Lattice Graph × Trane Technologies

Lattice Graph is a computational materials-discovery platform built around a knowledge graph spanning millions of compositions, with every candidate material validated by multiple independent machine-learning interatomic potentials — MACE, CHGNet, MatterSim, and ORB — before any DFT computation is committed. Stability is not inferred from a single model; consensus across phonon stability and thermodynamic free-energy calculations is required before a candidate advances. That multi-oracle approach removes the single-model blind spots that have historically sent chemistry teams down dead-end synthesis routes. The result is a dramatically higher hit rate in targeted simulation campaigns and a sharply lower cost per qualified candidate. The platform also carries a large atlas of labeled negative results — failed experiments and compositions that looked promising on paper but did not survive validation — giving every search the benefit of institutional memory that no single lab could accumulate. On top of discovery, Lattice Graph runs composition- and claim-level freedom-to-operate and patent-whitespace screening across more than 300,000 materials patents, so a team can identify defensible IP positions alongside the chemistry. Together, these capabilities compress the front-end of materials R&D from years of trial-and-error into weeks of directed simulation.

Trane Technologies occupies a rare position in the data-center cooling market: it owns the thermal-management hardware and the channel relationships, and through its strategic investment in LiquidStack it has a direct line into two-phase immersion-cooling deployment at scale. The one gap in that stack is the qualified fluid. That is precisely where Lattice Graph enters. 3M's exit from its approximately $12.5 billion PFAS and Novec coolant business at the end of 2025 removed the dominant incumbent fluid supplier from the market at the exact moment demand is accelerating — the immersion-cooling fluid market is moving from roughly $2 billion toward an estimated $11 billion — and left hardware providers like Trane without a drop-in qualified replacement. The PFAS-free dielectric immersion-cooling system is a publicly stated performance spec: verified dielectric retention, a tested corrosion-inhibitor package, multi-hundred-hour reuse life, and no fluorinated chemistry. No shipping product currently meets that specification in full. That is not a product gap Trane can close with formulation intuition and bench iteration alone — the combinatorial space of PFAS-free dielectric candidates, stabilizer packages, and operating-condition interactions is simply too large for conventional R&D timelines. Lattice Graph's computational platform is designed to navigate exactly this kind of high-dimensional candidate space, filtering millions of compositions down to a short list of experimentally testable leads with high confidence in thermodynamic and dielectric stability. The alignment extends beyond fluids. As AI accelerator packages push power densities higher, thermal-interface materials become a co-equal bottleneck to immersion-fluid performance. Trane's data-center cooling ambitions require chemistry that works from the die face out to the bath — a full thermal stack — and Lattice Graph holds validated leads across both layers of that stack, including zone-modulated thermal-interface materials tuned to actual GPU heat-flux maps. The platform can serve Trane's fluid program and its accelerator-packaging work from the same computational knowledge graph.

Trane Technologies occupies a rare position in the data-center cooling market: it owns the thermal-management hardware and the channel relationships, and through its strategic investment in LiquidStack it has a direct line into two-phase immersion-cooling deployment at scale. The one gap in that stack is the qualified fluid. That is precisely where Lattice Graph enters. 3M's exit from its approximately $12.5 billion PFAS and Novec coolant business at the end of 2025 removed the dominant incumbent fluid supplier from the market at the exact moment demand is accelerating — the immersion-cooling fluid market is moving from roughly $2 billion toward an estimated $11 billion — and left hardware providers like Trane without a drop-in qualified replacement. The PFAS-free dielectric immersion-cooling system is a publicly stated performance spec: verified dielectric retention, a tested corrosion-inhibitor package, multi-hundred-hour reuse life, and no fluorinated chemistry. No shipping product currently meets that specification in full. That is not a product gap Trane can close with formulation intuition and bench iteration alone — the combinatorial space of PFAS-free dielectric candidates, stabilizer packages, and operating-condition interactions is simply too large for conventional R&D timelines. Lattice Graph's computational platform is designed to navigate exactly this kind of high-dimensional candidate space, filtering millions of compositions down to a short list of experimentally testable leads with high confidence in thermodynamic and dielectric stability. The alignment extends beyond fluids. As AI accelerator packages push power densities higher, thermal-interface materials become a co-equal bottleneck to immersion-fluid performance. Trane's data-center cooling ambitions require chemistry that works from the die face out to the bath — a full thermal stack — and Lattice Graph holds validated leads across both layers of that stack, including zone-modulated thermal-interface materials tuned to actual GPU heat-flux maps. The platform can serve Trane's fluid program and its accelerator-packaging work from the same computational knowledge graph.

The PFAS-free dielectric and process fluids portfolio was built for the exact market moment Trane now faces. Every asset in it was developed under the constraint that fluorinated chemistry is off the table — not as a regulatory hedge but as a hard engineering requirement — so the dielectric performance, thermal stability, and materials compatibility benchmarks are already set against a PFAS-free baseline. For Trane, the most direct entry point is the closed-loop dielectric immersion-cooling system designed specifically for AI accelerator and data-center environments, paired with the fluid-agnostic purification and sensor-gated release process that can qualify any PFAS-free fluid candidate to electronics-grade purity at production volumes. That combination gives Trane a qualified fluid plus the process infrastructure to supply it consistently, not just a lab sample. The High-power thermal-interface materials portfolio addresses the other half of the thermal stack. Heat-flux-mapped, zone-modulated thermal-interface materials are designed to the actual power-density profiles of current AI accelerators — reducing hotspot temperatures by 10 to 25 degrees Kelvin without adding filler mass — while the integrated package architecture covering TIM-1, TIM-2, and lid-attach layers provides a single ordered thermal path from die to heat-spreader. For Trane, which needs to offer data-center customers a complete cooling solution rather than a point product, these thermal-interface assets extend the company's value proposition from the immersion bath through the chip package itself. Taken together, the two portfolios map cleanly onto Trane's stated strategic objective: own the full thermal-management stack for AI data centers.

Trane's materials scientists and process engineers can access the Lattice Graph platform through a web application designed around the workflows of a chemistry team running parallel discovery programs. Candidate materials surfaces can be explored interactively across composition space, with stability consensus scores from the full ensemble of machine-learning potentials displayed alongside DFT-validated results and experimental flags from the negative-result atlas. The patent-whitespace screening interface allows an engineer to enter a candidate composition or formulation class and receive a claim-level map of the existing patent landscape, identifying both crowded regions and open whitespace without requiring a formal IP engagement. For Trane's ongoing fluid qualification and thermal-interface development work, the platform's targeted simulation workflow allows a team to specify operating conditions — temperature range, voltage, substrate materials, heat-flux profile — and receive a ranked shortlist of candidates that have cleared multi-oracle stability validation under those constraints. Results are exportable into standard formats compatible with lab information management systems, and the API access layer means Trane's own data science team can integrate Lattice Graph outputs directly into internal R&D pipelines without manual data transfer steps.

A Lattice Graph engagement with Trane would typically begin with a scoped discovery sprint — usually four to eight weeks — focused on a specific target chemistry space: in this case, the PFAS-free dielectric candidates most likely to meet Trane's immersion-system performance specification. Lattice Graph delivers a ranked candidate shortlist with full stability dossiers (phonon, thermodynamic, and where warranted DFT validation), a freedom-to-operate map of the relevant patent landscape, and a synthesis-priority recommendation identifying which leads are most accessible given Trane's existing process capabilities. That deliverable is designed to go directly into a lab team's experimental queue, not into a shelf report. From there, engagements typically expand into a platform access arrangement that gives Trane's chemistry team ongoing query rights against the knowledge graph and API products, enabling the team to run incremental candidate screening independently between deeper simulation campaigns. Pricing is structured around the scope of the initial sprint and the volume of API access required, and is available on a per-engagement or annual-subscription basis. Given the market timing — qualified fluid supply is the rate-limiting factor for immersion-cooling deployment right now — the initial sprint can typically be scoped and launched within two to three weeks of agreement.