The US invested $500 million in SandboxAQ to reduce chip production’s reliance on China

The US has allocated $500 million to the AI startup SandboxAQ and become its direct shareholder — an unprecedented step for government technology financing. The bet: artificial intelligence will help create an alternative to Chinese materials, without which chip production is impossible today.

What happened

The US Department of Commerce entered into an agreement with SandboxAQ under the CHIPS Act: the startup receives $500 million to develop new chemicals and metals for domestic semiconductor production. In return, the government received a stake in the company — a move atypical even by the standards of active industrial policy during the CHIPS Act era. Reuters first published information about the deal. SandboxAQ grew out of an internal division of Alphabet (Google). The company specializes in quantum technologies and the application of AI in physics and chemistry. Its key products are platforms for molecular simulation and prediction of material properties: without physical synthesis, only computational models.

Why materials matter more than it seems

When people talk about dependence on China in the semiconductor industry, they usually recall rare earth metals and manufacturing equipment. But the real vulnerability runs deeper and is less visible: chip production requires dozens of specialized chemical substances and metals of ultra-high purity. A significant portion of these components are manufactured or processed precisely in China, and there is no quick replacement for them.

• Chemical gases for plasma etching of silicon wafers High-purity metals for PVD deposition: tungsten, molybdenum, cobalt, ruthenium Precursors for chemical vapor deposition (CVD) CMP slurries — chemicals for polishing surfaces between layers Photoresist components for EUV lithography Each of these items is a separate supply chain with its own vulnerabilities. It is impossible to quickly switch them to domestic production: building capacity takes years, and accumulating technological know-how takes decades.

How AI shortens the discovery cycle This is where SandboxAQ enters the game.

The company proposes using quantum simulations and machine learning models to predict the properties of new materials before their physical synthesis. The classical path from idea to industrial application of a new chemical compound takes 20–30 years. The computational approach promises to reduce it to 3–7 years. The logic is similar to what happened in biopharmaceuticals: DeepMind's AlphaFold system radically changed the speed of predicting protein structures. Now researchers are trying to transfer similar methodology to industrial materials — with direct implications for national security.

"The speed of material discovery is becoming a new field of competition between states," — this is the argument with which

SandboxAQ built dialogue with the US government.

What this means The deal with SandboxAQ is a double precedent.

First, the US has for the first time become a shareholder of a technology startup through the mechanism of industrial subsidization. Second, the government is betting on AI not as a product, but as a tool for solving supply chain problems. If the startup delivers on its promises, this will change both the economics and geopolitics of the semiconductor industry. And the first results are separated from us by several years.