What Is DRAM? The Investor Guide to Memory, HBM and the Chip Cycle

HBM gets most of the attention in the AI market.

But HBM is still DRAM.

And DRAM remains one of the most important technologies behind PCs, smartphones, servers, GPUs and AI accelerators.

For investors following Samsung Electronics, SK hynix or Micron, understanding DRAM helps explain not only how memory works, but also why memory-company earnings can rise and fall so sharply.

Bottom Line

DRAM is the temporary working memory that supplies active data to processors.

Its importance to investors comes from five variables:

Pricing × Bit Shipments × Product Mix × Cost/Yield × Capex

That framework matters more than simply asking whether DRAM prices are going up.

A memory company can benefit from higher prices and still disappoint investors if:

shipments fall,
product mix weakens,
manufacturing yield deteriorates,
capital spending rises too quickly,
or the market already priced in the recovery.

What Is DRAM?

DRAM stands for Dynamic Random Access Memory.

It temporarily stores data that a CPU, GPU or other processor is actively using.

Unlike NAND flash, DRAM is volatile.

That means the data disappears when power is removed.

A simplified computing system works like this:

Storage → DRAM → Processor

An SSD stores data permanently.

DRAM loads the information needed now.

The processor works on that data.

This makes DRAM much faster for active workloads than trying to process everything directly from storage.

Why Is It Called “Dynamic”?

A conventional DRAM cell is commonly explained using:

one transistor
one capacitor

The capacitor temporarily stores an electrical charge.

But that charge gradually leaks.

So the memory must periodically refresh the stored state while power remains available.

That is the “dynamic” part of Dynamic RAM.

A simplified process is:

Store charge → charge leaks → refresh → data restored

DRAM therefore trades some refresh complexity for much higher density and lower cost per bit than SRAM.

DRAM vs SRAM vs NAND

These technologies solve different problems.

Technology	Main Role	Keeps Data Without Power?	Typical Use
DRAM	Working memory	No	PCs, servers, smartphones, AI systems
SRAM	Very fast temporary memory	No	CPU/GPU cache
NAND	Persistent storage	Yes	SSDs, smartphones, storage devices

SRAM is generally faster than DRAM but much more expensive per bit.

NAND is slower than DRAM but can retain data without power and provides much cheaper storage capacity.

The correct comparison is therefore not:

Which technology is better?

It is:

Which layer of the computing system does the workload need?

The Four DRAM Families Investors Should Know

DRAM is not one single product.

Different DRAM families are optimized for different combinations of bandwidth, power, capacity and cost.

DDR

DDR — Double Data Rate

Used primarily as general-purpose system memory in:

PCs
servers
workstations

DDR5 is the current major generation used across modern computing platforms.

For investors, DDR demand is closely linked to:

PC demand
server demand
memory capacity per system
inventory
contract pricing

LPDDR

LPDDR — Low-Power Double Data Rate

Designed for high performance with lower power consumption.

Common applications include:

smartphones
tablets
thin laptops
vehicles
edge AI systems

Samsung's current LPDDR portfolio includes LPDDR5, LPDDR5X and LPDDR6, reflecting the expanding role of low-power DRAM beyond smartphones into AI and automotive systems.

The key advantage is not simply speed.

It is:

performance per watt

That becomes increasingly important as AI moves from data centers to devices.

GDDR

GDDR — Graphics Double Data Rate

Designed for bandwidth-intensive graphics and parallel workloads.

Typical uses include:

graphics cards
game consoles
graphics workstations
selected accelerators

GDDR provides high data rates without the same packaging structure as HBM.

This can make it attractive where extremely high HBM bandwidth is unnecessary or too expensive.

HBM

HBM — High Bandwidth Memory

HBM stacks DRAM dies vertically and places them close to high-performance processors.

Its purpose is to provide extremely high aggregate bandwidth for:

AI accelerators
GPUs
high-performance computing

HBM therefore should not be viewed as a completely separate memory technology.

It is a specialized, high-bandwidth form of DRAM.

For a detailed explanation, see KMI's dedicated HBM Investor Guide.

DDR5 vs HBM: Why AI Systems Need Both

A common misconception is that HBM will replace conventional DRAM.

That is usually the wrong way to think about it.

In high-end AI systems:

DDR5

often provides CPU and system memory.

HBM

provides extremely high-bandwidth memory close to GPUs or AI accelerators.

NAND SSDs

provide persistent storage.

A simplified hierarchy looks like this:

CPU Cache → DDR → HBM → SSD / Storage

These technologies complement one another.

AI growth can therefore increase demand across several memory layers at the same time.

Why AI Is Increasing DRAM Demand

AI systems move enormous amounts of data.

Training may require memory for:

model parameters
training batches
activations
gradients
temporary working data

Inference may require:

model weights
KV cache
simultaneous user requests
retrieval data
temporary states

The important investment point is:

More AI compute does not automatically mean only more HBM.

AI infrastructure can also increase demand for:

server DRAM
low-power DRAM
enterprise SSDs
networking memory
storage

The entire memory hierarchy matters.

Why DRAM Prices Are So Cyclical

DRAM manufacturing requires enormous capital investment.

Factories, semiconductor equipment, process transitions and yield improvements take time.

Demand can change much faster.

That creates a structural mismatch.

A typical upcycle can look like this:

Demand rises
→ inventory falls
→ orders increase
→ supply tightens
→ prices rise
→ earnings improve
→ suppliers invest
→ supply eventually increases

Then the cycle can reverse.

Demand slows
→ inventory rises
→ prices fall
→ margins weaken
→ investment slows
→ supply adjusts

This delay between investment and actual production is one reason memory cycles can be so volatile.

Spot Price vs Contract Price

Investors often focus heavily on DRAM spot prices.

That can be misleading.

Spot prices

reflect relatively small, short-term transactions and can move rapidly.

Contract prices

represent pricing negotiated between memory manufacturers and large customers.

Contract pricing often has a closer relationship with actual company revenue.

However, neither should be viewed alone.

A spot-price increase does not prove that:

Samsung's average selling price increased,
SK hynix's margins improved,
Micron shipped more bits,
or HBM pricing changed.

Investors should combine pricing data with:

inventory
bit shipments
product mix
end demand
capacity plans

Samsung, SK Hynix and Micron

Samsung Electronics, SK hynix and Micron are major global DRAM manufacturers.

But comparing them only by DRAM market share misses much of the investment story.

Samsung Electronics

Important variables include:

conventional DRAM pricing
server DRAM
HBM qualification and shipments
LPDDR demand
manufacturing yield
memory capex
advanced packaging

Samsung's broad semiconductor portfolio gives it exposure across memory, logic and packaging.

But breadth alone does not guarantee stronger earnings.

Execution still matters.

SK Hynix

Important variables include:

HBM shipments
customer programs
server DRAM
product mix
packaging capacity
manufacturing yield
capital spending

Its growing AI-memory exposure makes HBM particularly important.

But investors should still distinguish:

product announcement
→ qualification
→ production
→ shipment
→ earnings

These are different stages.

Micron

Micron provides useful global comparison because it participates across:

conventional DRAM
HBM
mobile memory
automotive memory
NAND
enterprise storage

Important variables include:

data-center mix
HBM shipments
bit output
cost reductions
gross margins
capex

The same framework should be applied to all three manufacturers.

The KMI DRAM Earnings Framework

A simple DRAM revenue model is:

Revenue ≈ Average Selling Price × Bit Shipments

But that is only the starting point.

For investors, a better framework is:

DRAM Earnings = Pricing × Volume × Product Mix × Cost/Yield − Capital Intensity

Pricing

Are contract prices rising?

Not just one headline spot-price indicator.

Volume

Are bit shipments actually increasing?

Higher pricing with falling shipments can limit revenue growth.

Product Mix

Is the company selling more:

HBM
server DRAM
advanced LPDDR

or more lower-value commodity products?

Cost and Yield

Can the company manufacture those products efficiently?

Higher prices do not automatically create higher margins if manufacturing costs rise.

Capital Intensity

How much investment is required to sustain growth?

A memory boom can eventually create the supply that causes the next downturn.

Why DRAM Matters So Much for Korea

The Korean equity market has unusually high exposure to global memory.

Samsung Electronics and SK hynix are major KOSPI constituents and major global memory manufacturers.

That means changes in:

DRAM pricing
AI infrastructure spending
HBM demand
server demand
capex
manufacturing yields

can affect not only individual semiconductor stocks but also the direction of the broader Korean market.

This is one reason headline KOSPI performance can sometimes be dominated by a small number of semiconductor companies.

The Five Indicators KMI Watches

Instead of asking simply:

Are DRAM prices rising?

KMI focuses on five indicators.

1. Contract Pricing

Are realized pricing conditions improving across major DRAM categories?

2. Bit Shipments

Is physical volume growing?

3. Inventory

Are customers rebuilding inventory or reducing orders?

4. Product Mix

Is revenue shifting toward higher-value server DRAM and HBM?

5. Capex and Supply

Are suppliers adding future capacity faster than demand is growing?

Together, these provide a much clearer picture of the memory cycle.

What Investors Should Watch Next

For Samsung Electronics, SK hynix and Micron, the most important signals include:

DRAM contract prices

Are prices improving across PC, server and mobile products?

Server demand

Does AI infrastructure spending continue to support conventional server DRAM as well as HBM?

HBM qualification and shipments

Are product announcements converting into commercial volume?

Inventory

Are customers maintaining lean inventory or beginning to rebuild?

Yield and cost reductions

Are process transitions improving profitability?

Capital spending

Are suppliers staying disciplined or preparing the next oversupply cycle?

Key Takeaways

DRAM is the working memory behind modern computing.

DDR, LPDDR, GDDR and HBM are different DRAM families designed for different workloads.

AI has increased the importance of HBM, but it has not made conventional DRAM irrelevant.

For investors, the key framework is not simply:

DRAM prices are rising.

It is:

Pricing × Bit Shipments × Product Mix × Cost/Yield × Capex

That framework helps explain why memory-company earnings and share prices can behave very differently even during the same industry cycle.

For Korean-market investors, understanding DRAM is essential because Samsung Electronics and SK hynix sit at the center of both the global memory industry and the KOSPI.

Sources

KMI prioritizes primary sources, including:

Samsung Semiconductor — DRAM, DDR, LPDDR and HBM product information
SK hynix Newsroom — DRAM and AI-memory technology disclosures
Micron — DRAM and memory product documentation

Product specifications and market conditions may change over time.

Korea Market Insight

Evidence Before Narrative

This article is for informational and educational purposes only and does not constitute investment advice.