Progress in materials science is incremental. Experimentation in this space, be it in the lab or in silico, is driven by first-principles computational methods, characterization equipment, synthesis equipment, machine learning models, and theory that collectively represent thousands of lifetimes’ worth of work. Despite all this labor and toil, there is still so much more to build. The advent of the modern GPU, the rapid buildout of compute driven by the LLM boom, and increasingly widespread access to these technologies have sparked a private-sector renaissance in the sciences. Organizations like Periodic Labs, Lila Sciences, Radical AI, and many others have raised substantial amounts of capital on the promise of discovering the next generation of materials. The self-driving lab concept often employed by these organizations promises to close the discovery loop by connecting computational predictions with real-world experiments where the resulting data could help train increasingly capable AI systems for scientific research. Their aim is a system that supports materials discovery as broadly as AI tools now support software development. We think this is great for materials science, but we also believe there is a much more targeted and pragmatic use of capital.

1. We believe wholeheartedly in a deliberate focus on pushing the frontier of computational methods. This is the golden age of software development. Computational methods like DFT, CALPHAD, phase-field modeling, and MD simulations are full of potential optimizations, improvements, and new features just waiting to be built. Alongside these tools, research into machine learning methods for problems that have traditionally been difficult to model, such as solidification, novel preconditioners, crystal nucleation, linear-time scaling solvers, grain growth, and the diffusion of interstitials and vacancies, has never been easier. We can spend more time simulating and modeling materials, developing methods to improve the accuracy of these simulations, and ultimately working toward a future where you can make a material in the lab on the first try.

2. We believe in specialization. We’re solely focused on permanent magnets, and our immediate priorities are advancing the modeling of intermetallic microstructures, finding novel preconditioners to accelerate SCF convergence in magnetic materials, and discovering more promising AFLOW prototypes. Materials science is a multifaceted and interdisciplinary field where you encounter a new set of challenges at every length scale. Equipping labs to work across multiple material families, hiring the specialized talent to operate them or develop software to control them autonomously, and running thousands of experiments across each family can become incredibly expensive. By focusing all of our resources on a narrow problem, we intend to take a measured approach to delivering innovation.

3. We believe carefully scoped, targeted real-world experimentation can be just as powerful as mass autonomous experimental data collection. There is a wealth of borderline wizard-like experimental talent in academia. These folks have accumulated their arcane expertise over decades of working in some of the most advanced labs in the world. Whether through collaborations with these institutions or by bringing that talent in-house, we believe focused experimentation with these folks is the most effective way to improve our simulations at this stage.

Above all else, we embrace the incrementality that has defined progress in this field. There is no “great man.” Overnight successes have always been decades in the making, and we feel that it’s our responsibility to openly contribute at every step of our journey.

Project 14 is dedicated to fully open science.

An optimization, a dataset, a model - whatever it might be, if we built it, everyone will have access to it. Science is about progress, and the future is everyone’s right.