Chipflation Is a Wafer Allocation, Not a Demand Shock

chipflation-is-a-wafer-allocation · conviction low · status open · horizon 2027 · as of 2026-08-05

Consumer device costs rise because HBM consumes the wafer capacity commodity DRAM would have used, not because consumer demand recovered. An HBM bit occupies roughly 2.5x the area of a commodity bit, so ~9% of DRAM bits is ~20% of DRAM capacity — and at 20% industry output growth against 100% HBM bit growth, commodity bit supply grows about 0%.

Rests on filed figures, not on modelled shares. 3 premises (3 entity); no derived cell is involved, so the undisclosed supply weights that put a range on other pages in this bank cannot move this one.

Exhibits

Exhibit 1Relative performance, indexed to 100How the names in this thesis have traded against SOXX.

Series available as data/chipflation-is-a-wafer-allocation.csv

Exhibit 2Who pays Conventional DRAM and NAND supply (HBM crowding-out), and who keeps the moneyCapturers average 56.8% operating margin against payers' 32.4% — the owners of the scarce thing capture the rent, as expected.

Green/blue = model marks it as CAPTURING the rent (unbound and supplies the scarce good); faded = PAYING it (bound severe or moderate). Operating margin, live.

Exhibit 3What the conviction is actually made ofEach premise and the number it composes to. A conjunction of plausible premises is far weaker than any of them.

Commodity DRAM supply is flat-to-shrinking even while total memory output grows, because HBM is paid for out of the same wafers

23% if the 3 gates are independent, 55% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.

Weakest link: Consumer Electronics Demand at 0.55 — Consumer devices buy commodity DRAM and NAND, not HBM, so they absorb the allocation loss directly rather than competing for the product that displace

The variant

Consensus

Memory pricing is cyclical and mean-reverting. Rising DRAM and NAND contract prices signal a normal upturn driven by recovering end demand across phones, PCs and servers, and will correct when that demand normalises.

Variant

This cycle has a mechanical floor the previous ones did not, because AI demand does not compete with consumer demand for BITS — it competes for WAFERS, at a 2.5x area penalty per bit. Commodity supply can therefore contract in absolute terms while total industry output grows, which is what makes 'memory output +20%' and consumer BOM inflation the same sentence rather than contradictory ones.

Differentiator

Consensus watches contract prices and infers demand. The variant watches HBM's share of DRAM WAFER capacity and infers supply. The two diverge at the point that matters: a demand-driven squeeze ends when demand cools, and an allocation-driven one ends only when HBM bit growth falls to roughly the industry output growth rate — currently 20% against an assumed 100%.

Open questions

HBM's actual share of DRAM wafer capacity is the load-bearing input and is not directly disclosed. The three makers discuss it obliquely in capex commentary; obtaining it would move this from a mechanism to a measurement.
NAND is treated as exposed alongside DRAM here, but NAND does not lose capacity to HBM directly — its exposure runs through eSSD demand for agentic KV spillover, which is a different channel and is not modelled.

Reasoning chain

Commodity DRAM supply is flat-to-shrinking even while total memory output grows, because HBM is paid for out of the same wafers VALID

premises

Conventional DRAM and NAND supply (HBM crowding-out)0.60 moderate
2.5x9%20%
An HBM bit occupies roughly 2.5x the wafer area of a commodity DDR5 bit — TSV overhead, stacking, and yield loss.
why
So an HBM share of ~9% of DRAM BITS is ~20% of DRAM CAPACITY, and the capacity share is what the supply arithmetic runs on.
HBM40.70 moderate
20%100%0%
HBM bit growth tracks accelerator unit growth times rising per-unit content, which is running far above industry output growth.
why
At 20% output growth and 100% HBM bit growth, commodity bit supply grows about 0% — HBM absorbs the entire increment.
Consumer Electronics Demand0.55 moderate
Consumer devices buy commodity DRAM and NAND, not HBM, so they absorb the allocation loss directly rather than competing for the product that displaced them.

AND rather than OR — the area penalty, the growth differential and the consumer exposure must all hold. THE SIGN FLIPS INSIDE THE PLAUSIBLE RANGE, which is why conviction is LOW rather than medium: at 7% HBM bit share commodity supply grows +4.9%, at 9% it is +0.2%, at 12% it contracts -7.3%. The bit share is a desk estimate and it decides the answer. THE FALSIFIER IS SHARP: commodity DDR5 contract prices flat-to-down through 2027 while HBM bit growth stays above 80% would mean either the area penalty is smaller than assumed or capacity additions are larger, and the cannibalisation channel is weak.

Sources

Industry memory output growth ~20%/yr acc 2026-08-05

Write-up

Pre-filled skeleton: chipflation-is-a-wafer-allocation.md