The AI-native neutron company

The software layer between
every neutron and its target

Every nuclear facility wastes >95% of its neutrons at the wrong energy. We built the AI platform that computationally designs optimal neutron spectra — multiplying isotope output from existing infrastructure by 10–50×. No new reactors. No new sources. Just physics, done right.

10–50×
Yield from the same source
$54.6B
Radiopharmaceutical TAM by 2040
0
Competitors doing this
Design space no human can search
Spectrum Design AI Monte Carlo Transport Neural Surrogates Bayesian Optimization Additive Manufacturing Spectrum Shaping OpenMC GPU Clusters
Spectrum Design AI Monte Carlo Transport Neural Surrogates Bayesian Optimization Additive Manufacturing Spectrum Shaping OpenMC GPU Clusters
Spectrum Design AI Monte Carlo Transport Neural Surrogates Bayesian Optimization Additive Manufacturing Spectrum Shaping OpenMC GPU Clusters
The Problem

The nuclear industry has a brute-force problem

Need more isotopes? Buy a bigger reactor. More beam current. Longer irradiation. Every facility on Earth solves the same problem the same way: maximize total flux and hope enough neutrons land at the right energy.

Most don't. >95% of neutrons arrive at energies where the target nucleus barely interacts. Billions in infrastructure, wasted on thermodynamic noise. Nobody is computationally designing the spectrum itself. We are.

R = ∫ N · φ(E) × σ(E) dE
NTarget atoms — fixed by your material choice
σ(E)Cross-section — fixed by quantum mechanics. Not negotiable.
φ(E)Neutron energy spectrum — the only variable. The one we design with AI.
Cross-section resonances are narrow quantum windows where capture probability spikes by 10,000× or more. A neutron at the right energy is worth ten thousand at the wrong one. Our AI learns to concentrate flux exactly at these peaks — across millions of candidate geometries, simultaneously.

Indium-115

30,000 barns at 1.46 eV

vs. <1 barn at 1 MeV. A 30,000× contrast ratio — the AI finds geometries that concentrate flux exactly here.

Uranium-238

Dozens of resonance peaks

Dense resonance structure between 6 eV and 10 keV. Multi-objective optimization across competing resonances.

Molybdenum-98

12, 39, 95 eV clusters

Diagnostic backbone. 40,000+ US procedures/day depend on Mo-99 — with zero domestic production.

The AI Engine

The software is the product. The geometry is the output.

We run millions of Monte Carlo neutron transport simulations on GPU clusters, train neural surrogates on the results, then use Bayesian and evolutionary optimization to converge on designs no human could find. Every new isotope, every new source, every new customer makes the model smarter. The data compounds.

Simulate

Millions of OpenMC neutron transport runs across massive parameter spaces. Brute-force the physics.

Learn

Neural surrogates compress weeks of simulation into millisecond inference. The model gets smarter with every campaign.

Optimize

Bayesian & evolutionary search navigates a combinatorial design space no human can visualize. Hours, not months.

Fabricate

AI-optimized geometries manufactured via metal & ceramic 3D printing. Ship to any neutron source on Earth.

0
Simulations processed
0%
Surrogate convergence
1.0×
Flux improvement vs. baseline
Nuclear engineers who run Monte Carlo codes don't train neural networks. ML researchers who build surrogate models don't know what a barn is. We are the team that works in both worlds. That interdisciplinary gap is not a hiring problem — it's a structural moat.

5 materials. 50 possible layers. Arbitrary thicknesses. Continuous geometry. That's a design space larger than the number of atoms in the universe. Our AI searches it.

How It Works

From target isotope to optimized geometry — in five steps

Where neutrons actually land
Wrong energy — 95.2%
Useful — 4.8%
Industry response: scale everything
Bigger reactor
$1B+
More beam current
10× cost
Same waste ratio
95%

Every facility wastes >95% of its neutrons

Reactors, accelerators, and neutron generators all produce neutrons across a broad energy spectrum. But each target isotope only reacts at specific, narrow energy windows. The rest of the neutrons? Absorbed, scattered, lost. Completely wasted.

The industry's answer has always been brute force: build a bigger source, run it longer, spend more money. The waste ratio stays the same.

40,000+daily US procedures $0spent optimizing spectra
Cross-section: Mo-98 → Mo-99
12 eV 120 barn 39 eV 95 barn 95 eV 68 barn conventional φ(E) Neutron energy (eV) → σ (barn)
The sharp peaks are quantum-mechanical — fixed by nature. The dashed line is where conventional reactors put most neutrons. Almost zero overlap.

Every isotope has a resonance fingerprint

Quantum mechanics dictates that each target nucleus absorbs neutrons only at very specific energies — called cross-section resonances. At these peaks, capture probability spikes by 10,000× or more. A single neutron at the right energy is worth ten thousand at the wrong one.

Conventional sources produce a broad, unoptimized spectrum. The peaks and the flux barely overlap. That's the fundamental waste.

10,000×peak-to-baseline ratio <5%spectral overlap (baseline)
AI design campaign
Target
Mo-99 at 12, 39, 95 eV
Monte Carlo
10M+ neutron transport sims
Neural Surrogate
geometry → spectrum in 2ms
Optimizer
Bayesian + evolutionary search

Brute-force the physics, then learn from it

We run millions of Monte Carlo neutron transport simulations on GPU clusters — each one a different combination of materials, thicknesses, and ordering. The neural surrogate learns the nonlinear mapping from geometry to output spectrum. What took weeks of simulation now takes milliseconds of inference.

Then Bayesian and evolutionary optimizers search the vast combinatorial space — materials × thicknesses × ordering × continuous geometry — converging on designs no human engineer could find by hand.

10M+simulations per campaign 2mssurrogate inference
AI-optimized assembly for Mo-99
SOURCE
Iron14cmfast filter
D₂O30cmmoderator
PE10cmthermal
Bi3cmγ filter
LiF3cm
TARGET
Each material filters, moderates, or absorbs neutrons at different energies. The ordering and thicknesses are what the AI optimizes — the combinatorial space is astronomical.

Not a cylinder. An AI-designed geometry.

The optimizer converges on an assembly of materials — specific compounds at specific thicknesses in a specific order. Iron strips away fast neutrons. Heavy water moderates them down. Polyethylene thermalizes. Bismuth filters gammas. LiF absorbs the thermal tail.

The output geometry is complex and asymmetric — nothing a human would design by hand. Metal and ceramic additive manufacturing makes it buildable. We ship the physical geometry to any neutron source on Earth.

5material classes thickness combinations
Before & after: Mo-99 production
Baseline yield
23×
AI-optimized yield
4.8%
Baseline overlap
61%
Optimized overlap
conventional AI-shaped flux
Gold peaks now align with the blue cross-section resonances. Same source, same neutrons — 23× more reactions.

Same source. Same neutrons. 23× more isotope.

The AI-designed geometry reshapes the neutron energy spectrum to overlap with the target's cross-section peaks. No new reactor. No bigger beam. Just a physical insert — manufactured via 3D printing and shipped to the facility.

This is what it means to computationally design a neutron spectrum. The improvement isn't incremental. It's an order of magnitude — from the same infrastructure, at a fraction of the cost of scaling.

23×yield improvement 61%spectral overlap $0new infrastructure
1 / 5

People are dying on waitlists because we can't make isotopes fast enough. The bottleneck isn't reactors — it's physics.

The Market

$15B in pharma M&A. Zero companies designing the spectrum.

$12B+
Radiopharmaceutical market today
20%
CAGR — fastest-growing drug class
$15B+
Pharma M&A in radioisotopes (2 yrs)
40K+
US procedures/day — zero domestic production

Pluvicto broke the dam

Novartis's Lu-177 radioligand therapy hit $1.39B revenue in 2024 (+42% YoY), targeting $5B peak. First blockbuster radiopharmaceutical ever. Now every major pharma company is racing to build a radioisotope pipeline — and they all need the same scarce isotopes.

Every FDA approval multiplies demand. Every indication expansion multiplies patients. New reactors take a decade and $1B+. Spectrum shaping multiplies the output of what already exists — in weeks, not years.

Bristol-Myers Squibb
$4.1B
RayzeBio — Ac-225 platform
AstraZeneca
$2.4B
Fusion Pharmaceuticals
Eli Lilly
$2.5B
Point Biopharma + Aktis
Novartis
$4.7B+
Mariana + PeptiDream + others
Sanofi
$356M
RadioMedix licensing
Total Firepower
$1.3T
Top 25 pharma (all-time high)
Lu-177

Lutetium-177

$3.4B → $14.7B (2034)

Powers Pluvicto and Lutathera. 20% CAGR. Every pharma company wants in. Production already strained.

Ac-225

Actinium-225

Global supply: ~60 GBq/yr. Demand: limitless.

Alpha emitter, 1000× more potent than Lu-177. Market capped at $75M only because supply caps demand. The biggest bottleneck in medicine.

Mo-99 / Tc-99m

Molybdenum-99

80% of all nuclear medicine. Zero US production.

40,000 daily US procedures depend on a supply chain that routes through 5 aging reactors, none in America. 2024 shortage canceled procedures globally.

Platform, Not Product

We sit between every neutron source and every application

The spectrum is the product. The target changes — the platform doesn't. Every new isotope, every new source type, every new customer adds training data to the surrogate models. The moat compounds.

Medical Isotopes

Mo-99, Lu-177, Ac-225, Tb-161 — each with unique resonance fingerprints. One platform designs optimal spectra for all of them. Higher specific activity, less waste, fewer processing steps, from the same source.

Fusion Tritium Breeding

Every D-T fusion reactor must breed its own tritium. Li-6 captures at 940 barns thermal. Li-7 needs fast neutrons above threshold. Same problem, same AI — different target nucleus.

Radiation Effects Testing

Space and defense electronics qualified under specific neutron spectra. AI-shaped beams test exact failure modes with precision no manual design achieves. DOD and NASA need this yesterday.

Waste Transmutation

Long-lived waste isotopes have specific capture resonances. Spectrum-matched irradiation turns the economics of accelerator-driven transmutation from theoretical to viable.

Why Now

This was impossible 5 years ago. Today it's inevitable.

Compute

GPU Cost Collapse

Training surrogate models on millions of Monte Carlo sims was science fiction in 2020. OpenMC went open-source. GPU clusters dropped campaign costs by 100×. The physics was always there — the compute finally caught up.

Sources

Commercial Neutron Market

SHINE ($779M raised, revenue since 2023), Adelphi, Phoenix — a commercial neutron market now exists. They sell neutrons. We make each neutron 20× more productive. Perfect symbiosis.

Fabrication

3D Printing Unlocks the Geometry

AI-optimized assemblies are complex, asymmetric, multi-material — nothing like the stacked cylinders humans design by hand. Metal and ceramic additive manufacturing makes them buildable for the first time.

Demand

Therapeutic Isotope Crisis

Every radioligand approval multiplies isotope demand. Pluvicto alone expanded to 3 indications. 8 more drugs in pipeline. New reactors take a decade. The supply crisis is here now.

We don't build reactors.
We make every reactor 10× more productive.

The company that owns the AI-optimized mapping from target isotope to optimal geometry sits between every neutron source and every application on Earth. That position compounds — with every new isotope, every new source, every new customer, and every simulation that makes the model smarter.