DDR4 Memory Went From $1.40 to $19. Somebody Has to Pay.

How this happened

The cause is straightforward and structural. Samsung, SK Hynix and Micron have redirected DRAM capacity toward high-bandwidth memory for AI accelerators and have been phasing out DDR4 production. That tightens supply of conventional memory even as demand for it continues.

The result is a price move with very few precedents in semiconductor history:

Korea's DRAM export price in June was around $60,000 per kilogram, down 1.7% from May — the first per-unit-weight decline in nine months, attributed to shifting volume mix between advanced and general-purpose chips rather than falling prices.

Meanwhile China exported roughly 179.44 billion integrated circuits worth $177.28 billion in the first half of 2026, up more than 96% year over year by value. Analysis of those figures suggests the surge tracks the memory price cycle rather than any step change in Chinese output — value inflating faster than volume.

Who eats the cost

Memory is not a specialty component. It is in essentially every electronic product manufactured. The companies buying it fall into three groups, with very different ability to pass the cost on:

Group one — can pass it through. Server and AI infrastructure builders selling into a demand environment where customers are capacity-constrained. Price increases are accepted because supply is the binding constraint.

Group two — partially absorb. PC and smartphone manufacturers in competitive consumer markets. Consumer Reports has already flagged that AI data centres consuming RAM could push laptop prices higher in 2026. Some cost passes to consumers; some compresses margins; some shows up as reduced specifications at the same price point.

Group three — absorb it fully, with a lag. Automakers, appliance manufacturers and industrial equipment makers, where memory is a small line item in a product with long design cycles, fixed pricing contracts and no ability to reprice mid-model-year. These companies will report the effect in gross margin without ever naming it as the cause.

Group three is where the unpriced risk sits. A modern vehicle contains substantial memory content. A thirteenfold increase in a component that was previously too cheap to hedge is exactly the kind of cost that does not appear in guidance until it appears in results.

The consumer inflation channel nobody has modelled

There is a macro dimension here that has gone entirely unexamined. Central banks are currently focused on energy-driven inflation from the Middle East conflict. Goods disinflation from electronics has been a reliable, decades-long deflationary force in every developed-market CPI basket — the "quality-adjusted price of computing falls every year" assumption is embedded in how inflation is measured.

If memory costs stay elevated into 2027, that assumption breaks. Electronics stop being a disinflationary contributor and become a neutral or inflationary one, at precisely the moment central banks are trying to determine whether an energy shock is transitory.

That is not a forecast. It is a mechanism that is currently absent from mainstream inflation commentary.

What resolves this

Supply. Samsung, SK Hynix and Micron are accelerating capacity expansion into 2027. TrendForce projects a DRAM supply-demand gap of roughly 1% to 2% in 2026, widening further in 2027 — which sounds like tightness persisting, but capacity additions arrive in step functions, not smoothly.

Chinese suppliers are the other release valve. CXMT in DRAM and YMTC in NAND sell directly into the commodity segment where prices have moved most. Their capacity expansion is aimed at exactly the products in shortest supply.

Taiwan's memory names have already begun pricing the order-transfer opportunity — five Taiwanese memory stocks hit limit-up on August 3, with the combined market value of four bellwethers rising roughly NT$220.8 billion in a single day, on expectations of order transfers and price increases as international manufacturers shift capacity to HBM and DDR5.

What to watch

• Gross margin commentary from consumer electronics and auto manufacturers in Q3 and Q4 reporting. This is where the cost surfaces.
• Contract DRAM pricing rather than spot, which is what actually flows into product costs.
• CXMT and YMTC capacity announcements — the fastest path to relief in commodity tiers.
• Electronics components within CPI baskets in Canada and the U.S.

Screen your holdings for memory-cost exposure using the Quorum AI scanner, and if you hold semiconductor names through a Canadian-listed ETF, note that the dividend treatment differs by account — see our dividend tracker.

Bottom line

The memory supercycle has been covered exclusively as a story about who is making money. The larger and less examined story is who is paying — and that group includes most of the consumer and industrial economy, none of which has warned about it yet.

Frequently asked questions