The World Foundation Model
for critical minerals.
Physical AI that understands the entire operation, orebody, plant, and tailings, and recovers the metal that proven ground still holds. We bring the intelligence and the capital. You share the upside.
The world isn't short of metal. It's short of recovery.
Every proven deposit gives up only part of what it holds. The rest, often a third or more, ends up in tailings, locked in low grade, or stranded in ground that doesn't pay at today's recoveries. That metal is already found, already mined, already paid for. Getting it back is the largest prize in mining, and the one that's stayed out of reach, until now.
One model for the entire operation.
We built a World Foundation Model (WFM) for mining: physical AI that learns how rock behaves, how it breaks, floats, leaches, and gives up its metal, and uses that understanding to run the operation, not just describe it.
It senses the whole site, holds a working replica of it in simulation, and acts on what it sees. It runs on QUAEL, our engine of quantum-accelerated optimization, physical-AI world models, and AI-designed silicon, the engine behind everything we build.
A world model isn't a map. It's a working replica of your operation that thinks ahead.
It runs as a continuous loop, sense, simulate, act, learn, and tightens every shift.
Sense the whole site
A sensing fabric reads the operation end to end: hyperspectral core scans and assays underground, geophysics across the deposit, online analyzers, froth cameras, particle-size and mass-balance data through the plant, and vision and LiDAR across the pit. AI-designed edge silicon processes it where it’s generated, so the model sees the operation live, not in last month’s report.
Build the twin
Those streams fuse into a living digital twin, one model holding the orebody, the plant, and the equipment as a single physical system. It doesn’t just store what’s known; it learns the underlying physics: the mineralogy, the way metal is locked in the rock, the kinetics of how it’s freed.
Simulate before you touch the plant
Inside the twin, the model runs thousands of what-ifs, a finer grind, a different reagent dose, a new blend of feed, a longer flotation residence time, and watches recovery, grade, throughput, energy, and water respond. QUAEL’s quantum-accelerated optimization searches a space too large for trial and error and returns the move that pays.
Act in closed loop
The winning settings flow back to the operation, reagent dosing, grind targets, blend ratios, equipment routing, under autonomous control or with your operators in the loop. The model doesn’t hand over a recommendation and walk away; it holds the set-point and adjusts as the ore changes.
Learn and carry it forward
Every shift’s real result feeds back and sharpens the model. And because it’s a foundation model, what it learns on one orebody transfers to the next, so each operation starts smarter than the last.
The value lives in the couplings between stages, the blend that sets your recovery, the schedule that sets your stockpile. One model spans the whole chain, and recovery (the amber stages) is where the dollars concentrate, so it's where the model earns its place first.
One world model, learned as three connected layers.
The geoscience model
Turns sparse drilling and geophysics into a dense, probabilistic picture of the orebody, not only where the metal sits, but how it’s bound: grain size, hardness, mineralogy, the deleterious elements that wreck recovery. It predicts how each block will behave before it’s ever mined.
The recovery model
The metallurgical brain. It treats comminution and separation as live physics, ties the mineralogy of the incoming feed to plant set-points in real time, and finds the recovery that fixed recipes leave on the table, geometallurgy, automated.
The equipment model
The embodied layer: perception, prediction, and control for haulage, drilling, and the plant, the autonomous site that executes what the other two decide.
A predictor scores the present. A world model changes the outcome.
It simulates futures and tests them. It doesn’t just score the present.
Trained on the physics of recovery, fine-tuned per orebody, so a new operation cold-starts in weeks, not years.
Every tonne it processes is new training data; the advantage grows with use and lives in operating data no static dataset can hold.
QUAEL handles the combinatorial search across blends, set-points, and schedules that classical optimization can’t finish in time.
Sense, decide, act, and learn in one system. The intelligence is in the operation, not beside it.
Every layer is intelligent.
Not one model bolted onto old software, a full stack of AI and advanced compute, each piece earning its place in recovery. It all runs on QUAEL, the shared engine, so every advance compounds across the company.
World Foundation Model
A physical-AI base model that learns the physics of rock and process, the simulator the whole operation runs on.
Physics-informed ML
Hybrid physics + machine learning, so the model generalizes from little data and never breaks the laws it’s modeling.
Reinforcement learning
Control policies that hold recovery at the set-point as ore grade, hardness, and feed drift through the day.
Computer vision & sensor fusion
Froth cameras, hyperspectral core scans, LiDAR, and online analyzers fused into one live read of the plant.
Quantum-accelerated optimization
QFabric searches the combinatorial space of blends, set-points, and schedules classical solvers can’t finish in time.
Quantum chemistry
Simulates separation chemistry classical computers can’t, designing the extractants behind rare-earth and lithium recovery.
AI-designed silicon
QFoundry produces custom edge chips that run the model where the data is made, milliseconds on-site, not a cloud round-trip.
Digital twin & synthetic data
A physics-accurate twin generates the scenarios that train the model before it ever touches a live plant, sim-to-real.
Post-quantum security
Operational and geological data migrated to quantum-resistant cryptography, before the mandates land.
We start where the metal is already lost.
No new mine. No decade of permitting. We deploy the model on what you already own (your plant, your stockpiles, your tailings) and prove the uplift in metal, not slideware. Recovery first. Tailings next. Autonomy across the pit as the model earns its place.
We fund the build. We share the upside.
We're not here to buy your mine. We bring the intelligence, the modern processing, and the capital to install it, and we're paid out of the extra metal we recover. A royalty on results, not a bill for software.
Our partners keep their ground and their expertise; we make both worth more. Each side does what it does best, and shares in what they build together.
The operations we touch are proof sites and the seed of a cross-site data flywheel, not the balance sheet. Value accrues to the model and the royalty, not to any single asset.
Where the value compounds
Every plant the model runs makes it better and the next plant easier to win. The proprietary cross-site recovery data, not any one mine, is the moat. The data, and the mathematics on top of it, are ours.
Critical minerals and metals. The model travels.
Because it learns the physics of recovery, the model isn't bound to one metal or one orebody. Every deposit it works teaches it, and every lesson compounds into the next.
Already at work on proven ground.
Our first operation is live where it matters, proven deposits, modern processing, and the recovery model already learning the rock. What it earns on one orebody, it carries to the next. We say plainly what's deployed, what's in pilot, and what's still research.
A science company that happens to mine.
Metallurgists, geoscientists, process engineers, and world-model researchers working the same problem, AI and human expertise, not one replacing the other. We don't sideline the people who know the ground. We give them a model that learns as fast as they do.
Your ground is worth more than it's recovering.
Show us a plant, a stockpile, or a tailings facility. We'll size the metal you're leaving behind, and fund the build to get it back.
Forward-looking statements. This page describes the Critical Resources Mining platform under development within Moonshot Rodeo, including technology and capabilities at varying stages of development. Statements regarding future deployments, capabilities, and outcomes are forward-looking and not guarantees; actual results may differ materially. Nothing herein is an offer to sell or a solicitation to buy any security.