Methodology

What these prices are

Every current level is fact-checked against a named public source and carries one of three flags, exported with the data: reported (printed by the cited source — the DRAMeXchange/TrendForce spot board, an earnings release, a retail tracker snapshot), derived (arithmetic on reported numbers, such as an RDIMM price built from die spot, or a supplier's revenue built from a reported share and a reported industry total), and estimate (modeled, with no public print — treat as order-of-magnitude).

Intraday history is not tick data. Each series is back-cast over the window from today's level and the reported year-over-year move using a deterministic noise process tuned to that contract's volatility, so the shape is directionally faithful and identical on every render — but only the endpoint is a real quote. It is not a substitute for a paid feed like DRAMeXchange or a SemiAnalysis subscription.

HBM in particular is not an open market. Stacks are sold on multi-quarter allocation with hyperscaler prepayments, so the HBM3, HBM3E and HBM4 levels are published per-stack estimates of qualified-supply pricing, not executable quotes.

Units, and why indexing matters

A DRAM die, an HBM stack, an RDIMM and a retail kit are not comparable objects. The price explorer defaults to indexed mode, rebasing every selected series to 100 at the start of the window, so you are comparing percentage moves rather than nonsense like "a stack costs more than a die." Absolute mode shows native units:

• HBM3-24GUSD / stack
• HBM3E-36GUSD / stack
• HBM4-48GUSD / stack
• DDR5-16GUSD / die
• DDR5-16G-eTTUSD / die
• DDR4-8GUSD / die
• DDR4-16GUSD / die
• GDDR6-16GUSD / die
• RDIMM-64GUSD / module
• eSSD-30TUSD / drive
• RTL-16GUSD / module
• RTL-32GUSD / kit
• RTL-64GUSD / kit
• RTL-NVME2TUSD / drive

AI demand exposure score

Each contract carries a 0-100 score estimating what share of its incremental demand traces to the AI datacenter buildout. HBM is 100 by definition. Server RDIMMs score high because accelerator hosts pull two to four times HBM capacity in system memory. Retail kits score zero on demand but are the clearest downstream victims: they absorb the price effect without contributing to the bid. The scatter view plots this against year-over-year price change — the correlation is the whole thesis of the site.

Limitations

• No live exchange feed exists for memory; all public pricing is survey-based.
• Contract pricing settles quarterly, so spot and contract can diverge for months.
• Manufacturer figures blend reported financials with segment estimates.
• Nothing here is investment advice or a procurement quote.