CHIPS Act drops $500M into SandboxAQ to find chip materials, not solve chips

The US government just put $500 million behind a very specific bet: an Alphabet spinoff’s AI simulation system can help find new semiconductor manufacturing materials inside the US supply chain. SandboxAQ announced the award Wednesday, and it is part of the CHIPS and Science Act push meant to reduce dependence on foreign-sourced inputs as semiconductor manufacturing moves onshore.

But here is the twist executives should clock immediately: this is not money for chip plants. SandboxAQ says the funding from the Department of Commerce is an R&D grant to turn its AI simulation software toward discoveries needed for domestic chip industry development. The target is the “stuff” chips are made from and powered by: “novel molecules and formulations for semiconductor manufacturing,” plus a set of named material categories designed to break specific supplier bottlenecks.

SandboxAQ’s list is where the policy meets the practical pain points of global manufacturing. The company says its CHIPS Act-funded work will include chip production materials that are free of PFAS, described as “forever chemicals,” new semiconductor fabrication catalysts, magnets that do not rely on foreign-sourced neodymium and other rare earths, and fab-powering batteries that do not rely on majority foreign-sourced materials like lithium. In other words, the grant is trying to attack multiple choke points at once, from chemical feedstocks to power storage to components embedded in manufacturing equipment.

This matters because the broader CHIPS story is still a work in progress. The CHIPS and Science Act was signed into law by President Biden in 2022 and was designed in part to dole out $52 billion to US firms to reignite domestic semiconductor manufacturing. Four years on, the government’s many investments have delivered some wins, including an acquisition of a 10 percent stake in Intel aimed at helping keep the company afloat. Yet the central objective remains: reducing reliance on foreign supply chains and manufacturers, which is exactly what the $500 million R&D line item is meant to make easier.

Now add the AI layer, and the strategy gets interesting. SandboxAQ describes its Large Quantitative Models (LQMs) as “AI systems trained on the laws of physics, chemistry, and biology, not human language.” The pitch is that this training approach makes the models better suited to discover materials rather than merely pattern-match text. The hope is that LQMs can generate material predictions that researchers then test in the lab, mapping to the same “design-make-test” workflow people have been trying to apply in AI-assisted drug discovery for years.

If you are thinking, “Wait, AI designed drugs haven’t worked yet,” you’re not wrong. The source notes that despite AI industry leaders prognosticating that 2025 would bring AI-designed drugs, the US National Institutes of Health has not yet seen AI design a functional medicine. So why should anyone presume AI will succeed where drug research has struggled, especially in tightly constrained manufacturing domains like chip processes and battery chemistries?

SandboxAQ’s answer is basically: we are still doing science, not magic. The company’s announcement says its LQMs aren’t necessarily grounded in real-world data in every step, relying in part on synthetic data that is then fed into the LQMs to train design-make-test workflows. A company spokesperson told The Register in an email that it still uses real-world data where possible: “Where experimental data exists, we incorporate it,” SandboxAQ said. “Where it doesn't, we can still move forward and solve the problem.”

That’s paired with a frank acknowledgment that AI reasoning errors can waste researcher time. When asked whether an error could compound and lead to considerable lost time and lack of results, the company admitted that such a potential is exactly what “any rigorous AI-driven materials program has to answer.” The spokesperson added that models are “trained on the laws of physics and chemistry, so they are anchored to physical reality, rather than free to drift,” and that lab testing is the final check on AI accuracy. “A material either performs in the lab, or it doesn’t, and that validation gate is precisely what prevents a chain of reasoning from running away with itself.”

In plain terms: this grant is structured around the idea that AI can narrow the search, but qualification and validation still do the heavy lifting. SandboxAQ also says it is not starting from zero across all four target areas. It points to previous work on catalysts, battery materials, alloy discovery, and PFAS breakdown that will be incorporated into the CHIPS Act-funded effort. It also claims that in commercial deployment it has already cut development timelines “from months to weeks” at the candidate screening stage.

The semiconductor industry does not move fast, though, and SandboxAQ is careful about that distinction. It says PFAS mitigation work could be rolled out to existing fabs, and the same is true for new batteries and other items, but it admits different verticals will run on different timelines. It emphasizes that qualification is “genuinely rigorous” and takes time, and it does not frame the program as standing up new fabrication capacity from scratch. Instead, it frames the path forward as running through validation and industrial qualification with existing manufacturers.

For executives and boards, the second-order takeaway is simple: CHIPS Act dollars are not just about manufacturing subsidies. They are also buying downstream resilience by funding upstream material breakthroughs and the AI tooling that may accelerate them. If you operate across chips, batteries, or equipment supply chains, this is a signal that “security of inputs” is becoming a budget line item, and that AI in materials discovery is being treated as part of national industrial strategy, not just a science project.