Genesis Mission: The Manhattan Project for AI. Kind of.

Playback speed

Share post at current time

Share from 0:00

0:00

Transcript

Genesis Mission: The Manhattan Project for AI. Kind of.

The Drey Dossier

May 11, 2026

The last time I wrote to you about the ballroom, the ballroom was the lid. We knew the donor list. We knew which AI companies were paying for a renovation we couldn’t see the bottom of. Trump himself had confirmed that there was, in fact, a military facility being built underneath, and the ballroom was the shed on top. What we didn’t yet have was the name of the thing they were paying to install.

We have it now.

The program is called the Genesis Mission. The federal government announced it on November 24, 2025, in the same press release where they hacked the budget for federal scientific research by thirteen percent. They are defunding the thing while funding the thing. That alone would be worth a piece. The deeper story is the architecture of what is being built.

The Handshake

The Manhattan Project did not end when the war did. The factories stayed open, the contracts kept running, and by 1977 the whole operation had been folded into a new cabinet-level department called the Department of Energy. Same labs, same buildings, same contractors. New name on the door.

The deal that came out of it has now been running for eighty years. The seventeen national laboratories that descended from the original wartime cities are owned by the federal government. The land, the buildings, the supercomputers, the data, all of it sitting on federal property and paid for with taxpayer money. Private companies run the day-to-day under contracts worth billions. Argonne is run by a company called UChicago Argonne LLC, which the University of Chicago set up specifically to hold the contract. Oak Ridge is run by UT-Battelle. Los Alamos is run by Triad National Security LLC, a partnership between the University of California, Battelle, and Texas A&M.

The arrangement goes back to 1942, when the federal government needed thousands of scientists and needed them fast, and federal hiring couldn’t move that quickly. The deeper reason it has lasted eighty years is that science and government don’t really fit together cleanly. Science needs flexibility, risk tolerance, the freedom to fail, and the kind of pay that gets you the best people. Federal bureaucracy is built for the opposite. So at some point somebody decided the private side would handle the day-to-day, and in exchange the public would stay in charge of the mission.

Public funding. Public ownership. Private operation. Public oversight.

It is a handshake, and it works only if all four pieces remain intact, because each piece is the only thing keeping the others honest. The public funds the work, so the work serves the public. The public owns the labs, so the labs cannot be sold out from under us. Private companies run the day-to-day, because the day-to-day needs more flexibility than a federal agency can offer. And public oversight keeps the whole arrangement balanced, because without it, the people running the labs have every incentive to forget who paid for them.

In April of 2025, the Trump administration fired all twenty-two members of the National Science Board. The same pattern hit advisory committees at the National Science Foundation, the EPA, the CDC, and the FDA. The National Science Foundation lost more than thirty percent of its staff and over a billion dollars in canceled grants. The federal scientific oversight infrastructure was being torn down at the same moment Genesis Mission was being built.

The handshake is now three pieces. The piece that is missing was the only one that was never optional.

Genesis Mission, Officially

On paper, the Department of Energy has been ordered to take the federal scientific datasets it holds, train AI agents on them, and turn those AI agents loose on the research more or less by themselves. The press release named two dozen private partners. Oracle, Microsoft, Google, NVIDIA, OpenAI, Palantir, Anthropic. The largest AI companies in the world all signed on at the same time. The mission identifies twenty-six scientific challenges these AI agents are supposed to tackle, ranging from cancer research to fusion energy to quantum computing. The official pitch is that the program will double American scientific productivity within a decade.

The program is incredibly difficult to follow, in a way that goes beyond ordinary federal opacity and into the kind of deliberate disorientation you find in a carnival funhouse, the kind built by someone with a fat taxpayer budget and a very low opinion of his audience. Reporting on it is physical labor. It goes by about ten different names:

There is the Genesis Mission, which is the name you give a federal program when you want the press release to read like a movie trailer. There is the American Science and Security Platform. There is the American Science Cloud, which is somehow a different thing from the Platform. There is the Transformational AI Models Consortium, which appears to be a fourth or fifth distinct entity. And, definitely not finally but as far as I got, there is the Genesis Mission Consortium, which is somehow a separate entity from the Genesis Mission itself.

Each name has its own website, and each website is a different mirror in the same funhouse. Each one has its own partner list, its own funding documents, its own announcements. None of them link to each other in any consistent way, and most of them contradict each other on basic facts about what the program does, who is in it, and where the money goes.

This is the architecture working exactly as designed. A federal program called by ten different names is a federal program no journalist can summarize in a tweet, no congressional staffer can audit in less than a season, and no taxpayer can follow without quitting her day job. By the time you finish mapping which name goes where, the thing the names were built to hide has its roof on and its servers humming.

The funding follows the same logic. The Department of Energy announced 320 million dollars in Genesis Mission investment in December 2025. Most of that money turned out to be existing appropriations that had been renamed. The Trillion Parameter Consortium existed before. SciDAC existed before. DIII-D existed before. Genesis Mission was the new paint. The executive order itself orders DOE to identify existing federal computing resources to support the mission, which is a bureaucratic way of saying find the money we already gave you and rename it.

Solstice

The redirection needs a destination, and the destination is a computer. A very, very large computer. Several of them, in fact. Equinox at Argonne, Lux at Oak Ridge, Mission and Vision at Los Alamos, and Discovery, also at Oak Ridge.

The largest one, by a lot, is being built by Oracle. It is called Solstice. It will run on 100,000 NVIDIA GPUs and draw 200 megawatts of continuous power, which is twice the size of the current world record holder for AI supercomputers. It is designed for autonomous AI agents, which means AI systems that run experiments on their own, analyze the results on their own, design the next experiment on their own, and execute it on their own, without humans in the loop. Solstice is also classified-capable, which means the same hardware running open scientific research can run intelligence community workloads alongside it.

According to Glenn Lockwood, who used to design federal supercomputer storage systems at the Department of Energy and now tracks these projects at a level of detail nobody else in public reporting touches, the basic disclosures that come with a system this large are missing. As of April 2026, Lockwood wrote that it remains unclear who is paying for Solstice, where it will be sited, or what science research it is actually for.

It is a remarkable amount of supercomputer to not yet know the address of.

The official story is that Solstice is going to live at Argonne National Laboratory in Illinois. Argonne already has two supercomputers on that campus, and Solstice would need roughly ten times the floor space and more than twice the power, with an entirely new cooling infrastructure. The math doesn’t work. Where it actually goes is, for the moment, an open question.

What It Eats

A supercomputer that size, designed to run autonomous AI agents, has to be trained on something. Genesis Mission is the federal layer that wires the something into the somewhere.

The Department of Veterans Affairs runs a program called the Million Veteran Program. A million American veterans have already had their full genomes sequenced as part of it, which makes it the largest genomic database in human history. The data is stored at Argonne National Laboratory.

Every cancer case in the United States gets reported to a federal database, and Oak Ridge has been integrating that data with imaging and treatment records for years.

The Advanced Photon Source at Argonne is the most advanced biological imaging instrument in the world. It can see proteins, viruses, and molecular structures at atomic resolution.

And on top of the genomes and the cancer data and the biological imaging, there is something else. Years before Genesis Mission existed on paper, Oracle won a contract from the Pentagon to run a completely separate program. It is an electronic health records platform that covers 9.6 million Americans, including active duty service members, veterans, their families, and their dependents. It launched in 2017, it connects more than 700 military medical facilities around the world, and it holds every diagnosis, every prescription, every lab test, every imaging study, every mental health visit, every immunization, and every deployment-related health screening for nearly ten million Americans, going back almost a decade. It is the most comprehensive longitudinal health dataset the federal government has ever assembled in a single system. It is currently being migrated into Oracle’s cloud. It is currently being deployed inside the National Security Agency.

The name of that program is MHS GENESIS.

The largest clinical health dataset the federal government has ever built is called GENESIS. The largest AI program the federal government has ever announced is called the Genesis Mission. Both are run by Oracle. One holds the data. The other will train models on data of exactly that kind. Both are being routed through the National Security Agency, which is not a hospital.

If there is a longer version of subtle in federal naming conventions, I have not found it.

Where It Goes

Back to Solstice. A supercomputer twice the size of any AI machine the federal government has ever built, designed for autonomous agents that operate without humans in the loop, classified-capable so intelligence workloads can run on the same hardware. Built by the same company that runs the clinical health records of ten million Americans, on the same cloud those records are being migrated into, under the same broader program whose name happens to be half of the dataset’s name.

Where do you put a thing like that?

Argonne can’t fit it. None of the other labs Genesis Mission has named in its various press releases can fit it either. You put it somewhere defensible. Somewhere under the operational control of the executive branch. Somewhere already engineered to survive the kinds of attacks a facility like that would attract.

We have already met that building. It is in the news for being demolished and rebuilt as a ballroom.

The last piece I wrote argued the ballroom is the lid. This piece argues what the lid is on. No one in the federal government has announced that Solstice is going under the East Wing, and no one is going to. What they have admitted, in court filings related to the demolition lawsuit, is that a “medical facility” is being built underground.

In a piece written before any of us had heard of Genesis Mission, the phrase “medical facility” sounded like a clinic. After MHS GENESIS and Solstice and the federal health records of ten million Americans and an AI program built to train autonomous agents on exactly that kind of data, the phrase means something else. It means the place where medicine becomes infrastructure. Where health data becomes national security data. Where the body becomes the battlefield.

I think we got a little closer to knowing what is underneath.

A note. To my mom, who gave me her courage, her curiosity, and apparently her audacity, because now I’m on the internet asking what octogenarian tech billionaires are doing with federal data. She says I’m her hero. She’s mine. Always has been.

Happy Mother’s Day, I love you a whole lot.