AI Theft, Chip Wars, and the Invisible Metals Powering It All

Highlights

• U.S. officials accuse China of using AI model distillation to replicate frontier capabilities at lower cost, though public evidence remains thin and the practice itself isn't inherently theft.
• Beyond algorithms, AI dominance depends on hardware—China controls ~90% of rare earth processing and ~94% of magnet production, creating structural supply chain asymmetry.
• Chip export bans won't solve the materials problem: midstream refining bottlenecks and heavy rare earth separation gaps remain unaddressed in the West's AI competition strategy.

---

Is China conducting “industrial-scale” AI model distillation from U.S. labs?  Yes, according to a Financial Times (opens in a new tab) piece. Breaking down what’s credible, what’s narrative, and what’s missing—especially the overlooked role of rare earths and critical minerals in the AI hardware race. Designed for investors seeking clarity beyond headlines. The headline is simple, almost cinematic: the U.S. accuses China of stealing artificial intelligence at scale. But beneath the rhetoric lies a more grounded reality—an escalating contest over compute power, cost efficiency, and control of the physical supply chain that actually makes AI possible.

Distillation or Duplication? The Real Fight Over AI Efficiency

The Financial Times reports that U.S. officials believe Chinese entities are using “distillation”—training smaller models on outputs of larger ones—to replicate frontier AI capabilities at lower cost.  This is not pure fiction. Distillation is real, widely used, and economically powerful. It compresses intelligence into cheaper systems—exactly what a capital-constrained competitor would pursue. But here’s the nuance: Distillation is not inherently theft. It becomes contentious only when done via unauthorized access, scraping, or exploitation of proprietary systems.

Where the Story Holds—and Where It Stretches

There is a solid core of truth anchoring the narrative. China is, in fact, constrained by U.S. export controls on advanced AI chips—an engineered bottleneck designed to slow its access to cutting-edge compute. In that environment, it would be entirely rational for Chinese firms to pursue lower-cost pathways, including model distillation, to narrow the gap. This is not speculative; it is consistent with how innovation behaves under pressure. And notably, leading U.S. AI companies such as OpenAI and Anthropic have voiced concerns along these lines, lending credibility to the idea that competitive replication—whether fair or not—is actively underway.

But the story begins to stretch as it moves from inference to assertion. The phrase “industrial-scale theft” carries weight, yet the public evidence remains thin, more suggestive than definitive. Attribution is another gray zone. While the implication of coordinated, state-backed activity looms in the background, it is not concretely established in what has been disclosed. Even the technical reality is more nuanced than the rhetoric allows: distilled models, by admission, do not fully match the performance of their progenitors, though this limitation is often overshadowed by the alarm. What emerges, then, is not purely a technical dispute, but a geopolitical framing—one that blends real competitive pressures with strategic narrative shaping.

The Missing Layer: Rare Earths, Not Algorithms

Here’s what the FT piece doesn’t say—and what matters more to investors:

AI dominance is not just about models. It’s about hardware.

• Advanced semiconductors require rare earth elements for precision manufacturing
• Data centers depend on permanent magnets, cooling systems, and power electronics
• China still controls ~90% of rare earth processing and up to ~94% of magnet production

In other words:

The U.S. may lead in algorithms—but China still anchors the physical stack.

Chip Bans Won’t Solve a Materials Problem

Calls to halt AI chip exports to China may slow compute access. But they do not solve:

• Midstream refining bottlenecks
• Heavy rare earth separation gaps
• Magnet manufacturing dependence

This is the structural asymmetry.

Bottom Line for Investors

This is not just an AI story. It is a supply chain power story. The real question isn’t whether China can copy models.

It’s whether the West can build the industrial base to compete—before cost, scale, and materials advantage decide the outcome.