Runner-up · AI · Critical-mineral processing
Rare-earth separation gets an autonomous chemistry lab.
USA Rare Earth, Riven Systems and Pasqal will combine autonomous experiments with classical and quantum machine learning to search for better rare-earth separation molecules, then validate promising candidates at USA Rare Earth’s Wheat Ridge research facility.
The planned workflow is closed-loop: models propose extractant candidates, Riven’s self-driving laboratory runs thousands of automated experiments, and the resulting data trains selectivity models. Pasqal’s neutral-atom system will be benchmarked against classical machine learning rather than assumed to be superior.
The program will use feedstocks from the Round Top deposit, third-party mixed rare-earth carbonates and recycled magnet swarf. The partners have not yet disclosed a validated molecule, separation yield, energy reduction, pilot throughput or commercialization timetable.
Read original story ↗01
What changed
Rare-earth separation optimization has typically relied on slow, sequential laboratory work. What changed on September 17 is a project that connects model proposals, thousands of automated physical experiments and process validation on operating feedstocks in one industrial learning loop.
02
Why it matters
Separation is one of the hardest and most energy-intensive stages in a domestic rare-earth supply chain. A self-driving laboratory can search more molecular and process combinations than a manually sequenced program, while real feedstocks keep the work tied to plant chemistry rather than a benchmark. The architecture is also a useful template for industrial AI: automate the experiment, capture structured results and let the model propose the next test. The quantum component deserves restraint; it will be benchmarked against classical methods and may add no advantage. Commercial value begins only when a candidate improves selectivity, cost, energy use or plant footprint in a validated flowsheet.
03
What to watch
Watch for a disclosed selectivity gain, recovery rate, energy reduction, cycle-time improvement and a validated flowsheet at Wheat Ridge. A pilot using Round Top or recycled feedstock would be stronger evidence than another modeling result.
Why it was a runner-up
The autonomous experimentation loop is industrially specific, but the project has not yet produced a validated molecule, commercial flowsheet or plant-scale result.
Impact: 75/100 · Confidence: 84/100
The partners specify the experimental workflow, feedstocks and validation site, and independent reporting confirms the collaboration. Confidence is moderated because the technical outcomes are forward-looking and no experiment result is yet public.
Read this edition’s Daily Signal →