikin’s “The End of Elon?” presentation, and it is the rare promo centerpiece that is a real, named, locatable object: “The official name for this new breakthrough device is Lux… the world’s first FP64-powered AI supercomputer… The designers call Lux ‘the first dedicated U.S. AI factory for science.’” It is being assembled at Oak Ridge National Laboratory in Tennessee, the same site whose WWII-era secrecy earned it The Atlantic’s name, “the secret city.” For investors trying to separate the American Atlas thesis from promo fog, Lux is the anchor: here is what it is, when it arrives, and who builds it.
What Lux is
Lux is the first of a new class of AI data center built for double-precision, FP64, scientific computing rather than the low-precision FP16 inference that runs ChatGPT-style AI. The scale contrast with today’s flagship AI campuses is the presentation’s most vivid material: Lux is “about the area of a 7-Eleven convenience store… Literally 0.001% the size of Colossus,” Elon Musk’s 1,156-acre campus, drawing “as little as 24 megawatts” against Colossus’s roughly two gigawatts, yet, for precision-dependent scientific work, “more than a trillion times more powerful.”
The context that makes the small footprint possible is the whole point: the floating point problem. Chaikin’s demonstration, ChatGPT failing to count the C’s in “broccoli,” shows that low-precision hardware rounds numbers in ways that are harmless for chat and fatal for science: “That lack of precision can’t be solved with better programming… Because the floating point problem isn’t a software problem. It’s a hardware problem.” Fusion simulation, drug discovery, quantum algorithms, all of it needs FP64, and Lux is the first AI-era machine purpose-built to deliver it.
The timeline
The presentation dates the milestone twice: the opportunity window “starting as soon as September 29th,” and the physical machine “when they fire up this machine, which is expected before the end of this year… They’re assembling Lux as we speak… they’re connecting the final cables right now.” The project’s authorization runs through Executive Order #14363, which “charged the Department of Energy with accelerating AI breakthroughs in America,” with IBM veteran Dario Gil recruited to lead. A senior administration official calls it “the world’s most complex and powerful scientific instrument ever built,” and the White House framing is “a historic national effort, comparable in urgency and ambition to the Manhattan Project.”
The network it seeds
Lux alone is the first node. The plan Chaikin describes, tracking public DOE announcements, is “nine new AI micro clusters” (named Minerva, Janus, Tara, and Discovery among them) plus “more than 50” converted DOE supercomputers, “Frontier, El Capitan, Aurora, Tuolumne, Venado, Perlmutter, Crossroads, and Polaris… all becoming part of one giant AI breakthrough machine… America’s national AI super brain.” Chaikin names the network himself: “So I’m calling this AI supercluster American Atlas.” The claimed effect, “a sudden 360-fold acceleration in major AI-powered breakthroughs… going from one mile per hour to 360 miles per hour,” and “new breakthroughs in five days compared to five YEARS for Colossus.”
The investment angle
The picks follow the hardware. The free one is AMD, “the first company on my American Atlas ‘buy’ list… Advanced Micro Devices, ticker symbol AMD,” whose Instinct MI-series accelerators already power Frontier and El Capitan, with the presentation tying “at least $1 billion” of incoming demand to the buildout. The paid American Atlas Portfolio report covers “17 companies that have signed deals… They’re all working directly on building American Atlas,” some “absolutely tiny compared to AMD,” led by a networking company whose “advanced networking solutions” are “the unsexy stuff that will make the next stage of the AI revolution possible.” The honest note on the chip claim: the transcript shows the MI430X as the sovereign-AI accelerator before pivoting to the “brand-new MI455X chips. And we believe they’ll soon be powering Lux,” a hedge readers should register.
Questions readers ask about Lux
What exactly is Lux? The Department of Energy’s new supercomputer at Oak Ridge National Laboratory in Tennessee, described in the presentation as “the world’s first FP64-powered AI supercomputer,” and the first node in a network the pitch names American Atlas: nine new AI micro clusters plus more than 50 converted DOE supercomputers. The facility footprint is about the area of a 7-Eleven convenience store, drawing as little as 24 megawatts.
Why does Oak Ridge matter historically? It was the secret city of the Manhattan Project, and the administration’s own framing leans on that lineage: “a historic national effort, comparable in urgency and ambition to the Manhattan Project.” A machine aimed at fusion-reactor design, drug discovery, and materials science, built where the atomic age was engineered, is a story that writes itself.
Which chips will power it? The presentation shows AMD’s MI430X Instinct Accelerator, “engineered specifically for sovereign AI and scientific computing,” then pivots to the brand-new MI455X generation with the claim “we believe they’ll soon be powering Lux.” That “we believe” is doing real work: treat the specific chip-generation claim as expectation rather than contract. AMD is still the free pick on solid ground, since its Instinct accelerators already power Frontier at Oak Ridge and El Capitan at Livermore. Our full review covers the timeline and the 17-company portfolio.
Where to learn more
Start with our End of Elon? American Atlas review, then the FP64 explainer and the Oak Ridge supercomputer detail. The full presentation and the American Atlas Portfolio offer are at the Chaikin Analytics page, discounted from $499 with a 30-day refund guarantee.
NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.