The Factory Nobody Knows Is Building America's AI Chips

In the winter of 2021, a group of engineers in Bloomington, Minnesota stood in a fabrication plant that most of the semiconductor world had written off. The facility, originally built by Control Data Corporation in the 1980s and later operated by Cypress Semiconductor, had spent decades making chips that weren't glamorous enough to attract attention from the Nvidias or TSMCs of the world. Analog circuits. Radiation-hardened processors for satellites. MEMS sensors for industrial applications. Useful stuff, but nothing that would land on the cover of Wired.

Then something shifted. The CHIPS and Science Act was still months from becoming law, but the anxiety driving it was already acute. Pentagon procurement officers were waking up to an uncomfortable fact: the United States had effectively outsourced its most sensitive semiconductor manufacturing to a small island 100 miles off the coast of China. For certain categories of chips (the kind that go into missile guidance systems, satellite communications, and nuclear weapon monitoring), this wasn't just a supply chain inconvenience. It was a national security emergency.

The problem was worse than most people realized. Advanced logic chips get all the headlines. But the defense and aerospace world runs on a different class of silicon altogether. Rad-hard chips that can survive cosmic radiation. Mixed-signal circuits that convert real-world phenomena into digital data. MEMS devices that sense acceleration, pressure, and rotation. These aren't the kind of chips TSMC builds on its cutting-edge 3-nanometer process. They're made on older nodes (90nm, 130nm, sometimes 200nm) using specialized processes that the big foundries consider low-margin distractions. And for years, only a handful of facilities in the entire United States could produce them.

One of those facilities sat in Bloomington. But its owner was barely solvent.

The backstory matters. When Cypress Semiconductor decided to exit the foundry business in the mid-2010s, the fab could have easily been shuttered. Equipment gets sold for scrap. Cleanrooms get converted into data centers or demolished. It happens constantly in the semiconductor industry. Instead, a private equity group bought the operation, reorganized it, and took it public in 2021 at the exact moment Washington decided that domestic chip manufacturing was a matter of existential importance.

The timing looked lucky. But luck doesn't explain what happened next.

The company began winning contracts that nobody expected a sub-billion-dollar foundry to land. The Department of Defense selected it for a technology development program aimed at creating radiation-hardened chips using advanced packaging techniques. NASA tapped it for sensor fabrication. The Department of Energy signed on for detector work. And then came the big one: a Department of Commerce CHIPS Act award, not the multibillion-dollar prizes that went to Intel and TSMC, but a targeted investment specifically because this was the only domestically owned, U.S.-based pure-play technology foundry left standing.

Read that again. The only one.

There are other fabs in America, obviously. GlobalFoundries operates a massive plant in Malta, New York (but it's owned by Mubadala, the Abu Dhabi sovereign wealth fund). Intel makes chips in Oregon, Arizona, and Ohio (but Intel designs its own processors and only recently started serious foundry work for outside customers). Texas Instruments runs fabs in Texas and Utah (but those are captive facilities making TI's own analog chips). Samsung has a fab in Taylor, Texas (South Korean ownership). TSMC is building in Arizona (Taiwanese ownership).

For a Pentagon program manager who needs a domestically owned foundry with ITAR compliance, a security-cleared workforce, and the ability to run custom process flows for small-volume specialty chips, the options narrow fast. They narrow, in fact, to one.

This creates a strange and powerful dynamic. The company isn't competing with TSMC on leading-edge logic. It isn't trying to out-scale Samsung on memory. Instead, it occupies a niche that is small by global semiconductor standards but irreplaceable by national security standards. And the AI boom is making that niche bigger.

Here's why. As AI systems move from cloud data centers to the physical world (autonomous vehicles, battlefield robots, satellite constellations, industrial automation), they need specialized sensor chips, rad-hard processors, and mixed-signal components that work in harsh environments. The volume isn't enormous. But the strategic value is immense, and the willingness to pay premium prices is growing. When Lockheed Martin or Northrop Grumman needs a custom MEMS device for an autonomous drone, they can't send the design files to Shenzhen.

The company's revenue is still modest. Its margins are lumpy. It burns cash in some quarters and barely breaks even in others. Wall Street analysts who cover it tend to focus on the gap between its current financials and its future potential, and that gap is real. But the bears are pricing this like a struggling small foundry. They're ignoring what it actually is: the last domestically owned pure-play technology foundry in America, operating at the exact moment when the U.S. government has decided that reshoring semiconductor manufacturing is worth spending $52 billion to achieve.