Memory Prices 

Historic and current memory and storage prices, collected in the spirit of John C. McCallum's classic memory-price dataset — interactive, with the raw data downloadable. Hover for details, click the legend to toggle series, drag or use the slider to zoom, and use the camera icon to export an image.

⬇ Download CSV Loading…

Price per gigabyte over time

Historical lowest $/GB on a log scale — one line per memory type: DRAM, NAND flash, HBM, and hard disks. Toggle nominal vs inflation-adjusted dollars (constant 2024 $, US CPI-U).

DRAM price by generation

The DRAM line above, broken out by generation across the full history — Pre-DDR (SDRAM/core), DDR, DDR2, DDR3, DDR4, DDR5. (Generation is inferred from product descriptions, so older points are approximate.)

Accelerator cost breakdown

Modeled estimates from Epoch AI: quarterly accelerator cost across the four largest AI-accelerator designers — Nvidia, AMD, Google (TPU) and Amazon (Trainium) — stacked by component (HBM, logic die, packaging/CoWoS, auxiliary), a production-volume-weighted average.

HBM price by generation

By HBM generation (HBM2e → HBM3 → HBM3e → HBM4). HBM is sold only to accelerator makers on confidential contracts — there is no public spot market — so these are sparse industry-analyst estimates (TrendForce / SemiAnalysis), not transaction prices. HBM4 is projected (launches Q3 2026). $/TBps is cost per unit of memory bandwidth (stack price ÷ per-stack bandwidth).
Methodology note. $/GB is the cheapest listed retail price in nominal USD — not contract, average, inflation-adjusted, or a confirmed sale price. DRAM history is the McCallum dataset (extended from mid-2024 by Keepa Amazon prices); NAND is Keepa's cheapest consumer-NVMe price from 2016 (approximate anchors before); hard disks are McCallum's disk dataset (1956–2024) extended by Keepa's cheapest new consumer drive sold by Amazon itself; HBM figures are modeled estimates. Sources are listed below and in the downloadable dataset; please check before citing.
Methodology, sources and caveats

Sources and method

CategoryWhat we trackSource and methodReliability
DRAM $/GB cheapest retail $/GB, overall and by generation (DDR3/DDR4/DDR5) Deep history (1957–2024): the McCallum memory-price dataset (jcmit.net, via the Internet Archive). Mid-2024 onward: the cheapest new consumer DIMM each month from Keepa (Amazon retail price history), refreshed monthly. Reference + live
NAND $/GB cheapest retail SSD $/GB, 2010–present 2016 onward: the cheapest consumer NVMe SSD each month from Keepa (Amazon retail price history), refreshed monthly; SATA and enterprise/datacenter drives are excluded, and per-drive posting glitches are filtered (see caveats). 2010–2016: four approximate pre-NVMe anchor points (no McCallum-equivalent flash dataset exists). Live + approximate
Hard disk $/GB cheapest new consumer hard drive $/GB, 1956–present 1956–2024: the McCallum disk-price dataset (jcmit.net, via the Internet Archive; floppy entries and one RAID-array system, priced 4–6× above single drives of the same months, excluded). After mid-2024: the cheapest new consumer drive sold by Amazon itself each month, from Keepa — Seagate BarraCuda/IronWolf/SkyHawk, WD Blue/Red/Purple and Toshiba P300/N300/X300 only. Renewed, recertified, refurbished, used and server-pull listings, enterprise drives, relabelled ex-enterprise drives and multi-packs are excluded (see caveats). Reference + live
HBM spend and cost breakdown quarterly HBM spend ($B) and each component's share (%) of the accelerator bill of materials (HBM, logic, packaging, auxiliary) Epoch AI (CC-BY): a modeled estimate, production-volume-weighted across the four largest accelerator designers (Nvidia, AMD, Google, Amazon); aggregate only, no per-company split. External estimate
HBM $/GB by generation HBM price per GB and per TB/s of bandwidth, by generation Industry-analyst estimates — TrendForce and SemiAnalysis (HBM has no public spot market); bandwidth from JEDEC/Rambus. Observed periods only; no forward projections. Sparse estimate

Caveats

Updates

DRAM, NAND and hard-disk $/GB refresh monthly from Keepa; HBM updates quarterly (Epoch AI). The McCallum backbone and HBM estimates are fixed. The downloadable CSV lists every point with its source.

Columns in the CSV

The first ten columns are unchanged; new columns are only ever added at the end, so spreadsheets that read columns by position keep working.

ColumnMeaning
dateFirst day of the period. A Keepa (Amazon) row is the whole calendar month, published only after the month has ended.
category, series, metric, unit What the value measures, e.g. DRAM / DDR4 (Keepa) / usd_per_gb.
valueThe measurement in unit (nominal USD).
source, n_samplesWhere it came from; for Keepa, how many listings were priced that month (the value is the cheapest of them).
representative, notesThe listing or item behind the value, and how the point was derived.
capacity_gbCapacity of that item in GB as marketed (a 16 GB DIMM is 16, a 2 TB SSD or hard disk is 2000). McCallum rows use his own Size column; Keepa rows use the capacity the $/GB was computed from. Blank for estimates and aggregates. For Keepa this is the capacity of the cheapest-per-GB listing, so it is not a capacity trend.
unit_price_usdPrice of that item, nominal USD (McCallum's Cost column; for Keepa, capacity × $/GB).
generationDRAM generation (DDR5, DDR4, DDR3, DDR2, DDR, SDRAM, EDO, or SIMM for 30/72-pin modules whose DRAM type isn't stated); for SSDs, the PCIe generation (PCIe Gen3, PCIe Gen4, …); for hard disks, the interface (RLL, SCSI, IDE, SATA-1/-2/-3); or HBM generation. For Keepa DRAM it is the search basket the listing came from. An SSD's PCIe generation and a hard drive's interface are taken only from what the seller states (title or listing bullet points) and is never inferred from speed, so a few SSD rows are blank. Also blank before ~1990 (boards, core memory, chips) and for aggregates.

McCallum's memory $/MB uses binary megabytes, and those historical rows were converted at ×1000, so for McCallum DRAM rows unit_price_usd ÷ capacity_gb runs about 2.4% above value. His disk table uses decimal megabytes, so hard-disk rows have no such offset.

About

Compiled and maintained by David Shim, Stanford DAM project. Questions or corrections: hsshim@stanford.edu.