Colleagues —
Five issues ago I started from the 278. If that list is a closed caste, Genesis is a two-year program with a nice website. The order, the collaboration page, and the user-facility system all say it is not supposed to be closed. This is the field guide I would hand a founder, a university PI who missed the RFA, or a lab-adjacent operator who has something the platform needs.
The Factory has the same outsider problem. If the only desk that ever runs is Sudipa's, I built a family tool, not a shop. This is how a broker who was not in the room attaches: one loop, a two-week build, a price.
The consortium is the intended front door
DOE's collaboration page is the current map. The Genesis Mission Consortium connects industry, academic, nonprofit, and philanthropic organizations to the 17 labs and NNSA. Working groups named on that page:
- AI Model Development and Validation
- Data Integration and Standards
- High-Performance Computing and Cloud Infrastructure
- Robotics and Automation
- Workforce Development and Training
As of the July 22, 2026 summit, the consortium included all 17 DOE national laboratories, five NNSA plants and sites, and 41 industry, nonprofit, and philanthropic organizations, with DOE "now expanding engagement to academic institutions." Partner commitments announced that day exceeded $800 million — compute, model access, cloud, expertise, research partnerships, direct funding.
The inbox they publish is still genesismission-partnerships@hq.doe.gov. Subscribe on the same page for consortium updates. Under Secretary for Science Darío Gil is DOE's director for the mission.
I will not pretend an email is a CRADA. It is the documented starting point.
- Genesis Mission Collaboration
- U.S. Department of Energy Announces More Than $800 Million in Partner Commitments to the Genesis Mission
The named industry list (December 2025 MOUs)
On December 18, 2025, DOE announced MOUs with 24 organizations. The list, as published:
Accenture; AMD; Anthropic; Armada; Amazon Web Services; Cerebras; CoreWeave; Dell; DrivenData; Google; Groq; Hewlett Packard Enterprise; IBM; Intel; Microsoft; NVIDIA; OpenAI; Oracle; Periodic Labs; Palantir; Project Prometheus; Radical AI; xAI; XPRIZE.
DOE's conditions on that page are easy to miss and worth keeping: products for Genesis are to be architecture-agnostic; organizations had either responded to an RFI or already had active lab projects. The consortium later grew to 41. Treat the 24 as the public seed list, not the census.
If you are a founder, the useful question is not "how do I become the 25th logo." It is: which working group (models, data, HPC/cloud, robotics, workforce) is the honest fit, and do you have a CRADA-shaped piece of work a lab already wants.
User facilities: the door that already existed
Before Genesis had a name, DOE Office of Science already ran 28 user facilities with access "determined on a competitive basis using peer review," used by tens of thousands of researchers a year. Supercomputers, accelerators, light sources, neutron sources, nanoscience centers, genomics. That system is still the most reliable way for an outsider with a scientific question to put hands on the experimental half of the American Science and Security Platform.
AmSC's job, per Berkeley Lab, is to collapse the "dozens of different login systems" across those 28 facilities and 17 labs into one identity and API fabric, and to extend that fabric to commercial cloud. You do not wait for a Genesis award to submit a general-user proposal to APS, LCLS, SNS, NERSC, or OLCF. You do the proposal. Then you watch whether your data path is AmSC-native yet.
OLCF's Lux adds a paid variant: half the node-hours are proprietary, six months to five years, 175k node-hour minimum. That is how a company that will never lead an RFA still trains on the same MI355X cluster the mission uses.
- Office of Science User Facilities
- How the Genesis Mission's American Science Cloud Advances Innovation
- Lux: ORNL's AI + HPC Supercomputer
SBIR/STTR: the small-business door
The July 22, 2026 OTC round was the first Genesis-branded Phase I under the Office of Technology Commercialization and ConnectWerx: ~$10 million, ~40 awards, four topics (biotech, quantum+AI, materials inverse design, autonomous labs). Pitch-first, invitation-only full applications, $250k cap. DOE also opened about $147 million in Phase II for eligible prior awardees, and promised an additional Phase I in fall 2026.
If you are reading this in February 2027, the September 10, 2026 pitch window is closed. The durable facts: OTC redesigned the process; ConnectWerx is the intermediary; Genesis topics are a standing way to attach; a broader topic set was scheduled behind the first four. Check the ConnectWerx hub and the OTC announcement for whatever succeeded the fall 2026 round — I will not invent a 2027 solicitation that I cannot fetch.
- DOE Announces $10M in SBIR/STTR Funding Opportunity Supporting the Genesis Mission
- FY26 Phase I – Genesis Mission
Open models and cards: the contribution door
You can contribute without a CRADA if you have data, evaluations, or fine-tunes. Genesis-Science-1 (with Arcee) was announced as an open-weight science model with rolling contribution windows every three months after the August 2026 close. ModCon wants contributed agents, models, workflows, and tools through a registration path, documented with agent/data/model cards.
This is the lowest-friction path for a research group that already has a corpus the labs do not. It is also the path that will be judged on provenance. DOE asked for transparent training data and evaluation procedures. Bring those, or do not bother.
- U.S. Department of Energy Launches the Genesis Open Models Initiative – Apply Now!
- ModCon: Transformational AI and Data
Talent: what the order asked for, not what I wish it created
Section 5(b) of EO 14363 directs OSTP to coordinate competitive programs for research fellowships, internships, and apprenticeships focused on AI applied to the national challenges, including placement at DOE labs and other federal research facilities, with access to the Platform. I have not seen a single, named Genesis fellowship program I can verify as of the sources I fetched. The mandate is real. The implementation is something to watch, not something to announce.
Workforce is, however, a named consortium working group. If you run a training program that actually puts people on AmSC or on a self-driving lab, that is the working group to walk into.
PNNL's public Genesis page is a useful org chart for the models side: Court Corley (ModCon deputy director / co-PI), Nathan Hodas (Baseline Capabilities co-lead), Neeraj Kumar (AmSC liaison), plus AmSC commercial-cloud leads Roger Cass and Robert Rallo. Named people beat "we should partner with the labs."
How I would attach, ranked
- You have a scientific experiment — general-user proposal at the relevant Office of Science facility. Do not wait for Genesis branding.
- You have a model, dataset, or agent — ModCon cards + Open Models windows + AmSC contribution path.
- You are a U.S. small business in robotics, materials, quantum, or biotech — SBIR/STTR Genesis topics; watch OTC/ConnectWerx for the next Phase I.
- You need GPUs at OLCF scale and can pay — Lux proprietary half.
- You want a structural partnership — consortium working group, then
genesismission-partnerships@hq.doe.gov, then a CRADA/user agreement, not a tweet. - You hire — watch for the fellowship/internship machinery the EO required; until it exists, hire people who have already used a user facility.
The mission's own success metric is on genesis.energy.gov: double the productivity and impact of American research and innovation within a decade. That does not happen inside 278 teams. It happens when the platform those teams are standing up is boringly usable by the next thousand.
The AI Automation Factory
outsiders attach — first client, then the next desk
Five issues ago I started from the 278. If that list is a closed caste, Genesis is a two-year program with a nice website. The Factory has the same failure mode: if the only desk that ever runs is Sudipa's, I built a family tool, not a shop.
How I would attach, ranked, for a broker or sales operator who was not in the room:
- You have one painful loop (missed-call, text-back, booking, file chase). Reply with that loop. Do not wait for a platform brand.
- You can pay $2,500 to stand it up and $2,000/month to keep it running.
- You do not need to become the 25th logo. You need a bounded build with a runbook — the CRADA-shaped piece of work a shop already wants.
- After the first paid client, 2027 is a lead machine: this newsletter, one weekly ask, no spray.
Sudipa is the first door, the way a general-user proposal at a user facility is the door that already exists. The 278 are a cohort, not a caste. The first client is a cohort of one. The next desks only happen if the assistant is boringly usable.
Worth clicking
- Genesis Mission Collaboration — working groups, subscribe, partnership email.
- Energy Department Announces Collaboration Agreements with 24 Organizations to Advance the Genesis Mission — the named companies.
- Office of Science User Facilities — the access path that already works.
- DOE Announces $10M in SBIR/STTR Funding Opportunity Supporting the Genesis Mission — small-business template, even after the first window.
- Launching the Genesis Mission — Section 5 is the external-engagement chapter: CRADAs, IP, vetting, fellowships.
That is six months. The arc was: the platform is real; power is the constraint; labs have to close the loop; agents need contracts; the machines have to stay full; outsiders have to be able to walk in. If a later issue needs a seventh, it should be about measured scientific output — publications, qualified materials, licensed reactors, not another commitment total.
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.