THE AI CHIP SUPPLY CHAIN

The AI chip is the output of a supply chain with a single point of failure at nearly every stage. Each step — design, fabrication, the machines that fabricate, the packaging that assembles — is controlled by one or a few firms. The economics follow directly from that structure: scarcity at each choke point concentrates pricing power and profit there. To understand AI compute economics, you have to understand who controls each link.

This is not an ordinary supply chain that competition smooths out. It is a sequence of near-monopolies stacked end to end, and that is what makes it both lucrative and fragile.

Why is the chip supply chain a series of choke points?

Each stage of making an advanced AI chip demands extreme specialization, vast capital, and expertise accumulated over decades. Those barriers are so high that only one or a few organizations clear them at any given stage.

The consequence is a chain of choke points rather than a competitive market. At each link, a handful of firms — sometimes one — control whether anything passes through. Where supply is that concentrated, the controlling firm sets the price and captures the margin, because buyers have nowhere else to go.

This is the opposite of an efficient, commoditized market. In a textbook competitive market, profit gets competed away; in a chain of monopolized choke points, profit pools at each one. The deviation from market efficiency is precisely where the money is.

Why is advanced fabrication the tightest choke point?

The single narrowest point is leading-edge fabrication. The most advanced chips that train frontier AI models can be manufactured at the required scale and quality by essentially one contract manufacturer in the world.

That concentration exists because a leading-edge fabrication plant costs tens of billions of dollars, takes years to build, and depends on process knowledge that cannot be bought or rushed. The few firms that ever reached the frontier mostly fell away; staying there requires continuous reinvestment most cannot sustain.

The result is a structural dependency. The entire AI industry's most advanced compute funnels through one fabrication bottleneck, which gives that single firm extraordinary leverage and makes the whole chain vulnerable to anything that disrupts it. It is the clearest economic moat in the technology economy.

What is the upstream monopoly almost no one sees?

Behind the fabricator sits an even narrower choke point. The most advanced chips require extreme-ultraviolet lithography machines to pattern their circuits, and those machines are supplied by a single company, based in the Netherlands.

This is a monopoly beneath a monopoly. The leading fabricator depends entirely on this one equipment maker, whose machines are among the most complex devices ever manufactured, assembled from a global web of specialized suppliers and shipped in tiny numbers. No alternative source exists.

So the supply chain's true upstream is one firm making the tools that the one firm making the chips cannot operate without. Control concentrated this tightly, this far upstream, is why the chip supply chain is a matter of national strategy and not merely commerce.

Why has packaging become a second bottleneck?

Even when chips are fabricated, they are not yet usable accelerators. Modern AI processors must be combined with high-bandwidth memory and stitched together through advanced packaging, and that packaging step has itself become a binding constraint.

Advanced packaging capacity is limited and slow to expand, so it gates how many finished AI accelerators can actually ship. An operator can have chips designed and ordered and still be unable to take delivery because the packaging line is full. The bottleneck simply moved one stage downstream.

This matters for anyone forecasting AI supply. The headline often focuses on fabrication, but packaging capacity can be the real limiter on units shipped in a given period. It is another fixed, hard-to-replicate input, and the capital allocation decisions to expand it are made years before the demand they serve.

Why is this scarcity durable rather than temporary?

Most supply shortages resolve as competitors enter and capacity expands. The AI chip chain resists that correction, because each choke point is protected by barriers measured in tens of billions of dollars and decades of expertise.

A new leading-edge fabricator cannot appear in a year, and neither can a rival lithography supplier or a flood of advanced packaging capacity. The lead times and knowledge requirements that created the choke points also defend them, so the scarcity persists far longer than in ordinary markets.

That durability is the whole investment thesis and the whole strategic risk. The firms holding the choke points hold structural moats that compound over time, while everyone downstream remains exposed to constraints they cannot remove. Where that leaves the profits is the question we take up in who actually profits from the AI buildout, and it builds on the cost picture in the real cost of AI compute.