A three-person San Francisco outfit called Rasyn, which participated in Y Combinator, has filed provisional patent applications for PFAS-free datacenter cooling fluids and two other industrial chemistry targets. The timing reflects an industry squeeze: AI infrastructure is pushing thermal limits while regulators tighten the noose on legacy chemicals.
The company released three chemistry foundation models in early September and plasters its website with a claim to make "industrial chemicals 100,000x faster." RuntimeWire noted the figure lacks a disclosed baseline or test protocol, which is either confident salesmanship or the kind of imprecision that makes potential customers nervous.
An Industry Under Dual Pressure
The stakes are real enough. Data centers consumed approximately 2 percent of global electricity in recent years, per the International Energy Agency, and projections show that share climbing steeply by decade's end as AI workloads multiply. A Lawrence Berkeley National Laboratory report from June pegged 2030 U.S. datacenter draw at somewhere between 521 and 843 terawatt-hours. That range spans 6.7 to 12 percent of total U.S. generation, depending on how aggressively deployments scale.
Immersion cooling has emerged as one thermal escape hatch. Rather than air-cooling rows of servers, operators submerge them in dielectric fluids. Fortune Business Insights valued that market at $359.18 million in 2025 and forecast growth to $2.47 billion by 2034, a compound annual rate near 24 percent. Grand View Research published similar numbers in June, estimating $2.02 billion by 2033. Yet adoption remains patchy. Uptime Institute's 2026 cooling survey found direct liquid systems gaining ground but still occupying a minority share, with full immersion reserved mostly for high-density edge cases.
Supply chains tightened when 3M announced its exit from PFAS manufacturing in late 2022, with production of Novec and Fluorinert fluids that had anchored two-phase immersion deployments for years ceasing by the end of 2025. The EPA expanded Toxics Release Inventory reporting in January to cover 206 PFAS compounds, with compliance deadlines looming in mid-2027. The European Union imposed restrictions on PFAS in firefighting foams last October. Datacenter operators and fluid suppliers are now scrambling to test single-phase hydrocarbon, silicone, and ester-based substitutes.
Three Founders, Three Models, Three Patents
Rasyn's founders are Ansh Tiwari, Ayush Chauhan, and Daood Hashmi. They launched in 2025 with a pitch centered on compressing formulation cycles. RuntimeWire reported last August that two team members have Caltech ties; the institute's Green Labs page lists Tiwari as a Fume Hood Project Associate since fall 2025. Whether that academic connection translates to bench credibility remains an open question.
The startup unveiled three foundation models in early September: Synthon 350M for single-step retrosynthesis (66.23 percent top-1 accuracy on the USPTO-50K benchmark), Marcus 670M for reaction-condition prediction (35.37 percent top-1 on USPTO-CONDITION), and Larmor 2.8B for NMR chemical shift forecasting. The company says the models run through its Marigold software platform, priced from $89 monthly for individual researchers to north of $2,000 for institutional licenses.
The three provisional applications Rasyn disclosed cover asymmetric ester base fluids designed to sidestep PFAS restrictions, boron precatalysts for direct amidation, and agricultural cocrystals pairing spirotetramat with isoproturon. The company's news page noted that "experimental validation remains to be completed" and provided no application numbers or filing dates, leaving the IP claims in a murky zone between ambition and execution.
Rasyn is building a lab to generate proprietary training data, according to RuntimeWire. That mirrors a strategy employed by better-funded peers. Orbital Materials raised a $50 million Series B in May, Fortune reported, and operates a wet lab to feed its LINUS foundation model for crystals and small molecules. Fellow YC cohort members 83 Sciences and Ångström AI are chasing similar AI-native materials discovery, the latter promoting a $150,000 pilot focused on solubility prediction.
Benchmarks and the Credibility Gap
Established materials-informatics vendors have published customer results that offer some comparative yardstick. Huntsman told AZoM in April that Citrine Informatics' platform accelerated polyurethane formulation work. Uncountable reported that Repsol achieved 30 to 40 percent formulation savings and faster R&D cycles, according to a case study published last year. Intellegens highlighted Domino Printing's ink formulation optimization in a webinar series. None of those projects disclosed cycle-time reductions approaching the "years to weeks" language Rasyn used in its YC pitch. RuntimeWire flagged the absence of published customer results demonstrating such compression at scale for Rasyn itself, which is perhaps unsurprising given the company's youth but leaves its boldest claims in the realm of aspiration.
Traditional chemical suppliers are not sitting idle. Castrol launched its ON DC 20 single-phase hydrocarbon fluid for immersion cooling, with product sheets and safety data now available. Shell introduced its Immersion Cooling Fluid S5X and S3 lines, the latter debuting in November 2024. Engineered Fluids maintains a library of technical and safety data sheets for its EC-series dielectric fluids. Wacker Chemie drew trade coverage last August for its Helisol EC silicone-based immersion product. All are navigating the same regulatory gauntlet: PFAS scrutiny, TSCA new-chemical review timelines that can stretch one to three years per EPA hearing records, and datacenter operators' qualification protocols that favor incumbents.
Federal Signals and European Compliance
The National Science Foundation announced $50 million for two Materials Innovation Platforms in August, signaling federal appetite for infrastructure that accelerates discovery and qualification. Department of Energy funding-opportunity announcements published between March and May emphasized AI-enabled materials acceleration, with program language targeting reductions from "years to months–few years," a less breathless formulation than Rasyn's marketing copy. The EU AI Act's general-purpose AI obligations became enforceable in early August, adding transparency and documentation requirements for foundation models offered in Europe, including those marketed for scientific use.
Rasyn's defensive position will hinge on the proprietary data its lab produces and whether the startup can publish validated customer results, RuntimeWire argued. The company's August LinkedIn post promoting Marigold emphasized self-hosting: "runs on your cluster… data stays in the building." That pitch targets enterprises wary of sending formulation IP to external clouds, a legitimate concern in an industry where a novel polyurethane formula or catalysis pathway can represent years of R&D investment.
Whether that reassures chemical manufacturers accustomed to multi-year qualification cycles, or whether Rasyn's benchmark claims translate into contracted work, remains unproven. The immersion-cooling market is growing fast, PFAS alternatives are urgently needed, and a cohort of AI-for-chemistry startups is racing to compress development timelines. Rasyn has launched models and filed provisional applications. The next milestone is a named customer using those models to ship a qualified fluid that passes real-world thermal and safety testing. Until then, the startup's "100,000x faster" claim is a hypothesis in search of validation.
