Autonomous Resource Corporation. Augmenting the industrial base.
One sentence: we build the operating system for advanced manufacturing as a network — closing the loop from materials discovery through qualified production into one continuously learning system.
The subtitle is the thesis. Augmenting — not replacing — the people, the apparatus, and the institutions that already exist. The loop that follows is what augmentation looks like in steel and superalloy.
Discrete teams. Disconnected datasets. Qualification measured in decades.
Every step in the manufacturing graph — discover, design, simulate, produce, inspect, test, qualify — sits inside its own silo, owned by a different team, often a different vendor, never sharing data.
The cost is structural: parts that take twelve to twenty-four months to qualify in a world that needs them in weeks. ARC's premise is that the loop has to be one system, not seven. Continuous. Network-scale. Evidence-grounded.
Compression is physically possible.
SpaceX took the Raptor engine through three full generations in five years. The result: four times cheaper, one-and-a-half times the thrust, 36% lighter, 52% less interface mass.
The proof point we lean on is not the rocket. It is that compounding hardware improvement of this magnitude is not a fantasy when discovery, design, simulation, production and test live inside one continuous loop.
One company. One engine. Five years. ARC is building that loop as a network — across every part the industrial base no longer makes.
The anchor customer faces a binding constraint.
Since 1984, 1,600 U.S. foundries have closed. PCC Airfoils now controls more than 80% of remaining domestic investment-casting capacity. PLA military pressure on Taiwan rose by roughly 300% in 2024.
The Department of War is the anchor customer because no other buyer carries a sustainment problem of this size against an industrial base reduced this far.
"The Manhattan Project for Digital Qualification of advanced manufacturing parts."Senior INDOPACOM operational leadership, 2026
ARCNet + ADAM. The operating system for the loop.
ARCNet — the Autonomous Resource Controller Network — is the orchestration, control, data and security layer that links every node in the network.
ADAM — Autonomous Discovery and Advanced Manufacturing — is the inference stack underneath: triage, world model, surrogates, anomaly detection, and the evidence-policy release gate.
Inference informs. Release is gated. ADAM is not a language model. It is a physics-grounded decision system whose output is binding only when the evidence package clears the gate.
The seven-step loop compounds.
Every loop run — discover, design, simulate, produce, inspect, test, qualify — generates training signal that feeds ADAM at the hub.
Every trained ADAM tightens the next loop. The flywheel is not metaphor; it is the mechanism by which a single facility cannot generate this volume of cross-modality, cross-material, cross-supplier evidence — but a network can.
Three pillars no competitor can replicate.
- 1-of-1 dataset. Desktop Metal's petabyte powder database plus ORNL's 40+ trillion voxel dataset. Decades of metallurgy and qualification data that cannot be retroactively acquired.
- 1-of-1 compute and apparatus. ORNL's Lux 1,000-GPU allocation plus the Manufacturing Demonstration Facility stack. Roughly $3B and three years of construction to replicate — and grid interconnection is the binding national constraint.
- 1-of-1 customer. The Department of War faces the largest sustainment problem against an industrial base significantly reduced over three decades of offshoring.
Standalone factories don't compound. Networks do.
Every advanced-manufacturing peer building today is building one factory. ARC is building the network that makes every factory smarter.
Evidence flows from nodes into ADAM at the ORNL hub; updated models and qualification authority flow back out. A smarter network is a better model; a better model wins more share; more share trains a smarter network.
Project Prometheus automates individual factories. ARC connects them into one continuously learning loop. This is the winner-take-most game.
The MPV network rollout.
Two staged deployments. ARCNet CONUS — the Oak Ridge hub plus seven advanced-manufacturing nodes — proves the topology under domestic conditions.
ARCNet INDOPACOM — an approximately $2B Greenfield digital backbone feeding forward nodes along the first island chain — proves it at theater scale.
ARC's posture across both is operator and backbone, not vendor. We run the network. Tenants run the parts.
Three compounding revenue streams.
Pay-per-use. Federal and commercial customers pay per task — dollars per GPU-hour, per kilogram, per build-hour. Fixed GPU cost remitted to SPV; neo-cloud margin eliminated.
Recurring software license. The "Bloomberg for advanced manufacturing" — a per-operator-seat platform license with compounding data-network effects.
AMaaS — advanced manufacturing as a service. Usage-based fulfillment through ARC Impact, our contract-manufacturing arm. Each stream compounds the data flywheel the other two depend on.
Prove one impossible thing completely.
The Series A trigger is a year-end 2026 proof package. Three milestones:
- 22 July 2026. NPS Monterey NVLink 72 pilot — already running.
- 3 November 2026. ARCNet continuous-loop demonstration.
- Year-end 2026. UxS and manned-aircraft component qualifications through ARCNet + ADAM.
The output is a computationally verified digital evidence package co-authored with ORNL. The round opens not on a deck, but on a part. Specific customer and partner identities are held NDA-required pre-disclosure.
Non-dilutive leverage shields the cap table.
The dilutive ask is $25M, with a $75M proof-gated follow-on after the year-end 2026 demonstration.
Sitting alongside is a potential $615M to $700M of non-dilutive capital across five years: an estimated $85–170M of Greenfield margin share, approximately $500M from DoW Pacific Theater AM, and approximately $30M from DoW advanced manufacturing funding. All non-dilutive figures are potential, not committed — specifics held NDA-required pre-signature.
The structure means equity buys ownership of a system most of whose buildout cost is appropriated, not raised.
Advanced manufacturing, AI for science, and ARC.
Pure advanced-manufacturing specialists — Hadrian ($1.6B), Divergent ($2.3B), Ursa Major ($600M) — solve one factory at a time. Hadrian and Divergent are candidate ARCNet tenants.
AI-for-science peers — Periodic Labs (~$1.5B), Lila Sciences (~$1.3B) — have the AI but no theater-scale apparatus.
The relevant showdown is ARC versus Project Prometheus ($38B). Three structural cuts:
- 1-of-1 substrate. Prometheus cannot retroactively buy the ORNL/Desktop Metal data.
- Network versus factory. Prometheus automates silos. ARC compounds across tenants.
- Anchor customer. Bezos has Blue Origin. ARC has the Department of War.