Physics-informed models and learned surrogate models estimate expensive simulations or experiments while retaining known conservation laws and boundary conditions. Generative models search structural families rather than creating decorative forms; active-learning agents select experiments that reduce uncertainty in valuable regions; vision systems inspect fabricated samples and connect defects back to process conditions. Models must express where extrapolation begins and defer to physical tests whenever confidence is weak. Scientists define objectives, interpret anomalous behaviour and decide which candidates deserve scale-up. AI accelerates the cycle between hypothesis and evidence, but no generated structure becomes a material claim until it has been fabricated, measured and reproduced.
New Matter
“What materials become possible when algorithms join the design process?”
The beginning
New Matter is a computational materials company focused on substances and structures whose performance emerges from geometry as much as chemistry. Nature repeatedly achieves strength, flexibility, porosity and repair through organisation across scales; industrial materials are still often discovered through slow cycles separated between simulation, formulation, fabrication and testing. I imagine a closed learning system in which algorithms, scientists and machines search together for manufacturable structures with lower material and environmental cost. The ambition is not to ask a model for magical molecules. It is to create a disciplined route from desired behaviour to geometry, composition, process and physical evidence, allowing previously unreachable trade-offs to become legitimate design territory.
Materials discovery is slow, expensive and divided across incompatible representations. A candidate may perform beautifully in simulation yet be impossible to manufacture, unstable outside laboratory conditions or dependent on an unacceptable supply chain. Experimental data is sparse and costly, failed trials are poorly preserved, and environmental consequences are evaluated after performance has already shaped the design. Generative systems can produce enormous candidate spaces faster than laboratories can test them, making selection rather than generation the central problem. Without uncertainty-aware models and tight experimental loops, computational discovery risks becoming a factory for plausible structures that never survive contact with production, economics or the full life cycle of matter.
What it could become
The product would be a closed-loop discovery platform connecting desired functions, generative geometry, physics simulation, robotic fabrication and experimental measurement. A team specifies performance ranges and non-negotiable constraints; the system proposes candidate microstructures or formulations, predicts behaviour and chooses the next experiment based on both expected value and information gained. Results—including failures—return to a shared material graph linking composition, topology, process, defects and measured properties. Manufacturing constraints enter before optimisation, so the search concentrates on candidates that can be produced through available methods. The first version would target one bounded domain where structure creates measurable advantage, then expand only after demonstrating repeatable simulation-to-test learning.
For whom
- Advanced manufacturing companies
- Packaging and construction innovators
- Mobility and energy systems
- Materials research laboratories
Core capabilities
- Generative microstructure design
- Physics-informed surrogate models
- Automated experiment planning
- Manufacturing-aware optimisation
The project becomes meaningful only when a new technical possibility is translated into a clear human advantage, an experience people can understand, and a system capable of earning trust over time.
Intelligence and mathematics
Topology optimisation and computational geometry define structures whose macroscopic behaviour follows from internal organisation. Finite-element methods approximate stress, heat, flow or acoustic response; Gaussian processes and Bayesian optimisation navigate expensive experimental spaces; and dimensional analysis keeps learned relationships physically coherent. Multi-objective Pareto fronts expose trade-offs among strength, weight, carbon, cost, recyclability and manufacturability. Uncertainty propagation distinguishes model confidence from tolerance to manufacturing variation, while optimal experimental design selects tests that separate rival mechanisms. At larger scales, life-cycle models prevent a locally efficient material from hiding extraction, toxicity or end-of-life costs elsewhere in the system.
How it might live
New Matter would begin as a deep-technology company around one high-value application where current materials impose a costly constraint and a structural improvement can be validated quickly. Revenue could combine paid discovery partnerships, licensing of proprietary materials and processes, and later access to the discovery platform for industrial R&D teams. Intellectual property may exist in composition, microstructure, manufacturing method and the data connecting them, but experimental credibility is more defensible than a large library of unmade designs. Scaling would require close partnerships with manufacturers and test laboratories from the beginning. The company should earn value by reducing material, energy and failure—not simply by accelerating the production of candidates.
For me, a venture is more than an interesting technology. It needs a narrow first user, a repeated problem, a distribution path, a credible advantage and a reason to improve as more people use it. I would test those conditions before deciding whether this idea should become a company, a product, an open technology or an ongoing research programme.
Rules for making it real
- 01
Manufacturability enters the model on day one.
- 02
Physical evidence outranks simulated promise.
- 03
Environmental impact is an objective, not a report.
- 04
Unexpected failure is valuable data.
From question to company
- 01Frame
Select one high-value material domain.
- 02Prototype
Build simulation-to-test data infrastructure.
- 03Prove
Demonstrate a closed-loop discovery cycle.
- 04Build
Scale a first material with an industrial partner.
What could go wrong
Serious imagination includes the possibility that an idea should change radically—or should not exist. These are the tensions the project would need to resolve:
- Simulation reality gaps.
- Laboratory success failing at production scale.
- Optimising one environmental measure while worsening another.