Lattice Graph × MP Materials

Lattice Graph operates a computational materials-discovery platform built around a knowledge graph spanning millions of compositions, connected to property data, synthesis recipes, peer-reviewed evidence, and a patent corpus covering more than 300,000 materials filings. When a candidate separation chemistry, recovery sorbent, or process route enters the pipeline, it does not pass through a single model: Lattice Graph runs multi-engine validation using machine-learning interatomic potentials including MACE and CHGNet alongside density functional theory, requiring phonon stability and thermodynamic consensus across independent physics engines before a composition advances. This cross-validation discipline substantially reduces the rate at which theoretically attractive candidates fail at bench scale, which matters most in separation chemistry where ligand geometry, pH selectivity windows, and competing-ion rejection all depend on structural details that single-model approaches routinely misrepresent. For a company operating in rare-earth separation and recycling, the most decision-relevant layer of the platform is what most computational tools omit entirely: a large, governed atlas of labeled negative results. Most machine-learning models for materials properties are trained on published successes; Lattice Graph holds an internal corpus of failed experiments, synthesis dead ends, and separation routes that looked promising on paper and did not perform. For a potential licensor or technology evaluator, that negatives moat converts a candidate chemistry from a theoretical prediction into a de-risked position informed by what has already been shown not to work, before capital is committed to a pilot circuit or a licensing agreement. The platform also runs composition- and claim-level freedom-to-operate screening across the full 300,000-patent corpus at query time, so that every asset in the discovery portfolio carries an IP posture that has been stress-tested against the actual claims landscape, not against a keyword search. Taken together, multi-engine physics validation, provenance-traced knowledge graph, and systematic patent screening give MP Materials a path from its stated separation and recycling ambitions to licensable, computationally validated chemistry with a clear IP path, faster and with lower technical risk than internal discovery alone.

MP Materials is executing a deliberate and publicly stated integration from bastnaesite concentrate at Mountain Pass through separated rare-earth oxides, NdFeB metal and alloy, finished sintered magnets, and magnet and battery-scrap recycling. The margin at each stage is determined not by mining yield, which Mountain Pass has already demonstrated, but by separation chemistry: the length of the solvent-extraction cascade, the reagent inventory per kilogram of separated oxide, the fraction of dysprosium and terbium recovered from magnet scrap, and the economics of the recycling loop. These are computational materials problems at their core, and they are precisely the lanes where Lattice Graph has built its discovery infrastructure. The strategic urgency is specific. Light rare-earth oxide pricing for neodymium and praseodymium is structurally exposed to Chinese processing dominance, and the heavy rare earths that give NdFeB magnets their thermal stability, dysprosium and terbium in particular, now sit under active and escalating export controls. MP's domestic recovery of Dy and Tb through magnet-scrap recycling is the primary lever available to reduce that exposure, but the separation chemistry that determines recovery yield from a complex leachate is exactly what the critical-mineral recovery and recycling separations portfolio was designed to address. Every improvement in separation factor, reduction in cascade stages, and increase in Dy/Tb recovery from scrap flows directly to MP's unit economics and to the domestic-supply narrative it carries into offtake agreements with OEMs and the U.S. government. And with the U.S. Department of Defense now an anchor investor in MP — a roughly $400 million convertible-preferred position and a neodymium-praseodymium price floor near $110 per kilogram that effectively backstops demand — MP's exposure shifts from price risk to throughput and yield, which makes separation and recycling efficiency the single dominant margin lever rather than one of several. Lattice Graph brings two things that internal R&D and conventional contract research cannot replicate on the same timeline. First, a portfolio of computationally validated, freedom-to-operate-screened separation and recovery chemistries that are ready for technical due diligence and pilot-scale evaluation rather than starting from literature survey. Second, an intelligence layer, supply-risk, deposit-concentration, and patent-whitespace data, that lets MP's own technical and strategy teams pressure-test feedstock assumptions, monitor the competitive IP landscape in real time, and anchor the domestic-supply arguments that support government engagement. The combination of licensable chemistry and working data infrastructure maps directly onto the two parts of MP's integration story that are still open: refining netback and recycling economics.

MP Materials is executing a deliberate and publicly stated integration from bastnaesite concentrate at Mountain Pass through separated rare-earth oxides, NdFeB metal and alloy, finished sintered magnets, and magnet and battery-scrap recycling. The margin at each stage is determined not by mining yield, which Mountain Pass has already demonstrated, but by separation chemistry: the length of the solvent-extraction cascade, the reagent inventory per kilogram of separated oxide, the fraction of dysprosium and terbium recovered from magnet scrap, and the economics of the recycling loop. These are computational materials problems at their core, and they are precisely the lanes where Lattice Graph has built its discovery infrastructure. The strategic urgency is specific. Light rare-earth oxide pricing for neodymium and praseodymium is structurally exposed to Chinese processing dominance, and the heavy rare earths that give NdFeB magnets their thermal stability, dysprosium and terbium in particular, now sit under active and escalating export controls. MP's domestic recovery of Dy and Tb through magnet-scrap recycling is the primary lever available to reduce that exposure, but the separation chemistry that determines recovery yield from a complex leachate is exactly what the critical-mineral recovery and recycling separations portfolio was designed to address. Every improvement in separation factor, reduction in cascade stages, and increase in Dy/Tb recovery from scrap flows directly to MP's unit economics and to the domestic-supply narrative it carries into offtake agreements with OEMs and the U.S. government. And with the U.S. Department of Defense now an anchor investor in MP — a roughly $400 million convertible-preferred position and a neodymium-praseodymium price floor near $110 per kilogram that effectively backstops demand — MP's exposure shifts from price risk to throughput and yield, which makes separation and recycling efficiency the single dominant margin lever rather than one of several. Lattice Graph brings two things that internal R&D and conventional contract research cannot replicate on the same timeline. First, a portfolio of computationally validated, freedom-to-operate-screened separation and recovery chemistries that are ready for technical due diligence and pilot-scale evaluation rather than starting from literature survey. Second, an intelligence layer, supply-risk, deposit-concentration, and patent-whitespace data, that lets MP's own technical and strategy teams pressure-test feedstock assumptions, monitor the competitive IP landscape in real time, and anchor the domestic-supply arguments that support government engagement. The combination of licensable chemistry and working data infrastructure maps directly onto the two parts of MP's integration story that are still open: refining netback and recycling economics.

The critical-mineral recovery and recycling separations portfolio is the primary match for MP Materials, and it covers both sides of the margin equation MP is working to improve. On the refining and co-product side, the portfolio includes tunable sorbent architectures designed to recover specific critical-mineral oxocations from complex industrial process streams, including chemistries with confirmed selectivity in high-competing-ion environments at the low pH levels typical of rare-earth leach circuits. On the recycling side, it includes both sorbent-based and solvent-process routes designed for the mixed leachates that come from magnet scrap and spent battery black mass, with system-level flowsheet claims that cover ordered recovery cascades rather than isolated compositions. For MP, the relevant assets span the magnet-scrap recycling train it is building and the co-product recovery opportunities in its existing separation circuits at Mountain Pass. The solid-state battery electrolytes and interfaces portfolio carries a lower but real relevance as a forward option. As MP considers value capture beyond finished NdFeB magnets, the lithium and sodium battery-material chemistries in that portfolio become strategically interesting, particularly given the convergence of magnet and battery supply chains in automotive electrification. The battery-recycling assets in the critical-mineral portfolio, specifically the chloride-free deep-eutectic-solvent cathode process, also create a bridge between MP's current magnet-scrap recycling line and a potential extension into lithium-ion battery-scrap recovery, which would widen the recycling economics and the volume of domestic critical materials MP can recover without importing from controlled sources. The two portfolios should be read as a near-term and a forward-option layer. The near-term assets, the sorbent and process chemistries in the critical-mineral recovery portfolio, address what MP is building right now. The battery-material portfolio represents where MP could extend its technology and supply-chain position if it pursues the recycling and materials-supply expansion that its public roadmap suggests. Lattice Graph's governance model allows MP to license specific asset families within each portfolio and to take options on others, structuring the commercial relationship around its actual integration sequence rather than a bundled portfolio acquisition.

MP Materials' technical and strategy teams would use the platform most heavily across two interconnected workflow surfaces. The supply and IP workflow pages let a refining or recycling chemist move from element-level supply-risk analysis on dysprosium and terbium through a map of relevant candidate chemistries and into composition-level freedom-to-operate screening without context-switching between tools. For a company where the margin question and the IP question are inseparable from the same separation or recycling decision, having supply risk, chemistry discovery, and patent whitespace in a single governed environment shortens the path from a technical question to a diligence-ready answer by a substantial margin. The composition-360 evidence views and the knowledge-graph explorer give individual researchers a single pane on any candidate sorbent, ligand architecture, or process route: predicted properties alongside the density functional theory and machine-learning potential stability evidence that supports them, the synthesis recipe precedent, and the surrounding patent neighborhood. The natural-language graph query capability lets MP's team interrogate the rare-earth separation and critical-mineral recovery space directly without needing to know the underlying data schema, batch-screening candidate chemistries, comparing agreement and disagreement signals across computational sources, and pulling structured provenance for any claim into a package that can serve as the starting point for a technical due-diligence review. For a team evaluating whether to license, co-develop, or pass on a given asset family, these workflows convert an abstract portfolio into something MP's own process chemists and metallurgists can pressure-test against their actual leach liquors, product slate, and recycling streams.

For an assets-archetype partner, the natural engagement structure opens with a scoped technical evaluation: a freedom-to-operate and patent-whitespace package on the chosen asset families, a knowledge-graph provenance data room covering the computational stability evidence and recipe precedent for each candidate chemistry, and a structured technical discussion with Lattice Graph's discovery team to map those assets against MP's specific process conditions, leach-liquor compositions, and recycling-stream characteristics. That evaluation produces a diligence-ready technical package, not a sales deck, and it gives MP's engineering and legal teams the material they need to make an informed licensing or co-development decision against their own internal criteria. The scoped evaluation is the right first step precisely because the most valuable assets for MP are those whose separation or recovery performance can be verified against its actual Mountain Pass and recycling-circuit conditions, not against generic benchmarks. Following the evaluation, commercial structures available include field-of-use licenses on specific asset families, structured as an upfront payment plus royalties tied to recovered-product volume or netback; co-development programs priced to shared bench and pilot milestones where open proof gates remain (for example, measured separation factors on MP's own magnet-scrap leachate rather than modeled predictions); and acquisition or exclusivity arrangements for families where MP wants to foreclose competitor access in heavy-rare-earth separation and magnet-recycling fields of use. Data and intelligence API access runs alongside any of these structures as a subscription, supporting ongoing feed, IP, and supply-risk diligence as MP scales production. Commercial terms are negotiated to fit the specific asset families and fields of use MP selects rather than as fixed catalog prices.