Historical Memory Prices 1960-2026: Trends and AI Impact

Memory and storage prices have followed a long-term downward trajectory in cost per gigabyte since the 1960s, but recent spikes in demand for AI accelerators have introduced significant volatility. While the cost of consumer-grade DRAM and NAND flash continues to decline overall, the emergence of High Bandwidth Memory (HBM) has created a specialized, non-public market where pricing is driven by confidential contracts between memory makers and AI chip designers.

Long-Term Memory Price Trends (1960–2026)

The cost per gigabyte for DRAM and NAND flash has dropped by several orders of magnitude over the last six decades. This decline is tracked across multiple generations of memory technology, from pre-DDR (SDRAM/core) through DDR5 and from early flash storage to modern NVMe SSDs.

DRAM and NAND Flash Pricing

• DRAM: The historical dataset, which builds upon the McCallum memory-price dataset, tracks the cheapest retail price per gigabyte in nominal USD. The data shows a clear progression across generations (Pre-DDR, DDR, DDR2, DDR3, DDR4, DDR5).
• NAND Flash: Tracking began in 2010, with NVMe SSD pricing being the cheapest consumer retail price per gigabyte from 2016 onward.

Key Methodology and Caveats

To maintain accuracy, the dataset uses a specific set of sources and reliability levels:

Category	Source	Reliability
DRAM $/GB	McCallum dataset (1957–2024) and Keepa/Amazon (mid-2024 onward)	Reference + Live
NAND $/GB	Keepa/Amazon (2016 onward)	Live + Approximate
HBM Spend	Epoch AI (modeled estimates)	External Estimate
HBM $/GB	TrendForce and SemiAnalysis (industry analyst estimates)	Sparse Estimate

Important Caveats:

• Nominal USD: Prices are not inflation-adjusted, which would otherwise make the historical costs appear even higher.
• Retail vs. Contract: The data tracks the cheapest listed retail price, which often reflects end-of-life generations being cleared out rather than leading-edge technology.
• Filtering: To avoid outliers, any SSD listing more than 60% below its typical price is excluded.

The AI Accelerator Market and HBM

High Bandwidth Memory (HBM) represents a shift in how memory is priced and priced. Unlike DRAM and NAND, HBM is sold via confidential contracts to accelerator makers (Nvidia, AMD, Google, and Amazon), meaning there is no public spot market.

HBM Generations and Projections

Memory bandwidth is measured as cost per unit of memory bandwidth (stack price divided by per-stack bandwidth). The current progression is HBM2e $\rightarrow$ HBM3 $\rightarrow$ HBM3e $\rightarrow$ HBM4 (projected for Q3 2026).

Accelerator Cost Breakdown

According to modeled estimates from Epoch AI, the cost of AI accelerators is a production-volume-weighted average across the four largest designers. The cost is split between HBM, the logic die, packaging (CoWoS), and auxiliary components.

Community Insights and Technical Counterpoints

Discussion among technical professionals on Hacker News suggests that the long-term decline in memory prices has not simply reduced costs, but has enabled entirely new classes of software and applications.

The "Regression" Observation

Some users noted that current DDR5 prices per gigabyte are roughly equivalent to DDR3 prices from around 2010. This has led to a discussion on whether we are experiencing a "regression" in pricing trends due to AI demand.

"So a price per GB today is about the same as it was in 2010. 16 year regression, wow!"

Software Bloat and Memory Hunger

Another point of contention is that while hardware costs have plummeted, the efficiency of software has decreased. Users argued that modern browsers and operating systems are significantly more memory-hungry than their predecessors, effectively neutralizing the cost gains.

"Someone needs to talk about how oppressively hungry browsers and OSes are compared to in the past."

The Log Scale Perspective

Critics of the graph's presentation argued that using a log scale can mask the real-world impact of recent price spikes. Furthermore, they pointed out that the absurdity of measuring price per gigabyte in the 1960s—when systems were measured in kilobytes or megabytes—is a conceptually flawed approach to historical analysis.

"It certainly doesn't look as bad as it really is when presented on a log scale chart."

Economic Impact of AI

Some users speculated that the AI boom is currently justifying huge upfront production costs for new memory fabs, which may lead to a future surplus of capacity once the AI demand spike settles, potentially making high-capacity RAM (e.g., 1TB) significantly cheaper in the long run.