DEV Community: AVAQ SEMICONDUCTOR

MCP, eMMC, and eMCP Explained: Differences, Structure, Connection, and Selection Guide for Engineers

AVAQ SEMICONDUCTOR — Tue, 04 Aug 2026 06:30:29 +0000

Introduction

In modern electronic products, PCB space is becoming more limited while memory requirements continue to increase. Smartphones, tablets, wearable devices, industrial controllers, and IoT products all need compact memory solutions that provide enough storage capacity, fast data access, and low power consumption.

A few years ago, designers commonly used separate memory chips:

NAND Flash for data storage
DRAM for system memory
External controllers for memory management

However, this approach increases PCB size, makes routing more difficult, and requires more hardware design effort.

To solve these challenges, semiconductor companies developed several memory integration technologies, including MCP (Multi-Chip Package), eMMC (embedded MultiMediaCard), and eMCP (embedded Multi-Chip Package).

Although these three terms look similar, they describe different concepts:

MCP is a packaging technology.
eMMC is an embedded storage solution.
eMCP combines eMMC and DRAM into one package.

For engineers, procurement teams, and embedded system designers, understanding the relationship between MCP, eMMC, and eMCP is important when selecting memory components for a product.

This article explains the differences, internal structures, connection methods, applications, and practical selection considerations from a hardware engineering perspective.

1. What Is MCP (Multi-Chip Package)?

### 1.1 MCP Definition
**
MCP stands for **Multi-Chip Package.

It is a semiconductor packaging technology that integrates multiple dies or chips inside a single package.

Instead of placing several independent ICs on a PCB, semiconductor manufacturers stack or combine multiple chips into one package.

A typical MCP may contain:

NAND Flash memory
DRAM
SRAM
NOR Flash
Logic ICs

The individual dies remain separate chips, but they share one external package.

A simple MCP structure looks like this:
`
MCP Package

+----------------+
|    Memory Die  |
+----------------+
|    Memory Die  |
+----------------+
|    Controller  |
+----------------+

      BGA Balls

         |
         |
       PCB

`
MCP technology is mainly used when manufacturers need:

Smaller product size
Lower system cost
Simplified PCB design
Higher integration

**1.2 How MCP Works

**
Inside an MCP package, chips can be arranged in several ways:

*1. Vertical stacking
*
Multiple dies are stacked together:

   Top Die

+-----------+
|   DRAM    |
+-----------+

+-----------+
| NAND Flash|
+-----------+

   Package

This saves PCB area because multiple chips occupy the same footprint.

2. Side-by-side placement

Different dies are placed horizontally inside the package.

This method can improve thermal performance and simplify manufacturing for some designs.

1.3 Advantages of MCP

The major advantages of MCP include:

*1. Smaller PCB footprint
*
A single MCP package replaces multiple separate memory devices.

This is especially important for:

Smartphones
Wearables
Compact IoT products

**2. Easier hardware design

**
Engineers only need to place and route one package instead of several memory components.

This reduces:

PCB routing complexity
Signal integrity problems
Manufacturing steps

*3. Better product integration
*
MCP enables manufacturers to create smaller and thinner devices.

2. What Is eMMC?

*2.1 eMMC Definition
*
eMMC stands for embedded MultiMediaCard.

It is an embedded storage device that combines:

NAND Flash memory
Flash controller
MMC interface

inside one BGA package. The eMMC standard is defined by JEDEC specifications for embedded storage applications.

A simplified architecture:
`
Application Processor

            |
            |
      eMMC Interface

            |

   +----------------+
   |     eMMC       |
   |                |
   | NAND Flash     |
   |                |
   | Flash Control  |
   |                |
   +----------------+

`
The key point is:

eMMC is not just NAND Flash.

It is a complete storage solution.

2.2 Why Was eMMC Developed?

Before eMMC became popular, engineers had to manage raw NAND Flash directly.

A traditional design looked like:
`
CPU

NAND Flash

External Controller`

The processor needed to handle complex Flash operations:

Bad block management
Error correction
Wear leveling
Flash translation layer (FTL)

This increased software and hardware development effort.

With eMMC:
`
CPU

eMMC

(NAND + Controller)
`

The internal controller handles NAND management automatically.

This makes product development much easier.

3. How Does eMMC Connect to a System?

The connection between an application processor and eMMC is usually through an MMC interface.

Typical signals include:
`
Application Processor

   |
   |

CMD
CLK
DAT0-DAT7

RST#

   |

 eMMC

`
Important signals:

CLK

Clock signal controlling data transfer timing.

CMD

Command communication between processor and eMMC.

*DAT0-DAT7
*
Data lines for transferring information.

*RST#
*
Reset control signal.

Modern eMMC devices support higher-speed modes defined by JEDEC standards, such as HS200 and HS400 depending on the device generation.

4. What Is eMCP?

*4.1 eMCP Definition
*
eMCP stands for embedded Multi-Chip Package.

Unlike traditional MCP products that simply combine multiple memory dies, eMCP is designed specifically for embedded systems by integrating:

eMMC storage
LPDDR DRAM

into a single BGA package.

The purpose is to provide both:

Non-volatile storage (NAND Flash)
Temporary working memory (DRAM)

while reducing PCB size and simplifying system design.

A typical structure:
`
eMCP

+----------------+
|   LPDDR DRAM   |
+----------------+

+----------------+
|      eMMC      |
|                |
| NAND + Control |
+----------------+

        |
      BGA

        |
       PCB

`
A practical example is Samsung's KMQX60013A-B419, an eMCP memory device combining:

32GB eMMC 5.1 storage
16Gb LPDDR3 DRAM
LPDDR3-1866 interface

inside one compact package.

In a smartphone or embedded device, this single component can replace two separate memory devices:

Before:
`
Application Processor

| |

eMMC LPDDR3

(Storage) (RAM)

After:
Application Processor

    |

 Samsung eMCP

eMMC + LPDDR3

`
This reduces PCB area, shortens memory traces, and simplifies hardware layout.

5. MCP vs eMMC vs eMCP: Main Differences

Feature MCP eMMC eMCP
Full name Multi-Chip Package Embedded MultiMediaCard Embedded Multi-Chip Package
Category Packaging technology Storage solution Memory integration solution
Contains NAND Possible Yes Yes
Contains Controller Optional Yes Yes
Contains DRAM Possible No Yes
Main purpose Combine chips Provide embedded storage Combine storage + RAM
Typical use Mobile, IoT Embedded systems Smartphones, tablets

6. Relationship Between MCP, eMMC, and eMCP

Many people confuse these three terms because they are closely related.

             eMMC

      NAND + Controller

`
The simplest explanation:

*MCP
*
A technology category.

It describes how multiple chips are packaged together.

*eMMC
*
A storage product.

It integrates NAND Flash and a controller.

*eMCP
*
A specific MCP implementation.

It combines eMMC and DRAM into one package.

7. eMMC vs eMCP in Hardware Design

*7.1 Separate Memory Solution
*
Traditional design:
`
CPU

    |              |

  eMMC          LPDDR

`
Advantages:

Flexible component selection
Easier memory upgrades

Disadvantages:

More PCB space
More routing complexity
Longer signal paths
*7.2 eMCP Solution
*
With eMCP:
`
CPU
```
       |

     eMCP
```
eMMC + LPDDR

`
Advantages:

Smaller PCB area
Reduced routing effort
Faster product development

This is why eMCP became popular in compact mobile devices.

8. Applications of MCP, eMMC, and eMCP

*MCP Applications
*
Common applications include:

Feature phones
Wearable devices
IoT products
Compact electronics

*eMMC Applications
*
eMMC is widely used in:

Industrial computers
Embedded Linux systems
Smart displays
Automotive infotainment systems
Consumer electronics

Industrial eMMC products typically integrate NAND Flash, controllers, and MMC interfaces to simplify system design.

*eMCP Applications
*
Typical applications:

Smartphones
Tablets
Wearable devices
Entry-level mobile platforms
Space-limited embedded systems

9. How Engineers Select MCP, eMMC, or eMCP

When selecting memory components, engineers should consider several factors.

*9.1 Storage Capacity
*
For eMMC:
Consider:

8GB
16GB
32GB
64GB
128GB+

For eMCP:

Consider both:

eMMC capacity
LPDDR capacity

Example:

32GB eMMC + 3GB LPDDR
64GB eMMC + 4GB LPDDR

*9.2 Interface Compatibility
*
Check:

Processor memory interface support
eMMC version
LPDDR generation

Examples:

LPDDR3
LPDDR4
LPDDR4X

*9.3 Package Size and Pin Compatibility
*
Verify:

BGA package dimensions
Ball pitch
Ball count
PCB footprint

A replacement part with similar memory capacity may still fail if the package or pin assignment is different.

*9.4 Temperature Rating
*
For industrial products, check:

Commercial temperature range
Industrial temperature range
Automotive qualification

10. Future Development Trends

Memory integration continues to evolve.

The industry is moving toward:

Higher integration

From:
`
Separate Memory

↓

MCP

↓

eMCP

↓

Advanced SiP Solutions
`

*Faster storage interfaces
*
Many high-performance systems are moving from eMMC toward:

UFS storage
Higher-speed memory technologies

*More compact AI and IoT devices
*
Edge AI products require:

More memory bandwidth
Lower power consumption
Smaller packages

Integrated memory solutions will continue to play an important role.

11. Frequently Asked Questions

**Is eMMC the same as MCP?
**No.
eMMC is a storage device containing NAND Flash and a controller. MCP is a packaging technology that combines multiple chips.

*Is eMMC RAM?
*
No.
eMMC is storage memory. It stores operating systems, applications, and user data.
RAM is temporary working memory.

*Is eMCP better than eMMC?
*
They serve different purposes.
eMMC provides storage only.
eMCP provides both storage and DRAM in one package.

*Can I replace eMMC with eMCP?
*
Usually no.
They have different package structures, signals, and system requirements.

Conclusion

MCP, eMMC, and eMCP are closely related memory technologies, but they solve different engineering problems.

MCP is a packaging method for combining multiple chips.
eMMC is an embedded storage solution integrating NAND Flash and a controller.
eMCP combines eMMC and LPDDR memory into one compact package.

For hardware engineers, the choice depends on system requirements:

Need only storage → choose eMMC.
Need storage plus RAM in limited space → consider eMCP.
Need customized multi-chip integration → consider MCP.

Understanding these differences helps engineers design smaller, faster, and more reliable electronic products.

LM324 vs LM324N: Understanding the Real Difference Between These Two Popular Op-Amps

AVAQ SEMICONDUCTOR — Mon, 03 Aug 2026 07:53:04 +0000

When engineers search for LM324 vs LM324N, they are usually trying to answer one practical question:

*“Can I use LM324 and LM324N as replacements for each other?”
*
This question looks simple, but semiconductor part numbers often create confusion. A small suffix after a component number can represent many different things:

Package type
Temperature grade
Performance level
Manufacturing option
Compliance requirement

For the LM324 family, the difference is easier to understand once you know how semiconductor manufacturers name their products.

The short answer is:

*LM324 and LM324N are functionally the same LM324 quad operational amplifier family in most applications. The “N” mainly identifies the package option, commonly a 14-pin PDIP package. However, engineers should still verify the exact datasheet before replacing one with another.
*
In this article, I will explain the difference from an engineer’s point of view, including real design considerations, replacement risks, and how to choose the correct LM324 variant.

1. A Quick Introduction to LM324

The LM324 is one of the most recognized general-purpose operational amplifiers in electronics history.

An operational amplifier, or op-amp, is an analog IC used to amplify voltage signals, perform filtering, create feedback circuits, and process sensor signals.

The LM324 is popular because it provides four independent op-amps inside one package.

Instead of using four separate amplifier ICs, designers can use one LM324 to build multiple analog functions.

Typical applications include:

Sensor signal amplification
Voltage monitoring
Battery management circuits
Industrial control systems
Analog filters
Signal conditioning
Power supply feedback loops

The LM324 family is designed for low-cost and general-purpose applications. It supports single-supply operation, which makes it convenient in systems powered by common voltages such as 5 V, 12 V, and 24 V.

For example, Texas Instruments specifies LM324N as a quad operational amplifier with a 14-pin PDIP package, wide supply range, and approximately 1 MHz bandwidth.

2. What Does LM324N Actually Mean?

Many engineers assume that LM324N is a completely different chip from LM324.

This is usually not correct.

The part number can be understood like this:

Part Number Meaning
LM324 Op-amp family
N Package identifier

The “N” suffix commonly indicates a plastic dual in-line package (PDIP) version.

A PDIP package is the traditional through-hole package used in:

Prototyping boards
Older industrial equipment
Educational electronics
Repair applications

Texas Instruments lists LM324N as a 14-pin PDIP device.

So when you see:

**LM324 → Basic device family

LM324N → LM324 in PDIP package**

The internal function is essentially the same.

3. LM324 vs LM324N: The Main Difference

Let’s compare them directly.

Parameter LM324 LM324N
Device type Quad op amp Quad op amp
Number of amplifiers 4 4
Input type Bipolar Bipolar
Supply voltage Similar Similar
Frequency compensation Internal Internal
Package Depends on order code PDIP-14
Mounting method Depends on package Through-hole
Pin function Same family Same family

The important difference is normally the package designation, not the electrical function.

However, there is one point engineers should remember:

A product name alone does not tell the whole story.

Different manufacturers may produce LM324-compatible devices with slightly different specifications.

Always check:

Datasheet revision
Manufacturer
Electrical limits
Package drawing

before approving a replacement.

4. Why Does the “N” Suffix Exist?

Semiconductor manufacturers use suffixes because one silicon design can be sold in many versions.

For example, the same LM324 circuit may be available as:

DIP package
SOIC package
TSSOP package
Different temperature grades
Lead-free versions

The suffix helps manufacturers and distributors identify the exact ordering option.

For LM324:

Common examples include:

LM324N
LM324D
LM324P
LM324N/NOPB

These may have the same basic amplifier function but different package or production options.

The naming history is also related to National Semiconductor, the original creator of the LM324 family. After Texas Instruments acquired National Semiconductor, TI continued supporting many of these legacy part numbers. TI engineers have explained that LM324N and related suffixes can be confusing because they come from different naming traditions.

5. LM324N Electrical Characteristics

Although the package is different, engineers still need to understand whether the LM324 specifications fit their application.

Supply Voltage

The LM324 family is designed for wide voltage operation.

Typical range:

Single supply: around 3 V to 32 V
Dual supply: approximately ±1.5 V to ±16 V

This makes it suitable for many industrial and embedded applications.

Gain Bandwidth Product

The LM324 has a bandwidth of approximately:

1 MHz

This means it works well for:

DC signals
Low-frequency sensors
Control loops
Slow analog processing

It is not designed for:

High-speed communication
RF circuits
High-frequency signal processing

For example, using LM324 as an amplifier for a fast ADC input is usually a poor design choice because the limited bandwidth and slew rate can affect signal accuracy.

Input Common-Mode Range

One reason LM324 became popular is that its input range can include ground when operating from a single supply.

This is useful in circuits such as:

Sensor interfaces
Battery voltage monitoring
Low-side measurements

However, engineers should not confuse this with rail-to-rail operation.

LM324 is not a modern rail-to-rail amplifier.

6. Can LM324 Replace LM324N?

In most practical situations:

Yes, LM324 can replace LM324N if the package and electrical requirements match.

For example:

A repair engineer replacing an LM324N on an old control board can usually use another LM324N-compatible device.

Common replacement situations:

Industrial control boards

Usually acceptable.

Examples:

Temperature controllers
Motor controllers
Power supplies
Sensor circuits

Usually acceptable if accuracy requirements are not strict.

Examples:

Light sensors
Pressure sensors
Simple analog measurement circuits
Precision measurement circuits

Be careful.

You should compare:

Input offset voltage
Bias current
Temperature drift
Noise

A replacement that works electrically may still affect measurement accuracy.

7. LM324N vs LM324N/NOPB

Another common confusion is:

LM324N vs LM324N/NOPB

The difference is related to manufacturing and compliance options.

NOPB generally indicates a lead-free/RoHS-compliant version.

Texas Instruments lists LM324N/NOPB as an active LM324-N device with a PDIP package and 14 pins.

For new production designs, engineers usually prefer RoHS-compliant versions.

The electrical function remains the same for normal applications.

8. LM324 vs LM324A vs LM324B

When selecting an LM324 replacement, engineers often encounter LM324A and LM324B.

These are not simply package changes.

They represent improved versions.

LM324

The standard version.

Suitable for:

General analog circuits
Low-cost designs
Non-critical applications
LM324A

An improved version.

Typical improvements:

Better offset performance
Improved accuracy

Useful when the circuit requires better DC performance.

LM324B

A newer generation device.

Compared with classic LM324, LM324B improves several specifications.

TI describes LM324B as a next-generation version with improved specifications while maintaining compatibility with LM324 designs.

For new designs, LM324B is often a better choice than the original LM324.

9. Common Mistakes When Replacing LM324 and LM324N

From an engineering and sourcing perspective, several mistakes happen frequently.

Mistake 1: Only Checking the Part Number

Two components may both say “LM324” but come from different manufacturers.

Always check:

Manufacturer
Full ordering code
Datasheet
Mistake 2: Ignoring Package Differences

Example:

LM324N:

PDIP
Through-hole

LM324D:

SOIC
Surface mount

They may perform similarly but cannot directly replace each other on a PCB.

Mistake 3: Using LM324 for High-Speed Applications

LM324 is reliable, but it is an old general-purpose amplifier.

Do not use it when you need:

MHz-level precision amplification
Low noise
High-speed ADC driving
Rail-to-rail performance

Modern op amps may provide much better performance.

10. When Should You Still Choose LM324N?

Despite being an old device, LM324N remains useful.

Choose LM324N when:

You are repairing legacy equipment

Many older systems were designed around DIP packages.

You need a low-cost analog solution

LM324 is inexpensive and widely available.

Your circuit requirements are moderate

For:

Simple amplification
Voltage buffering
Control circuits

LM324N is still a practical choice.

11. When Should You Choose a Modern Alternative?

For new designs, consider newer op amps if you need:

Lower power consumption
Higher bandwidth
Lower offset voltage
Rail-to-rail input/output
Better noise performance

Examples of newer alternatives may include:

Low-power CMOS op amps
Precision amplifiers
High-speed amplifiers

The correct choice depends on your circuit requirements, not simply the availability of a replacement.

12. Final Conclusion: LM324 vs LM324N

The difference between LM324 and LM324N is often misunderstood.

The key points are:

LM324 is the op-amp family name.
LM324N is a specific LM324 ordering version.
The “N” mainly identifies the PDIP package.
They are generally functionally equivalent.
Replacement is usually possible if package and specifications match.
Always check the exact datasheet before production use.

For engineers working on legacy electronics, LM324N remains one of the most practical and reliable general-purpose op amps available.

For new designs, however, it is worth evaluating newer LM324 variants or modern op amps that provide better performance.

A good engineering rule is:

Never select a replacement only because the part number looks similar. Confirm the electrical specifications, package, and application requirements first.

That small step can prevent unexpected failures during testing and production.

Micron Launches Hiroshima Fab Expansion to Scale Advanced 1γ DRAM and HBM Production

AVAQ SEMICONDUCTOR — Mon, 06 Jul 2026 03:27:59 +0000

Micron Technology has officially broken ground on a major expansion of its Hiroshima manufacturing site in western Japan, marking a significant step in the company’s long-term strategy to expand advanced memory output for AI-driven workloads.

The project represents a total investment of approximately ¥1.5 trillion (around $9.3–9.6 billion) and will significantly increase production capacity for next-generation DRAM and high-bandwidth memory (HBM), both of which are critical components in AI accelerators used by leading data center and GPU platforms.

Focus on 1γ DRAM and HBM Scaling

At the core of the expansion is Micron’s push to scale 1γ (1-gamma) DRAM technology, an advanced process node designed to improve density and performance for AI and high-performance computing applications. The new facility will also support expanded production of HBM products, which are increasingly constrained by surging demand from AI training and inference workloads.

The Hiroshima site is already a key global DRAM manufacturing hub for Micron, and the expansion is intended to further strengthen its role as a leading-edge production center for next-generation memory technologies.

Construction Timeline and Equipment Ramp

According to the project schedule, construction will proceed in phases, with equipment installation expected in the second half of 2028 (2H28). This timeline aligns with the anticipated ramp of advanced AI chips requiring next-generation HBM capacity.

Commercial output is expected to begin around 2028, positioning the facility to support the next wave of AI server and accelerator deployments.

Government Support and Strategic Importance

Japan’s Ministry of Economy, Trade and Industry (METI) is expected to provide up to ¥500 billion in subsidies to support the expansion, continuing Japan’s broader strategy to strengthen domestic semiconductor manufacturing capabilities and attract strategic foreign investment.

The Hiroshima expansion also reflects a broader geopolitical and industrial trend: diversifying advanced memory production outside of concentrated manufacturing regions while reinforcing supply chains for AI infrastructure.

Purchase Micron Semiconductor Products from Avaq Semiconductor Co., Ltd.

Part of a Global Capacity Expansion Wave

Micron’s Hiroshima investment is part of a wider global capital expenditure push across the memory industry, as companies race to address structural shortages in HBM and advanced DRAM. With AI workloads driving unprecedented demand, leading memory suppliers are rapidly expanding capacity to avoid long-term supply bottlenecks.

Once operational, the expanded Hiroshima fab is expected to play a central role in Micron’s roadmap for next-generation memory technologies, particularly as the industry transitions toward more advanced AI architectures requiring higher bandwidth and efficiency.

Anthropic Explores Custom AI Chip Partnership With Samsung

AVAQ SEMICONDUCTOR — Sat, 04 Jul 2026 02:41:11 +0000

Artificial intelligence startup Anthropic is reportedly exploring the development of its own custom AI chip and has entered early-stage discussions with Samsung Electronics over a potential manufacturing partnership, according to multiple media reports. The initiative reflects a growing trend among leading AI companies seeking greater control over their computing infrastructure as demand for AI accelerators continues to rise.

Sources familiar with the discussions said the proposed collaboration could involve Samsung's advanced 2-nanometer foundry technology along with its advanced semiconductor packaging capabilities. However, the project remains in its early planning phase, and neither company has officially confirmed that an agreement has been reached.

The reported negotiations are also consistent with the existing partnership between Anthropic and Samsung Electronics. Anthropic identified Samsung Electronics, SK hynix, and Micron as its "strategic infrastructure partners" during its Series H funding round in May, recognizing their contributions in supplying memory, storage, and logic chips. Among these three companies, Samsung stands out as the only one operating a leading-edge foundry business, making it the most likely manufacturing partner should Anthropic move forward with developing its own AI chips.

Reports indicate that Anthropic has not yet finalized the intended purpose of the custom processor. The company is still evaluating how the chip would be integrated into its AI infrastructure, what workloads it would target, and the performance objectives it should achieve. These decisions are expected to shape the overall architecture and production timeline.

At present, Anthropic relies on a diversified hardware ecosystem to train and deploy its AI models, including processors supplied by Amazon, Google, and Nvidia. In response to questions about the reported Samsung discussions, Anthropic stated that its computing strategy will continue to leverage multiple hardware platforms but declined to comment on the reported project.

For Samsung, securing a potential partnership with Anthropic would represent another opportunity to strengthen its contract chip manufacturing business. The company has been actively seeking additional customers for its leading-edge semiconductor manufacturing as it competes with other major foundries in the rapidly expanding AI semiconductor market.

The reported discussions also highlight a broader industry trend. As AI model developers face increasing demand for computing power, many are investing in proprietary silicon designed specifically for their workloads. Custom AI chips can improve performance, optimize power efficiency, reduce long-term infrastructure costs, and lessen dependence on commercially available GPUs.

While the Samsung-Anthropic talks remain preliminary, the potential collaboration underscores the growing importance of close partnerships between AI software companies and advanced semiconductor manufacturers. If the project moves forward, it could further expand Samsung's presence in the AI chip supply chain while giving Anthropic greater flexibility in optimizing the hardware that powers future generations of its AI models.

(Source: Bloomberg, Trendforce, The Information)

Micron Q3 Earnings Are Coming: Is the AI Memory Supercycle Still Just Getting Started?

AVAQ SEMICONDUCTOR — Mon, 22 Jun 2026 03:55:58 +0000

Micron will report its fiscal Q3 results on June 24, and many investors view this as one of the most important semiconductor earnings releases of the year. Expectations are extremely high, with analysts forecasting revenue around $34.8 billion, EPS near $19.8, and gross margins approaching 81%. HBM demand and AI infrastructure spending continue to drive bullish sentiment across the memory industry.

Several key topics will likely determine how the market reacts:

Can Micron maintain gross margins above 80%, or are we already near peak profitability?
How much progress has Micron made on HBM4 and HBM4E compared with Samsung and SK hynix?
Will HBM remain supply-constrained through 2027, or will aggressive capacity expansion eventually lead to oversupply?
Is Micron's rapidly increasing capital expenditure a necessary investment, or could it become a future drag on earnings?

One interesting point is that Micron previously indicated that some non-HBM memory products were generating higher margins than HBM itself. At the same time, DRAM pricing remains strong, while HBM capacity is reportedly sold out through 2026.

Personally, I think the biggest question is whether AI memory demand has fundamentally changed the traditional memory cycle. Historically, memory has always been cyclical, but many analysts now argue that HBM is becoming a structurally constrained product rather than a commodity. Others believe today's massive capital spending will eventually create the next oversupply problem.

Questions for the community:

Do you believe HBM demand can stay strong through 2027–2028?
Which company is best positioned for the HBM4/HBM4E generation: Micron, Samsung, or SK hynix?
Is Micron's current margin level sustainable?
Are we witnessing a true AI memory supercycle, or simply another memory boom that will eventually peak?

Interested to hear opinions from engineers, memory specialists, and anyone following the AI infrastructure market.