Quantum Machine Learning Targets Rare-Earth Separation

A new U.S.-French partnership is putting quantum computing into one of the least glamorous—and most consequential—parts of the critical-minerals supply chain: separating rare-earth elements from one another.

USA Rare Earth, French neutral-atom quantum company Pasqal, and AI chemistry firm Riven Systems announced the collaboration on September 17, 2026. The project will use Riven’s automated laboratory to run thousands of experiments on chemical extractants, then use the resulting data to train machine-learning models. Pasqal’s quantum processing unit will benchmark quantum machine-learning approaches against classical models to help identify molecules that bind more selectively to individual rare-earth elements. [1][2]

What changed

The announcement is not a claim that a quantum computer has already improved industrial separation. It is a plan to build and test a discovery pipeline around a specific feedstock and processing problem. USA Rare Earth expects the work to include mixed rare-earth carbonate from its Round Top project in Texas, along with other materials and recycled magnet-manufacturing waste. Promising candidates would ultimately be validated at the company’s research facility in Colorado. [1]

That distinction matters. Rare-earth separation is chemically difficult because the elements have closely related properties. Processing mixed material into individual oxides requires repeated stages, substantial equipment, and significant energy. China’s established dominance is concentrated not only in mining, but also in these downstream refining and separation capabilities. The partnership is therefore aimed at a chokepoint where better chemistry could have more practical value than another headline about raw tonnage.

Why it matters

The industrial logic is a hybrid one: autonomous experimentation supplies real-world chemical data; machine learning narrows the search; quantum hardware tests whether it can represent or optimize parts of the problem more effectively than conventional approaches. Mining Weekly described the same effort as an attempt to optimize processing through the combination of Pasqal’s quantum systems, Riven’s automation, and USA Rare Earth’s materials expertise. [2]

If the approach works, the payoff would not necessarily be a dramatic quantum speedup. A more plausible near-term benefit would be finding a useful extractant with fewer experiments, then designing smaller separation facilities that consume less energy and material. That could improve the economics of Western rare-earth processing while reducing environmental burdens.

The strategic stakes extend beyond one company. Rare earths such as dysprosium, terbium, and yttrium are important to high-performance magnets and other advanced technologies. More efficient separation could make non-Chinese supply chains more competitive, but only if laboratory results survive scale-up, impurities, recycling variability, and continuous industrial operation.

What remains uncertain

The partnership has disclosed no performance results, timeline for a validated molecule, financing terms, or evidence that the quantum models outperform classical alternatives. It also uses forward-looking language: the companies say the project aims to discover better extractants and could eventually reduce facility size, cost, and energy use. Those outcomes remain hypotheses, not demonstrated gains. [1]

The first meaningful test will be comparative: whether Pasqal’s models produce better candidates than classical methods trained on the same experimental data. Until that benchmark and a pilot-scale process are public, the announcement is best read as a serious application experiment—not proof that quantum computing has solved rare-earth refining.