Chapter 01 — The platform is real

Colleagues —

July's Genesis Mission Summit was the first time this program looked like a platform instead of a press release. I am not going to recap the slogans. I am going to map the stack that 278 selected teams can now touch, and the few remaining on-ramps for everyone else.

I am also writing this as the operator of an AI Automation Factory — a shop that builds virtual assistants for sales and broker desks. Genesis is the national platform. The Factory is the desk-scale version of the same bet: coordination, not another model.

The platform, not the photo op

On July 22, Energy Secretary Chris Wright announced the first selections under the Genesis Mission Request for Applications: 278 awards drawn from what DOE called the largest response to a funding opportunity in its history. The split is specific: 87 led by DOE and NNSA laboratories, 168 by universities, 19 by companies, 4 by nonprofits. Behind those leads sit 342 participating institutions — 16 DOE/NNSA labs, 142 universities, 157 companies, 13 nonprofits, and 14 others.

Read the numbers as a design choice. This is not a lab-only program with university window dressing. Universities lead the plurality. Companies are in the room. The official caveat still matters: selection is for award negotiations, not a commitment to issue funding.

The White House, the same day, put more than $5 billion in federal commitments behind a whole-of-government expansion. More than 15 agencies will contribute awards, datasets, and facilities to a set of National Science and Technology Challenges, riding DOE's American Science and Security Platform. Separately, DOE announced more than $800 million in committed partner support through the Genesis Mission Consortium — compute credits, foundation-model access, cloud, expertise, and cash — from a consortium that now includes all 17 national labs, five NNSA plants and sites, and 41 industry, nonprofit, and philanthropic organizations.

HPCwire reported that the 278 teams will tap into $250 million in Genesis Mission funding, and that remaining Phase II awards (beyond the one nuclear award already named) were slated for later this year. Watch that second wave. Do not treat it as already shipped.

AmSC: the connective tissue, not another login

If you work at a DOE user facility you already know the failure mode: 17 labs and 28 user facilities sit on ESnet, and still require dozens of login systems. Inder Monga, ESnet's executive director and deputy project lead for the American Science Cloud, described AmSC as the fix — a common identity layer, interoperable service interfaces, AI-enabled orchestration, and intelligent storage so the distributed complex behaves as one system. Berkeley Lab launched that description in mid-July. AmSC itself launched in October 2025.

Oak Ridge runs AmSC as the federated, science-optimized environment that links DOE computing, experimental facilities, data, and high-performance networks. Pacific Northwest National Laboratory, which co-leads the Models pillar of Genesis, describes AmSC as exposing unified APIs, secure enclaves, and commercial-cloud integration — the backbone that ModCon rides on.

This is the part of Genesis that matters if you are not on an award: the platform is being built as a federated cloud with APIs, not as a single building in Tennessee. Qualified researchers are supposed to get pieces of it. Watch the identity layer and the data APIs more closely than the summit footage.

ModCon: models, cards, and the agent layer

Argonne leads the Transformational AI Models Consortium. DOE's ModCon page is unusually concrete for a federal AI program. Four core services sit under it: Baseline AI R&D (BASE), Best Practices for Scientific Workflows (BPSW), Data Brokers & Standards (DBS), and IP & Partnership Formation (IPPF). All of it is delivered through AmSC.

The artifacts to actually read:

Agent cards — a structured description of what an agent does, who provides it, how to invoke it, skills and tools, inputs/outputs, auth, runtime, intended uses, limitations, and safety. In Genesis, agent cards are how humans and other agents decide whether something is safe to compose into a workflow.
Data cards — provenance, access, and use constraints for every dataset, regardless of size or sensitivity.
Model cards — training data, methods, evaluation, licensing, compute environment, limitations.

That is the interoperability bet. If 278 teams each ship a one-off notebook, Genesis fails. If they ship cards that AmSC can register, you get a catalog.

HPCwire's July 22 report is explicit that awardees get AmSC (ORNL), ModCon (Argonne), and "agentic frameworks and other software." DOE's own award notice says the same in plainer language: AI agent frameworks, industry models, and HPC across the labs.

A window into the work: nine ORNL-led projects

I am not going to list 278 titles. Oak Ridge's July 22 release is a useful cross-section because the lab both leads nine Phase I projects and contributes to 32 others, while also operating AmSC and installing Lux.

The nine ORNL-led awards, with named PIs:

AI-assisted fusion power plant designs (Katarzyna Borowiec) — HPC tools plus robotic system design.
Low-energy AI for scientific discovery (Prasanna Date) — neuromorphic computing against scientific data.
Pipit, an AI "foundation model factory" for materials and manufacturing (Stephen DeWitt).
AI-enabled discovery of rare earth elements (Xin Gu).
AI-powered operations for self-driving science (Mariam Kiran) — coordinating agentic and scientific workflows across institutional boundaries.
AI foundation models for water forecasting (Dan Lu).
Modeling soil processes with AI (Benjamin Sulman).
AI-driven Earth system prediction for the TVA region (Dali Wang).
AI for next-generation fusion materials (Ying Yang).

That list is the mission in miniature: fusion, critical minerals, water-for-energy, manufacturing models, and the unglamorous problem of moving agent traffic across lab boundaries. Phase I is explicitly a demonstration and evaluation round — design the workflow, measure whether it actually accelerates discovery.

Department of Energy selects ORNL to lead nine Genesis Mission projects

The one Phase II: Prometheus, $60 million, nuclear

Among the 278, DOE named a single Phase II award: a three-year, $60 million nuclear-energy investment "to help deliver nuclear facilities faster and safer while cutting operating costs." Idaho National Laboratory leads Prometheus, a 32-partner collaboration with ORNL, Argonne, Sandia, four universities (NC State, Penn State, Tennessee, Texas), and a long industry list that includes X-energy, Oklo, Westinghouse, GE Vernova, NVIDIA, and AWS.

World Nuclear News, citing INL, reports the team has raised more than $200 million in industry cost share and $30 million in industry capital, and is designed to cut reactor deployment timelines and operating costs in half via human-in-the-loop AI across design, licensing, manufacturing, construction, operations, fuel fabrication, and legacy-document management. DOE's National Science and Technology Challenges document is even more specific on the target: "at least 2x schedule acceleration and greater than 50% operational cost reductions," using surrogate models, agentic workflows, autonomous labs, and digital twins.

Treat those as program goals, not results. The award is entering negotiations. The rest of Phase II, DOE said in July, comes later.

Two doors still open this month

SBIR/STTR Phase I, Genesis topics. DOE's Office of Technology Commercialization opened a $10 million Phase I opportunity — about 40 awards, four topics (biotechnology, quantum systems, materials with predictable functionality, AI-driven autonomous laboratories). Pitches are due Thursday, September 10, 2026, at 2:00 p.m. ET via the ConnectWerx AMP portal. Full applications are by invitation only. Phase I requests cap at $250,000 ($256,500 with TABA). A broader Phase I is promised for fall 2026; about $147 million in Phase II is open for prior awardees.

If you are a small business that belongs in this stack and you have not drafted the four-question pitch, that is the only calendar item that matters this week.

Open-weight science models. On August 7 DOE launched the Genesis Open Models Initiative. The first model is Genesis-Science-1, developed with Arcee. The first contribution windows close this month (pretraining August 14; fine-tuning August 25). Additional deadlines are rolling, expected every three months. That is the public-science path into the model layer if you are not on an RFA award.

The AI Automation Factory

coordination, not models

The bottleneck on Genesis is no longer cognition. It is coordination. That is also the Factory.

I am building an AI Automation Factory: a shop that trains virtual assistants for sales and broker desks. Not a chatbot. A VA that sits on a real workflow — intake, follow-up, files, calendar — with a human checkpoint where a wrong send would hurt.

First workflow is my sister Sudipa's. She is a real-estate broker and operator. The product is the same as AmSC in miniature: one identity layer, one set of services, one catalog of what the assistant is allowed to do. If 278 teams each ship a one-off notebook, Genesis fails. If I ship a one-off prompt for Sudipa, the Factory fails the same way.

The offer, when I sell it: $2,500 setup + $2,000/month. One desk, one loop. Goal: first paid client and cash in the account by 31 December 2026. Then 2027 is a lead machine — this newsletter, plus one weekly ask. No spray.

September is the platform month. For Genesis that means AmSC and ModCon. For the Factory it means: stop treating a VA as a speech. Ship the connective tissue.

Worth clicking

Launching the Genesis Mission — Executive Order 14363. The legal architecture: American Science and Security Platform, agents, datasets, robotic-lab review, 270-day initial operating capability.
Genesis Mission — the official program site. Goal language, domain map, "how to get involved."
Genesis Mission National Science and Technology Challenges — the living challenge portfolio, not the summit talking points.
Energy Department Announces Collaboration Agreements with 24 Organizations to Advance the Genesis Mission — the December 2025 MOU list (AMD, NVIDIA, OpenAI, Anthropic, Cerebras, CoreWeave, and the rest). Names, not vibes.
Lux: ORNL's AI + HPC Supercomputer — half of Lux's 3.5 million node-hours are reserved for Genesis. The other half is paid proprietary access. That is a commercial door.

Next month: watts. The binding constraint on this platform is not another foundation model. It is power, interconnection, and whether data-center load eats the grid Genesis is supposed to help plan.

If you already have the award and not the keys

I am taking one Genesis team from award to a first cross-facility run on AmSC: accounts, Globus whitelist, storage, first job, honest status. Two weeks. If you are the computational lead still collecting logins, or you know that person, reply to this email. I will tell you on a twenty-minute call whether it is a fit.

The AI Automation Factory. Virtual assistants for sales and broker desks. First workflow: a real-estate broker/operator (Sudipa). Offer: $2,500 setup + $2,000/month. First paid client / cash by 31 Dec 2026. 2027: this newsletter + one weekly ask. No spray.

Reply if one loop on your desk is eating the week. deep-heap.com

Deep Datta
Deep Heap

Intelligence has a substrate.

I write this for people who already know the labs, the machines, and the constraints — and for operators who need a virtual assistant that actually runs a desk. If I cannot verify a source, I leave it out.