Oak Ridge National Laboratory AI: Frontier, Lux, and the “Secret City” Supercomputer Stocks

Marc Chaikin’s “The End of Elon?” presentation opens on “this high-security, triple-barbwire fence in rural Tennessee… At the same top-secret site that launched the Manhattan Project.” He borrows The Atlantic’s nickname, “the secret city,” and spends the next hour arguing that what is being built there will reshape AI. The lab is Oak Ridge National Laboratory, and unlike a lot of promo settings, it deserves the drama. This article explains what Oak Ridge actually does in AI, what Frontier and Lux are, and how an investor gets exposure.

Why Oak Ridge

Oak Ridge was built in 1942 to enrich uranium for the Manhattan Project. The town did not appear on maps, hence “secret city.” Today it is the Department of Energy’s largest science and energy laboratory, and for most of the last two decades it has hosted the fastest computer in the United States.

Chaikin gets the significance right: “If I’m right, this will be the biggest moment for technology since the first atom bomb exploded on July 16, 1945.” He is quoting the White House’s own framing of the DOE AI program as “a historic national effort, comparable in urgency and ambition to the Manhattan Project,” which is language the administration used when announcing the Genesis Mission.

Frontier is the existing flagship. Chaikin’s description: “Frontier can perform more than 1 quintillion calculations per second… if every person on Earth could complete one calculation per second… It’d take all of mankind four years to solve the problems Frontier can solve in one second.”

That is accurate. Frontier was the first exascale computer, exceeding one exaflop (a quintillion FP64 operations per second) in 2022. The other details check out too: “One Frontier cabinet weighs the same as two pickup trucks. Frontier can store 33 times the amount of data housed in the entire Library of Congress.” It runs on AMD EPYC CPUs and AMD Instinct MI250X GPUs, integrated by HPE Cray.

Chaikin’s key claim: “Without changing a single chip… Frontier can already crush every other AI model out there when it comes to scientific breakthroughs. That includes the Colossus data center located out of Memphis.” In the narrow sense he means, double-precision scientific simulation, that is defensible. Colossus is built for training language models at low precision and would be a poor fit for climate modeling or fusion plasma physics. In the broad sense, Frontier cannot train a frontier language model competitively. They are different tools.

Lux: the “AI micro cluster”

The new piece is Lux. Chaikin: “The official name for this new breakthrough device is Lux. It’ll be the world’s first FP64-powered AI supercomputer, Molly. The designers call Lux ‘the first dedicated U.S. AI factory for science.’”

Lux was announced in October 2025 as a partnership among the DOE, Oak Ridge, AMD, HPE, and Oracle. It is built on AMD Instinct MI355X accelerators and was slated to come online in early 2026, which matches Chaikin’s “they’re connecting the final cables right now.” Its purpose is to run AI models trained on DOE scientific data, which is a different job from Frontier’s simulation work. Its power draw is in the tens of megawatts; Chaikin estimates “as little as 24 megawatts,” roughly one percent of Colossus’s planned two gigawatts.

The promo also mentions Discovery, “another new AI supercomputer” that will “accelerate training of scientific AI models.” Discovery is the successor to Frontier, built on AMD’s next-generation MI430X chips and due around 2028. The presentation attributes MI430X to Lux, which conflates the two projects, but AMD is at the center of both.

Chaikin’s term “AI micro cluster” is his own. The actual descriptor used at Oak Ridge is “AI factory for science.” Either way, the point is real: a small, precise machine doing work that a giant, imprecise one cannot.

What Oak Ridge does with the machines

The presentation lists targets that are real Oak Ridge programs. “Oak Ridge National Lab runs the Quantum Science Center. And it just powered a new quantum-computing breakthrough called ‘ion trapping.’” The Quantum Science Center is a DOE National Quantum Information Science Research Center headquartered at Oak Ridge. Chaikin also cites the Fortune headline that “The U.S. just bet $1 billion that AI supercomputers can turn most cancers from ‘death sentences’ to ‘manageable conditions’ within 8 years,” which refers to the DOE and National Cancer Institute collaboration that runs partly on Oak Ridge hardware.

On fusion, he references “an AI-Fusion Digital Convergence Platform” and a Lawrence Livermore AI model that “accelerated calculation speeds 100 million-fold” for plasma simulation. That is Livermore rather than Oak Ridge, but Oak Ridge hosts the U.S. ITER project office, so fusion modeling is part of its portfolio.

The stocks

AMD (AMD). The free pick, and the clearest exposure. Frontier, Lux, and Discovery are all AMD machines. Chaikin: “At least $1 billion dollars are about to hit this company’s account.” AMD’s FP64 lead over Nvidia is the reason Oak Ridge has been an AMD shop since 2019.

Hewlett Packard Enterprise (HPE). HPE Cray is the integrator for every current Oak Ridge flagship. Its Slingshot interconnect is the fabric inside Frontier and will be inside Discovery. HPE is the closest thing to a pure-play “supercomputer builder” on public markets, though the segment is a modest share of HPE’s total business.

Oracle (ORCL). Named partner on Lux and Discovery for cloud and software integration.

Ciena (CIEN). Not named in the promo but a strong candidate for Chaikin’s teased networking pick. Ciena supplies ESnet, the DOE’s dedicated fiber network linking Oak Ridge to the other 16 national labs. We work through that reasoning in AI Supercluster Stocks.

Tennessee Valley Authority powers Oak Ridge but is a federal corporation with no stock.

Our take

Oak Ridge is the rare promo backdrop that is more impressive in fact than in marketing. The lab really does host the machines Chaikin describes, they really are AMD-based, and the DOE really is spending on an AI expansion. Where we part ways with the presentation is on the “1 trillion times more powerful than Colossus” framing, which is register-width arithmetic rather than a performance measure, and on the idea that Lux makes commercial AI clusters obsolete. Frontier and Colossus will both be busy for years doing different things.

For the full assessment of the pitch and the Power Gauge Report offer, read our End of Elon / American Atlas review. For Chaikin’s framing of the timeline, see The AI White Swan of 2027.

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.