SandboxAQ Gets $500 Million CHIPS Bet For AI-Designed Chip Materials

A Materials Grant Aimed Below The Fab Line

SandboxAQ has won a $500 million CHIPS Act award from the US Department of Commerce to use AI simulation software on materials needed for domestic semiconductor production.

The Alphabet spinoff is not being funded to manufacture chips.

Its task is research and development around the minerals, molecules and chemicals that sit underneath chipmaking capacity.

That distinction matters because the US onshoring push is not only about fab buildings or lithography tools.

Semiconductor manufacturing also depends on catalysts, magnets, process chemicals and battery inputs that can be exposed to foreign supply risk or environmental constraints.

SandboxAQ's award is aimed at those lower-level dependencies rather than finished processors.

The company said the work will cover novel molecules and formulations for chip production.

The target list includes PFAS-free manufacturing materials, new fabrication catalysts, magnets that avoid reliance on foreign-sourced neodymium and other rare earths, and batteries that do not depend on materials such as lithium when those inputs are mostly sourced abroad.

That list also explains why a materials award can matter even when it does not add immediate wafer capacity.

PFAS restrictions can affect process chemistry, rare-earth exposure can affect magnet supply, and battery inputs can shape the power systems around fabs.

The award therefore sits at the junction of industrial policy, environmental substitution and chip supply-chain resilience.

Physics Models Meet The CHIPS Act Supply Chain

SandboxAQ spun out of Alphabet in 2022 under the chairmanship of former Google chief executive Eric Schmidt.

It describes its large quantitative models as AI systems trained on physics, chemistry and biology rather than human language.

The company argues that this makes the models suited to materials discovery, where a predicted compound or formulation must eventually survive lab testing.

The grant sits inside the broader CHIPS and Science Act programme, signed in 2022 and designed in part to distribute $52 billion to revive US semiconductor manufacturing.

The public record already includes a 10 percent US government stake in Intel, but the SandboxAQ award shows a different layer of the policy: strengthening the material inputs that fabs need before production can be resilient.

The technical risk is that AI-generated materials can look plausible in simulation and still fail in qualification.

SandboxAQ uses synthetic data in its modelling workflow, while also using experimental data where it exists.

The company says final validation comes from lab performance, because a material either works under testing or it does not.

The cautionary comparison is pharmaceuticals: AI-designed drugs had been forecast for 2025, yet the US National Institutes of Health says AI has not designed a functional medicine.

For chip materials, that makes validation more important than model branding.

A simulated catalyst or PFAS substitute has to meet industrial requirements before it can reduce supply-chain risk.

The Proof Will Be Industrial Qualification, Not AI Claims

The clearest near-term test is not whether the models generate candidates.

It is whether candidates move through screening, lab validation and semiconductor qualification without creating new bottlenecks for manufacturers.

SandboxAQ said previous work on catalysts, battery materials, alloy discovery and PFAS breakdown will feed into the CHIPS Act-funded programme.

The company also says commercial deployments have reduced candidate-screening timelines from months to weeks.

That is useful if it holds inside semiconductor materials work, but it does not remove the qualification burden.

Chip industry adoption requires reliable performance, repeatable supply and compatibility with existing manufacturing processes.

There is also a deployment distinction inside the programme.

Some work, such as PFAS mitigation, new batteries and related substitutions, could be applied to existing fabs.

Other work will move on different timelines because semiconductor qualification is slow by design.

SandboxAQ says existing manufacturers still have to validate and qualify any material before adoption, so the programme is not a shortcut around factory approval or a plan to build new fabrication capacity.

That qualification burden is the practical boundary around the award.

A faster screening stage can reduce the number of weak candidates researchers carry forward, but fabs still need materials that behave consistently in production settings.

The useful outcome would be fewer imported dependencies without adding untested process risk.

The next checkpoint is whether the $500 million award produces qualified materials that reduce dependence on foreign-sourced inputs, not simply more AI-discovery demonstrations.

For US chip policy, the important result would be a verified material supply option that helps fabs operate with fewer external chokepoints.