America Is Building Chip Factories Nobody Wants to Use

Creative Robotics
America Is Building Chip Factories Nobody Wants to Use

There's a peculiar disconnect happening in American technology policy right now. On one hand, Apple just committed $30 billion to buy Broadcom chips manufactured on US soil. On the other, American rare earth miners—despite backing from the Trump administration—are selling the majority of their output to Japan and South Korea because domestic demand remains anemic. Meanwhile, US companies including Coinbase are increasingly adopting Chinese AI models like DeepSeek, even as the State Department moves to restrict their use.

This isn't just policy incoherence. It's a case study in how nationalist industrial policy collides with global market realities.

The Biden administration's CHIPS Act represented the most ambitious American industrial policy in decades, with $52 billion allocated to boost domestic semiconductor manufacturing. The goal was clear: reduce dependence on Asian supply chains and rebuild America's technological sovereignty. Apple's Broadcom deal seems like validation of that vision—a marquee commitment from the world's most valuable company to source American-made components.

But zoom out slightly, and the picture gets messier. Those semiconductors require rare earth elements for their production. American miners are extracting those materials domestically, exactly as policymakers hoped. Yet they're shipping them across the Pacific for processing and integration into Asian supply chains, because that's where the actual demand and manufacturing ecosystems exist. The US invested in extraction but failed to build the downstream processing infrastructure that makes those materials valuable.

The AI chip situation reveals an even thornier problem. DeepSeek, the Chinese AI company now developing its own inference chips to reduce reliance on NVIDIA, has been gaining traction with American enterprises precisely because its models offer compelling price-performance ratios. When intelligence becomes cheaper—as UC Berkeley researchers noted, inference costs have dropped 50-900x annually—companies follow the economics, not the flag.

The State Department can restrict corporate use of Chinese AI, but that doesn't change the fundamental calculus: if Chinese models deliver comparable performance at lower cost, American companies operating on thin margins will find workarounds. They always do. Export controls and entity lists can slow technology transfer, but they can't reverse the commodification of AI capabilities.

What these stories reveal is that reshoring isn't just about building factories or subsidizing production. It's about reconstructing entire ecosystems—the supply chains, the talent pools, the customer bases, the complementary industries that make a technology sector viable. You can't will a semiconductor industry into existence with subsidies alone when the rest of the value chain exists elsewhere.

Apple's Broadcom commitment is real, but it's also an outlier. The company has the scale and profit margins to absorb higher costs in exchange for supply chain security and political goodwill. Most American tech companies don't. They'll continue sourcing components and software from wherever delivers the best combination of price, performance, and availability—regardless of what policymakers in Washington prefer.

The uncomfortable truth is that economic nationalism works best when it aligns with economic incentives. When it doesn't, you get rare earths flowing to Asia despite explicit government support for domestic processing, and Chinese AI models gaining share despite official disapproval. The factories are one thing. Making anyone want to use them is another problem entirely.