Historical Memory and Storage Prices (1960-2026)

Memory prices have plummeted over six decades, though recent AI demand is introducing new volatility

Since 1960, the cost per gigabyte (GB) of memory and storage has seen a massive, multi-decade decline. While the long-term trend remains downward, recent data indicates that the price per GB for certain memory types has regressed to levels seen over a decade ago, driven largely by the surge in demand for AI accelerators and High Bandwidth Memory (HBM).

Historical Trends in DRAM and NAND Flash

Memory pricing has transitioned from an era where a single megabyte was a significant hardware investment to a commodity market where gigabytes are measured in cents.

DRAM Price Evolution

DRAM prices have followed a consistent downward trajectory across generations, moving from Pre-DDR (SDRAM/core) through DDR, DDR2, DDR3, DDR4, and now DDR5. However, the decline is not linear.

Generation Overlap: The cheapest retail prices often track end-of-life generations being cleared out rather than the leading edge. For example, recent data points in 2025 may still reflect DDR3 pricing for small capacities (e.g., 2 GB sticks), which can make current pricing appear lower than it is for modern standards.
Cyclical Nature: The market exhibits repeating price cycles every few years, often tied to new node sizes, generation shifts, or the opening of new fabrication plants.

NAND Flash and SSDs

NAND flash pricing, specifically for consumer NVMe SSDs since 2016, has mirrored the general decline of DRAM but with its own volatility. Some users note that recent pricing may feel higher than 2020 levels, despite what aggregate retail data suggests, highlighting the gap between nominal retail listings and actual transaction prices.

The Impact of AI and High Bandwidth Memory (HBM)

High Bandwidth Memory (HBM) represents a departure from the consumer commodity market. Because HBM is sold via confidential contracts directly to accelerator makers, there is no public spot market, and pricing is based on industry-analyst estimates from sources like TrendForce and SemiAnalysis.

Accelerator Cost Breakdown

AI accelerators from Nvidia, AMD, Google (TPU), and Amazon (Trainium) see a significant portion of their bill of materials (BOM) dedicated to HBM. The cost of these accelerators is split between:

HBM: High-speed memory stacks.
Logic Die: The primary processing unit.
Packaging/CoWoS: The complex interconnects required to bond memory to logic.
Auxiliary components.

HBM Generations and Projections

As the industry moves from HBM2e to HBM3, HBM3e, and the projected HBM4 (launching Q3 2026), the focus has shifted to the cost per unit of memory bandwidth ($/TBps). This metric is more critical for AI workloads than simple capacity (GB).

Critical Perspectives and Market Analysis

Technical community discussions highlight several nuances that a simple price-per-GB chart may obscure:

The "Software Bloat" Counter-Argument

While hardware costs have dropped, the efficiency of software has declined. Users point out that modern browsers and operating systems are "oppressively hungry" compared to the past, effectively neutralizing some of the gains in memory affordability.

"Now everyone's going to talk about how cheap everything is by comparison - but someone needs to talk about how oppressively hungry browsers and OSes are compared to in the past."

Logarithmic vs. Linear Perception

Because historical data is often presented on a log scale to accommodate the massive price drops from the 1960s, recent price spikes can appear minimal. In a linear scale, the recent increase in memory costs due to AI demand would be far more prominent.

The Commodity vs. Moat Debate

There is an ongoing tension between memory manufacturers, who are positioning their technology as a "state-of-the-art moat AI backbone technology," and critics who argue that memory remains a fungible commodity. Some analysts suggest that tech giants may attempt to vertically integrate and replace these suppliers to avoid the current "hundred year flood" of pricing volatility.

Data Methodology and Sources

The dataset is a synthesis of historical and live data:

Memory Type	Source	Reliability
DRAM (1957–2024)	McCallum memory-price dataset	Reference
DRAM (2024–Present)	Keepa (Amazon retail)	Live
NAND (2016–Present)	Keepa (Amazon retail NVMe)	Live
HBM Spend/Cost	Epoch AI (Modeled estimates)	External Estimate
HBM Price/GB	TrendForce / SemiAnalysis	Sparse Estimate