The Demand Cascade

The AI buildout is not flowing through the supply chain where you'd expect.

May 2026. Nvidia announces a partnership with Corning to build three new optical manufacturing facilities in the US, expanding domestic optical capacity by more than 1,000%. Tosoh Corporation in Japan is targeting large-scale production of plastic optical fiber for data centers by 2029. The signal isn't that optical fiber demand is growing — it's that copper is being exited as the substrate for AI infrastructure. Nvidia's rack-scale AI systems will run Corning's fiber instead of copper interconnects. This isn't capacity expansion within a category. It's a substrate transition, and the major players are moving in unison.

That announcement landed the same week as two other data points that sit on opposite ends of the same structural logic — one showing unexpected scarcity where abundance was assumed, one showing unexpected surplus where scarcity was assumed.

Where Oversupply Became Scarcity

Chinese MCU manufacturers — microcontrollers, the commodity end of the semiconductor industry — have entered demand-exceeds-supply conditions for the first time in years.

MCU chips are not AI chips. They're embedded in power management systems, thermal regulators, and infrastructure controllers — exactly the components that multiply as AI data centers scale. This sector had been chronically oversupplied, with Chinese manufacturers carrying large inventory buffers. As of May 2026, that inventory is absorbed. Demand now exceeds supply.

This is the first-order fingerprint of a major infrastructure transition. The primary demand story — AI accelerators, HBM memory, high-bandwidth networking — drives the headlines. The secondary demand wave, flowing into sectors no analyst modeled as constrained, moves quietly until inventory runs out. At that point, the market narrative catches up to the supply chain.

Where Scarcity Is Being Engineered Away

The inverse phenomenon is playing out in rare earths.

MP Materials' CEO stated in May 2026 that demand for dysprosium and terbium — heavy rare earths used in high-performance permanent magnets — will fall sharply as manufacturers develop magnets with minimal or zero rare earth content. This is not a cyclical call. It's structural: the technology to substitute away from these materials is advancing faster than their supply chains can mature.

Rare earths spent a decade framed as a permanent strategic chokepoint. Mining, processing, and midstream logistics infrastructure was built on the assumption that demand would only grow. What actually happened: the price premium on these materials directly incentivized the engineering effort to eliminate them. That substitution is now structurally confirmed.

The implication: price weakness for dysprosium and terbium is not standard demand destruction. It's engineering-driven demand destruction — a category that commodity models don't price well because they treat the demand curve as exogenous. The demand is being designed away from the outside.

The Energy Side Is Moving the Other Way

Against this AI expansion, the energy side is contracting.

Iran fired missiles at US destroyers in the Strait of Hormuz in May 2026. Active exchanges are occurring where 20% of global oil supply transits. Saudi Arabia and Kuwait blocked US airbase access when fighting escalated, then reversed six hours later — unable to remain neutral, unwilling to commit. The speed of that reversal reveals the structural bind.

The cost is already in income statements. US airline jet fuel spending in March 2026 reached $5.06 billion — up 56.4% from $3.23 billion in February. Air Canada has cut four seasonal routes early. easyJet and Ryanair are adding fuel surcharges. These are not anticipatory hedges. The conflict cost is realized.

Beijing adds a second layer. Chinese refiners are prohibited from raising output prices even as input costs rise — a policy-enforced margin squeeze with no natural adjustment mechanism. LNG liquefaction costs are rising separately, projected to push price floors higher globally. Two compressive forces operating simultaneously.

Two Structural Forces, Opposite Directions

The collision is direct: energy supply chains contracting from active conflict while AI infrastructure demand accelerates globally. Both are structural.

Standard sector analysis prices each signal in its own lane. The energy shock flows into aviation and petrochemicals. The AI buildout flows into GPU and HBM manufacturers. MCU chips, rare earths, and optical fiber get priced as disconnected small worlds. The reality is that the same underlying force — AI infrastructure scaling into a constrained energy environment — is driving all of them simultaneously. Simultaneous movements in linked systems compound. The aggregate effect is not the arithmetic sum of independent moves.

The practical implication: the markets that move first aren't always the obvious ones. MCU scarcity wasn't priced before inventory ran out. Rare earth demand decline wasn't priced before the CEO confirmed it publicly. Optical fiber didn't carry AI infrastructure pricing before the Nvidia-Corning announcement made the substrate shift explicit. Step functions aren't priced in advance. They're priced at the step.

The Pattern

Major infrastructure transitions share a consistent fingerprint. They create scarcity in commodity sectors treated as permanently abundant. They destroy value in "strategic" sectors priced for permanent scarcity. They shift the primary substrate to a new medium. All three happen simultaneously, faster than supply chains can reprice.

Copper interconnects didn't exit on a linear slide. MCU inventory didn't deplete on a smooth curve. Rare earth substitution didn't develop on a published schedule. These transitions arrive as step functions.

On-chain perp markets price these movements across oil, commodities, and industrial materials in real time — without the earnings-call lag of equity proxies. The thesis doesn't wait for the quarterly report.

Trade these markets at Blackboard.