USA Rare Earth Enlists Quantum Chemistry to Shrink Separation Plants

USA Rare Earth announced a strategic partnership with quantum computing firm Pasqal and AI specialist Raven Systems to use quantum machine learning to design new rare earth separation molecules, and its Nasdaq-listed shares rose 3.51% to 15.63 as of 20:00 GMT on 17 September 2026.

USA Rare Earth (NASDAQ: USAR) has recruited two computing partners to attack the least glamorous and most stubborn problem in the rare earth business: pulling seventeen chemically near-identical elements apart from one another without building an enormous, power-hungry plant to do it.

The Nasdaq-listed miner said it has formed a strategic partnership with Pasqal, a quantum computing company, and Raven Systems, an artificial intelligence specialist, to develop next-generation separation technology for the rare earth value chain. The stated aim is to use quantum machine learning — machine learning models run on or informed by quantum hardware — to identify separation molecules that bind more selectively to individual rare earth elements than the chemistry available today. Better-binding molecules would, in theory, allow processing facilities that are smaller, cheaper to build and less energy-intensive to run.

Investors liked it. USAR traded at 15.63 as of 20:00 GMT on 17 September 2026, up 3.51% on the day from a previous close of 15.10, with an intraday range of 15.39 to 15.94, according to live market data. That outpaced a broadly strong tape: the S&P 500 proxy SPY was up 1.13% at $762.60, the Nasdaq 100 proxy QQQ up 1.73% at $716.92 and the Dow 30 proxy DIA up 0.61% at $518.35.

Why separation, not mining, is the bottleneck

Rare earths are not especially rare in the ground. What is rare is the ability to separate them economically. Neodymium, praseodymium, dysprosium and terbium — the elements that matter most for permanent magnets in motors, wind turbines and defence systems — sit next to each other on the periodic table and behave almost identically in solution. The incumbent industrial answer is solvent extraction: hundreds of mixer-settler stages in series, each nudging the mix a fraction closer to purity, consuming reagents, floor space and power along the way.

That is why the chokepoint in the Western rare earth supply chain has been midstream chemistry rather than mine supply. A separation molecule with materially higher selectivity would cut the number of stages required. Fewer stages means a smaller building, less pumped volume, less reagent inventory and a lower energy bill — which is precisely the outcome the three partners describe.

What quantum machine learning is actually being asked to do

The task here is molecular design, and it is a good fit for the argument quantum computing companies have been making. Predicting how strongly a candidate ligand will bind to a specific rare earth ion means modelling electron behaviour — the kind of quantum-mechanical problem that classical computers approximate rather than solve. Screening a large chemical space for the handful of structures worth synthesising is the expensive part; a model that ranks candidates more accurately narrows the laboratory work that follows.

Raven Systems' role as the AI specialist implies the classical half of that loop: training on existing binding data, generating candidate structures, and feeding the most promising into the quantum workflow Pasqal supplies. As reported by Mining Weekly, the partnership is framed around identifying molecules that bind more effectively to rare earths than existing alternatives — a discovery objective, not a commissioned plant design.

That distinction matters for how the announcement should be read. Nothing disclosed says a molecule has been found, synthesised, or tested at pilot scale.

The timeline problem between a screen and a flowsheet

Chemical discovery announcements compress badly into share prices. Between a computationally identified ligand and a working commercial circuit sit several gates that no amount of compute shortens much:

• Synthesis. A molecule that models well has to be makeable at a cost that does not swamp the savings from a smaller plant.
• Stability. Extractants degrade in acid, heat and repeated loading-stripping cycles. Lifetime in a real circuit is an empirical question.
• Pilot validation. Selectivity measured in a beaker has to survive continuous operation with real feedstock, impurities included.
• Engineering. Flowsheets, permits and capital approvals run on their own clock regardless of how clever the chemistry is.

Investors weighing the move in USAR should treat this as an option on a lower-cost processing route rather than a change to any facility currently being built. Optionality is worth something — particularly in a sector where the cost of Western separation capacity is the whole argument — but it is not near-term throughput.

How it fits the wider push to rebuild Western midstream capacity

The strategic logic is unambiguous even if the science is early. Rare earth separation capacity outside China is thin, and the projects trying to add it face the same arithmetic: high capital cost per tonne of separated oxide, high energy intensity, and product prices set in a market they do not control. Anything that lowers the capital hurdle changes which projects clear it.

It also fits a broader pattern of miners and processors buying computational capability rather than building it. Materials discovery has become one of the more credible commercial applications pitched for both AI and quantum hardware, because the search spaces are vast, the physics is genuinely quantum, and the payoff from a single better molecule is concentrated. For Pasqal and Raven Systems, an industrial partner with a real separation problem and a real feedstock pipeline is a more persuasive proving ground than a benchmark.

What to watch from here

Three things will tell you whether this is more than a press release. First, whether the partners disclose a candidate molecule or a validated selectivity result against a named incumbent extractant. Second, whether any result migrates from computational screening into bench and then pilot testing, with timelines attached. Third, whether USA Rare Earth's own processing plans are subsequently re-scoped — a smaller footprint or lower capital estimate for a facility would be the first hard evidence that the chemistry is influencing engineering decisions rather than running alongside them.

Until then, the day's 3.51% gain is the market paying for a credible idea in a sector short of them, on a session when risk appetite was broadly strong anyway. The stock finished the day's range well off its 15.94 high, which is its own small piece of commentary: enthusiasm, tempered.

Key facts

• USAR share price: 15.63, +3.51% as of 20:00 GMT, 17 Sep 2026
• Partners: Pasqal (quantum computing) and Raven Systems (AI)
• Technology: Quantum machine learning to identify rare earth separation molecules
• Stated goal: Smaller, lower-cost, less energy-intensive processing facilities

Frequently asked questions

What did USA Rare Earth announce?

USA Rare Earth announced a strategic partnership with quantum computing company Pasqal and artificial intelligence specialist Raven Systems to develop next-generation separation technology for the rare earth value chain. The collaboration will apply quantum machine learning to identify separation molecules that bind more effectively to rare earth elements than existing alternatives, with the aim of enabling smaller and less energy-intensive processing plants.

How did USA Rare Earth shares react?

USAR traded at 15.63 as of 20:00 GMT on 17 September 2026, a gain of 3.51% from the previous close of 15.10. The intraday range was 15.39 to 15.94, meaning the stock closed the session well below its high. The broader market was also firm, with the Nasdaq 100 proxy QQQ up 1.73% on the day.

Why is rare earth separation so difficult?

Rare earth elements are chemically very similar to one another, so they behave almost identically in solution. Industrial separation typically relies on solvent extraction, using hundreds of sequential stages to gradually purify each element. That process demands large facilities, substantial reagent volumes and significant energy, which is why separation rather than mining is the main bottleneck in Western supply chains.

What can quantum computing contribute to this problem?

Predicting how strongly a candidate molecule binds to a specific rare earth ion is a quantum-mechanical problem that classical computers can only approximate. Quantum machine learning is intended to rank candidate molecules more accurately, narrowing the field of structures worth synthesising in a laboratory. The partnership's goal is discovery of better-binding separation molecules, not the design of a specific plant.

Does this change USA Rare Earth's existing processing plans?

Nothing in the announcement indicates that a molecule has been identified, synthesised or tested at pilot scale. Any resulting chemistry would still need to pass synthesis cost, stability in acidic circuits, continuous pilot validation with real feedstock and engineering approval before affecting a plant design. The partnership is best read as an option on lower-cost processing rather than a near-term change.

Who are Pasqal and Raven Systems?

Pasqal is a quantum computing company and Raven Systems is an artificial intelligence specialist. Both are named in the announcement as strategic partners to USA Rare Earth. Raven Systems' AI capability would typically handle the classical side of candidate generation and screening, while Pasqal supplies quantum computing capacity for the molecular modelling workload.

Sources

• USA Rare Earth announces AI, quantum computing partnership for optimised processing — Mining Weekly

Photo: cottonbro studio · Pexels Licence — source