Dear Readers,

AI is often framed as an abstract contest of models and algorithms, but today’s lead story pulls it sharply back to earth: the real bottleneck may not be code at all, but the rare earths, magnets, optics, and processing chokepoints that keep the AI hardware stack alive - and increasingly politicized. In this issue, you’ll see why Greenland, China, and seemingly obscure materials like neodymium or erbium are quietly shaping the balance of power in the AI era, how rare earths are turning into leverage rather than fuel, and why export controls now ripple straight into data centers and chip fabs. We also zoom out to map who actually wins and loses as supply chains harden, what “friend-shoring” really buys the US, Europe, and Japan, and why diversification may reduce risk without eliminating dependence. If you want to understand where AI’s physical limits and geopolitical fault lines are forming right now, and what that means for the next wave of compute, keep reading.

All the best,

“Greenland is rich in natural resources including iron ore, graphite, tungsten, palladium, vanadium, zinc, gold, uranium, copper, and oil. But the resources attracting the most attention to the region are rare earth elements (REEs). Vulnerabilities in U.S. REE supply chains for defense and commercial needs have recently been at the forefront of policy issues in Washington.”

— csis.org

AI often gets described like it’s pure math: weights, tokens, gradients, a kind of digital mind floating above the physical world. But the AI boom is, in practice, a materials boom. Every breakthrough model is anchored in warehouses full of servers; every training run is anchored in power electronics, cooling systems, optical networks, and precision manufacturing equipment. The “intelligence” is software, but the leverage is frequently hardware, and hardware is frequently geology plus chemistry plus geopolitics.

That’s why “rare earths” have returned as a strategic obsession. Not because neodymium is sprinkled into an AI model like a magic ingredient. In fact, most AI accelerators are still fundamentally silicon devices. The story is more subtle - and more dangerous: rare earth elements (REEs) and their close cousins (like gallium and germanium) show up at critical chokepoints along the AI supply chain: polishing steps in chip fabs, the highest-performance magnets inside motors and precision tools, fiber-optic amplification that keeps data moving, and the high-efficiency power and cooling infrastructure without which modern AI data centers would overheat or black out.

And then comes the second layer: who controls those chokepoints. Here, the inconvenient truth is that “who has the ore” matters less than “who can refine, separate, and turn it into industrial components at scale.” On that dimension, China’s advantage has been overwhelming for decades - and it increasingly behaves like a strategic asset, not a normal commodity market.

So the guiding question is straightforward, and a little unsettling: In the AI era, are rare earths becoming the new oil - not because they power computers, but because they can pressure the powers that build them?