The memory market has a property that nobody in the supply chain seems to have said out loud this year: the price you can see is not the price you can transact at, and the difference is not rounding error.
Today, on the index I publish, the DDR5 32GB group holds 63 listings. The cheapest visible kit is $349.99, $10.94 a gigabyte. It is out of stock. The cheapest kit I could actually order this afternoon is $429.99 — $13.44/GB, 23% higher. On a forty-kit line item that is $3,200 of margin that exists on the quoting spreadsheet and not in a pallet.
That is one capacity, on one day, at two retailers, measured by the script that regenerates this site every two hours. It is also not unusual: the DDR4 spot market has been reported at up to 172% over contract price (Goldman Sachs, carried by TechPowerUp, March 2026). In a shortage, the visible price is not the best available price. It is the panic price, and a buyer who quotes off it is not being careful, they are being late.
The vendors closest to this know it, because they have been writing it down for months. HIPER Global's mid-2026 customer advisory reports industrial lead times pushing past 30 weeks and legacy DDR4 at times trading *above* newer DDR5. spo-comm published a price-increase explainer in November and a follow-up in January. Compex Systems issued three price notices in 2026. Every one of those pages is an argument to a customer about a number — which means the exposure in this market is not the sticker price. It is the distance between your quote date and your buy date, and that distance is only visible if somebody checks the listing, not the search result.
Three things follow, and none of them require a forecast:
1. Date every number you quote. A quote without a date and a source link is a rumour with a total, and in this market it goes stale in days. 2. Quote off the cheapest listing you can order, not the cheapest you can see. If those are two different numbers, the second one is the one that will embarrass you. 3. Shorten or index the validity window. If a fixed price has to survive more than a week, tie it to a published figure you can cite — including when you are wrong, which is what protects the relationship.
What I cannot tell you is direction. This index has one day of history, so a trend claim from me would be theatre; anyone selling you a curve right now is reselling a TrendForce abstract. What I can sell you is a dated, linked, verified cross-check of the specific line items you are about to quote or publish — the sample is on this site, and it says which lines I could not stand behind.
Limits, stated before someone else has to. This index reads two retailers' visible listings, mostly desktop UDIMM and SODIMM kits, so for a buyer sourcing ECC RDIMMs through distribution it is a direction, not a landed cost. Buyability is verified on a sample of listings, in stock-marked order, and the count is published beside every page rather than buried. Prices are all-in with shipping, as read, not negotiated. If a number here ever contradicts a number in front of you, believe the one in front of you and tell me, because the ledger is public and I would rather be corrected in an email than in a lost order.
Sources read for this piece: the DDR5 32GB sweep group at componentregister.com (97 listings, 2026-09-25); HIPER Global, "Memory and SSD Supply Chain Mid-2026"; spo-comm, "Preiserhöhung 2026" and "Anhaltende CPU- und Speicher-Problematik 2026"; Compex Systems, "Notice: DDR & Flash Supply Update & Cost Adjustment"; CESIPC, "Will Industrial PC Prices Keep Rising in 2026?"; TechPowerUp on Goldman Sachs DRAM analysis; TrendForce consumer DRAM press notes.