The Chip Squeeze Is Not Hitting Everyone Equally

Archi4 min read

semiconductorssupplymanufacturinganalysis

Contents
  1. The line forms by size, not by need
  2. The 3 nm club has a waiting list
  3. Mature nodes: the quiet casualty
  4. Who absorbs the cost
  5. The design trap
  6. What actually changes the picture
  7. A note on the numbers
  8. The opinion, stated plainly

The line forms by size, not by need#

When chip supply tightens, the industry talks about shortages as if they were weather: a storm that passes over everyone. They are not. Capacity at the leading edge is allocated by contract, by prepayment, and by relationships built over years. The buyers who lose first are the ones without any of those.

That means small and mid-sized equipment makers, automotive suppliers buying mature-node parts, and the industrial firms that design a chip once and expect to reorder it a decade later. They do not get allocation meetings. They get told the lead time is now 40 weeks.

The 3 nm club has a waiting list#

Leading-edge capacity at 3 nm and below is concentrated among a handful of foundries. The largest customers, mostly designers of processors and accelerators for data centers, sign multiyear agreements with prepayments. That money funds new fabs, and in return those customers get first claim on output.

The effect compounds. A company that can commit $2 billion in advance gets both the wafers and the pricing. A company that needs 5,000 wafers a quarter gets the spot market and whatever is left.

This is not a conspiracy. It is how capital-intensive industries work when demand outruns supply. But it explains why two firms in the same sector can report wildly different results in the same quarter. One has locked supply. The other is bidding.

Mature nodes: the quiet casualty#

Most of the attention goes to the most advanced chips. The bigger problem, by volume, sits at 28 nm and above. These are the chips in power management, sensors, and motor controllers. They are cheap, unglamorous, and hard to replace once designed into a product.

Foundries have been adding mature-node capacity, but slowly relative to demand. Automotive and industrial buyers, who were late to sign long-term agreements, now compete with consumer electronics firms that can switch designs faster.

The companies hurt most here are the ones with long product life cycles. A car platform runs for seven years. If the microcontroller changes, the whole validation cycle restarts. A consumer gadget maker can redesign in months. A tier-two auto supplier cannot.

Who absorbs the cost#

The cost of a constrained chip does not stay with the chip buyer. It moves. A supplier paying more for a power management IC passes some of it to the automaker, which passes some to the dealer, which passes some to the buyer. The final price increase is small per vehicle, but the supplier's margin takes the first hit.

Smaller suppliers often lack the balance sheet to prepay or to hold inventory. They also lack the engineering staff to qualify a second source quickly. So they wait, and they pay.

The design trap#

There is a second group affected in a way that rarely makes headlines: firms that designed custom silicon on the assumption that capacity would be available when they needed it. A startup that taped out a chip in 2024 may find that its production slot in 2026 is not guaranteed.

This is where the analysis gets uncomfortable. The industry has spent a decade telling hardware startups to go fabless and let the foundries handle production. That advice works when capacity is loose. When it tightens, fabless companies with no long-term agreement are at the back of the line, and their investors notice.

Some respond by moving to older nodes, which are more available but slower and less efficient. Others delay product launches. A few sign agreements they cannot really afford, trading margin for certainty.

What actually changes the picture#

The obvious answer is more fabs. New capacity is being built in the United States, Japan, and Europe, backed by public money. But fabs take three to four years to come online, and the equipment to fill them is itself constrained.

The less obvious answer is design discipline. Companies that build in margin for a second source, or that choose parts available from multiple foundries, fare better. That is not a slogan. It is a procurement decision made at the schematic stage, often by an engineer who has never seen a supply agreement.

The firms most exposed are those that treat the chip as a commodity to be sourced at the last minute. They are the ones who discover, when allocation tightens, that they were never really in the queue.

A note on the numbers#

Lead times for some mature-node parts have stretched past 30 weeks. Advanced-node wafers are effectively sold out for the next several quarters at the largest foundries, according to public statements from the companies themselves. Those figures move, and they vary by part. But the direction has been consistent since 2023.

The point is not that the sky is falling. It is that the pain is unevenly distributed, and the distribution follows bargaining power, not need. A medical device maker and a gaming console maker may both need chips. Only one of them has a team whose job is to keep the foundry relationship warm.

The opinion, stated plainly#

The chip supply story is usually told as a technology story. It is a balance-sheet story. The companies with cash and contracts get served first. The companies without them get the leftover capacity and the higher prices.

Policy can help at the edges. Subsidies for new fabs matter. So do programs that help smaller firms qualify second sources. But the core dynamic will not change until capacity exceeds demand, and that is not a near-term forecast.

Until then, the firms that feel the squeeze first are the ones that cannot afford to lose a quarter. They are also the ones whose names rarely appear in the shortage coverage. That is the part of the story worth watching.

More articles