The U.S. Commerce Department has signed a definitive agreement awarding SandboxAQ $500 million in CHIPS Research and Development funding, putting a large federal bet behind AI-driven discovery of the materials used inside semiconductor manufacturing.
The award, announced June 17 by the Department of Commerce’s CHIPS R&D Office, is aimed at a less visible part of the chip supply chain: process chemicals, catalysts, magnets, and backup-power materials that fabs depend on before a finished processor ever reaches a data center or phone.
SandboxAQ will use the money to develop new molecules and formulations across four areas: PFAS-free process chemicals, catalysts for semiconductor fabrication, rare earth-free magnets, and battery chemistries for semiconductor facility backup power. The company plans to use its ReAQT platform and large quantitative models, or LQMs, to screen candidate materials virtually before passing the most promising results to lab and manufacturing partners.
The structure is also notable. Commerce will receive a minority, non-controlling equity stake in SandboxAQ in connection with the award, according to the agency. That makes the deal look less like a routine research grant and more like a public-sector attempt to share in the upside if federally backed materials work becomes commercially useful.
The bottleneck is not only fabs
CHIPS Act coverage often focuses on fabs, packaging sites, lithography tools, and advanced logic nodes. The SandboxAQ award points to a deeper problem: domestic chip manufacturing can still depend on foreign-controlled inputs even if the fab itself is built in the United States.
Commerce identified materials bottlenecks that are easy to overlook because they sit inside tools and factory operations rather than on chip spec sheets. PFAS chemicals are used in semiconductor manufacturing for heat transfer, lubrication, insulating coatings, and surface treatment. Catalysts are needed for upstream precursor generation and for abating exhaust gases. Neodymium-based permanent magnets appear in advanced chipmaking equipment, vacuum pumps, and precision actuators. Backup power has to be tightly controlled because even short disruptions can damage yields or force tool shutdowns.
The national-security angle is straightforward. Commerce says China controls more than 90% of global production of neodymium-based permanent magnets, while many battery systems rely on lithium, cobalt, or chemical precursors concentrated overseas. Replacing those dependencies is harder than placing a purchase order with a domestic supplier, because the substitute materials have to meet demanding performance, purity, durability, toxicity, and manufacturing requirements.
What SandboxAQ says its AI will do
SandboxAQ is not pitching a chatbot for chip engineers. In its own announcement, the company describes LQMs as models trained on physics, chemistry, and biology rather than human language. ReAQT generates physics-grounded training data through methods such as density functional theory, molecular dynamics, and reaction modeling, then uses those models inside design-make-test workflows.
The goal is to narrow the search space before expensive lab work begins. In materials science, the hard part is not imagining that alternatives might exist. It is finding candidates that satisfy many constraints at the same time: chemical performance, manufacturability, cost, safety, reliability, integration with existing fab equipment, and availability of feedstocks.
SandboxAQ says its AQCat work, built with NVIDIA-supported high-fidelity quantum chemistry calculations, can screen catalyst candidates far faster than traditional methods. The company also says its AQVolt work will be used for battery chemistry, while ReAQT and LQMs will be applied to rare earth-free magnets and PFAS alternatives. Those claims still have to survive the practical path from simulation to validated formulation to scaled domestic production.
Why PFAS alternatives are a serious fab issue
The PFAS portion may become one of the most watched parts of the program. Semiconductor production uses highly specialized chemicals that must work under tight thermal, electrical, and contamination limits. Replacing a “forever chemical” in a consumer product is difficult enough; replacing a process chemical inside a fab can require proof that the substitute will not reduce yield, damage equipment, introduce new contamination, or create a different environmental problem.
That is why the award matters even if no new material ships soon. Regulatory pressure on PFAS is rising, while chipmakers are under pressure to expand domestic capacity. If fabs cannot qualify safer substitutes at scale, environmental compliance and supply-chain resilience could collide with manufacturing targets. AI screening may help researchers test more candidates faster, but qualification in real semiconductor processes remains the slow, unforgiving part.
A test for scientific AI, not just chip policy
The award also broadens the definition of AI infrastructure. Training clusters, accelerators, and cloud capacity get most of the attention, but the AI boom depends on a semiconductor base that includes materials, chemicals, metrology, power systems, and industrial process control. Commerce is effectively treating AI-enabled materials discovery as part of the infrastructure needed to keep advanced computing supply chains from becoming brittle.
For SandboxAQ, the public test is now specific. The company must show that physics-based AI can produce candidates that are not merely computationally interesting but commercially manufacturable. For the Commerce Department, the test is whether a large award to one private platform company can produce public supply-chain benefits, especially when the government is also taking an equity stake.
The best case is not a single breakthrough material. It is a repeatable pipeline that can move from simulation to lab validation to American manufacturing partners across several categories. If that works, chip policy will have expanded from building fabs to rebuilding the material stack underneath them. If it does not, the award will become a reminder that AI can accelerate discovery without removing the hardest engineering and scale-up work.
