Climate change is a materials problem. We are building the AI to solve it, starting in polymers.

Almost every polymer still runs the linear path: fossil carbon in, product out, waste at the end. At scale, that is straining the climate; producing and burning the world's plastics alone could take a large share of the carbon budget left for 1.5°C by 2050, and plastics are only one polymer family.

RePolymer is building the AI for a different path: models built to predict how a polymer performs and degrades, and inverse design that proposes new ones made to be recovered, starting in polymers and reaching toward the wider materials economy.

Carbon-budget estimate: Zheng & Suh, Nature Climate Change (2019)

AILINEARWASTECIRCULARvalue returns

0.0°C

Warming limit at stake as plastics scale

0%

Share of plastic actually recycled worldwide

Possible polymers to search - the space is effectively unbounded

One line leaks value away. One loops it back. One moves it between domains.

Extracted, made, used, discarded: value and carbon leak out the open end of the linear path most polymers still take. Circular design bends that line into a loop, so the polymer keeps its value. RePolymer pushes further, toward a cross-domain path where one loop's output feeds the next (polymers into biopolymers, agriculture, coatings), one platform running many loops at once.

Linear · value leaks out

waste · CO₂ExtractProduceUseDiscard

Circular · value returns

DesignProduceUseRecoverreturned to the loop

Cross-domain · value moves between domains

BiopolymersCoatingsAgriculturePolymersone platform · many loops
none exact structurechain length / molecular weightnumber of chainslonger line = longer chainMany chains of different lengths, not one exact molecule.

A polymer is not a molecule. It is a distribution.

Small molecules have databases you can search. Polymers do not work that way. A single material is a statistical ensemble: a spread of chain lengths, sequences, and architectures whose real behavior only appears after it is processed. The space of what could be made is effectively unbounded.

So the field advances the only way it can, by trial and error, and a new polymer can still take years to reach the market. The recoverable, lower-carbon candidate rarely wins that race. The bottleneck here is not ambition. It is search.

what you needStrengthCostlowProcessabilityRecoverabilityRePolymermodelwhat you getnRP-114trust0.86nRP-207test0.63nRP-352test0.41Ask for properties, get candidates to trust or test.

Model the whole polymer before it exists.

RePolymer is being built to learn how structure, processing, and property connect from the sparse, hard-won data the field already holds, then run that link in reverse: state the performance you need and get candidate chemistries back, each with the confidence to trust it or test it.

Recoverability is treated as a property, not a cleanup. Feedstock, degradation, and end-of-life route are weighed in the same pass as strength, cost, and processability, so the circular option competes on performance from the first repeat unit.

on the horizonthe horizonnowPOLYMERSCompositesBiopolymersCoatingsAdhesivesPolymers now. The wider materials world next.

Fewer dead ends. Months, not years.

The aim is concrete: fewer dead-end experiments, discovery compressed from years toward months, and formulations where durability and recoverability are chosen together instead of traded blindly. Polymers are the sharpest place to prove it, a vast and high-stakes design space where better search changes what gets made.

Polymer intelligence now, materials intelligence next.

Help shape what comes next.

Researchers, partners, and builders who want polymers off the linear path: this is the work, and we want you in it.