Why Memory and Storage Define the Next Decade

Overview: Memory and Storage as the Foundation of Digital Civilization

From ChatGPT to autonomous driving, from edge computing to quantum communication, every disruptive technology depends on high-speed reading and writing, massive storage, and real-time data processing. Memory and storage are the land and the blood of the digital world.

In the next decade, whoever holds the technological high ground of memory and storage will hold the lifeline of digital civilization.

1. Technology Dimension: Memory and Storage at a Historic Turning Point

1.1 The End of Moore's Law and the Rise of a New Paradigm

Over the past half century, Moore's Law has dominated the pace of development in the semiconductor industry. Every 18 to 24 months, the number of transistors on a chip doubles and performance improves accordingly. However, as manufacturing processes approach physical limits at 2nm and 1.4nm nodes, the path of relying solely on shrinking transistor size to improve performance is approaching its ceiling.

In this context, innovation in memory and storage has become the core path to breaking through performance bottlenecks. Computing architecture is shifting from being compute-centric to being data-centric. No matter how fast a processor is, if data cannot be supplied in a timely manner, performance is still limited. This is the well-known Memory Wall problem.

Solving the memory wall will become one of the most important challenges for the semiconductor industry in the next decade.

1.2 HBM: Redefining Bandwidth Boundaries

High Bandwidth Memory, or HBM, is one of the most revolutionary memory technologies in recent years. By vertically stacking multiple layers of DRAM chips and connecting them through through-silicon via, or TSV, technology, HBM achieves bandwidth that is tens of times higher than traditional DDR memory.

NVIDIA's H100, A100, and other AI acceleration chips all use HBM precisely because AI training requires processing massive matrix operations within very short timeframes, and traditional memory cannot meet the bandwidth requirements. The emergence of HBM has directly made it possible to train large-scale AI models.

In the next decade, HBM4, HBM4e, and even the next generation of stacked memory technologies will continue to evolve, with bandwidth expected to exceed multiple TB/s levels, completely reshaping computing architecture.

1.3 The 3D Revolution of NAND Flash

The storage field is also undergoing profound transformation. Traditional 2D NAND Flash is limited by its planar area, leaving limited room for density improvement. 3D NAND technology has increased the number of layers from 32 in the early days to over 200 today by vertically stacking storage cells, and some manufacturers have announced technology roadmaps targeting over 300 layers.

3D NAND not only brings increased capacity, but also a continuous decrease in cost per bit, making high-capacity solid-state drives affordable for the consumer market and driving exponential growth in data center storage density.

1.4 Competitive Landscape of Emerging Storage Technologies

In addition to mainstream DRAM and NAND Flash, various new storage technologies are competing for development:

Among these, Processing-in-Memory technology is particularly noteworthy. It breaks the paradigm of separating computation and storage in the traditional von Neumann architecture by embedding computing units directly into memory chips, fundamentally eliminating the latency and energy consumption of data movement. This technology roadmap is considered one of the most significant computing paradigm revolutions of the post-Moore era.

2. AI Dimension: The Data Hunger Crisis of Large Models

2.1 AI as the Biggest Driver of Memory and Storage Demand

Artificial intelligence, especially the large language models represented by GPT-4, Gemini, and Claude, is consuming memory and storage resources at an unprecedented rate.

Taking GPT-4 as an example, with an estimated parameter count of over one trillion, storing only the model weights requires several terabytes of space. Training a large model requires parallel computing on thousands of GPUs for several weeks, generating intermediate data, gradient information, and checkpoint files that place extreme demands on memory bandwidth and storage capacity.

2.2 Inference Explosion: New Memory Requirements for Edge AI

The value of AI lies not only in training, but also in inference, which means deploying trained models into practical applications. As AI applications migrate from the cloud to the edge, smartphones, automobiles, industrial equipment, and household appliances all need to run AI models locally.

This creates new requirements for low-power, high-bandwidth, and compact memory:

• Mobile AI chips require LPDDR5X memory with bandwidth exceeding 68 GB/s.
• Automotive AI systems require industrial-grade memory that can operate stably under extreme temperatures.
• Edge servers need to deliver inference performance close to cloud-level within limited power budgets.

In the next decade, billions of edge devices will run AI globally, creating unprecedented demands for memory and storage.

2.3 Data Flywheel: AI-Generated Content Accelerates Storage Demand

AI not only consumes data but also generates data at an astonishing speed. AI-generated images, videos, code, documents, and audio are becoming a major source of internet content.

It is estimated that by 2030, the annual volume of data generated globally will exceed 180 ZB, or zettabytes, with a significant proportion generated or processed by AI systems. All of this data needs to be stored, indexed, retrieved, and analyzed, placing sustained pressure on storage infrastructure.

3. Industry Dimension: Strategic Restructuring of the Memory and Storage Sector

3.1 Supply Chain Concentration: A Highly Monopolistic Oligopoly

The global memory and storage industry presents a highly concentrated oligopoly competition pattern.

This highly concentrated market structure means that any fluctuation in the production capacity of a major manufacturer will have a profound impact on the global electronics supply chain. The chip shortage crisis from 2021 to 2022, as well as the subsequent sharp drop in memory prices in 2023, have clearly revealed the cyclical and fragile nature of this industry.

3.2 Capital Intensity: A Billion-Dollar Arms Race

The memory and storage industry is one of the most capital-intensive manufacturing sectors in the world. Constructing an advanced DRAM or NAND wafer fab requires an investment scale of USD 20 billion to USD 30 billion or more.

In the next decade, major manufacturers have announced unprecedented expansion plans:

• Samsung has announced the construction of the world's largest semiconductor park in Pyeongtaek, South Korea, with a total investment of over USD 200 billion.
• Micron has announced investment of over USD 150 billion in Idaho and New York states in the United States.
• SK hynix is building an HBM packaging factory in Indiana, USA.

This capital arms race will profoundly reshape the geographical distribution of the global memory and storage industry.

3.3 Technology Ecosystem: Standards Competition and Ecosystem Lock-In

Competition in the memory and storage industry is not only about technology and capital, but also about standards and ecosystems.

The DDR, LPDDR, HBM, and other standards developed by JEDEC, the Solid State Technology Association, determine the interface specifications between memory chips and processors. Whoever leads standard-setting holds the power of industry discourse.

In recent years, NVIDIA, AMD, Intel, and other chip giants have collaborated closely with memory manufacturers to jointly define the next generation of memory interface standards. The vertical integration trend of combining chips with memory will further deepen the lock-in effect of the industrial ecosystem.

4. Geopolitical Dimension: Memory and Storage as National Strategic Assets

4.1 The Core Battlefield of Semiconductor Competition

Memory and storage chips have evolved from commercial products into national strategic assets. The export controls that the United States has imposed on China's semiconductor industry have listed advanced memory technology as a key restricted category.

In October 2022, the U.S. Department of Commerce issued chip export control regulations that explicitly restrict the export of DRAM below 18nm, NAND Flash above 128 layers, and related production equipment to China. This policy directly affected the pace of development in China's memory industry.

4.2 China's Path to Breakthrough

Faced with technological restrictions, China's memory industry is striving to catch up:

• YMTC: 232-layer 3D NAND has been mass-produced, with a technology level approaching international first-tier standards. However, expansion has been hindered by equipment restrictions.
• CXMT: Focused on DRAM research and development, has mass-produced DDR4, and is moving toward DDR5.
• National funding: China's National Natural Science Foundation and related government programs continue to inject funds to support the localization of the memory industry chain.

However, the technological barriers in the memory industry are extremely high, requiring not only advanced manufacturing processes but also high-end equipment such as lithography machines, etching machines, and thin-film deposition systems. With the export ban on ASML EUV lithography machines, the road to catching up in China's memory industry remains full of challenges.

4.3 Supply Chain Diversification: Strategic Layouts Across Nations

Major economies including the United States, Europe, Japan, and South Korea are actively promoting localization of the memory and storage industry:

• United States: The CHIPS and Science Act provides USD 52.7 billion in subsidies to attract companies such as Micron and TSMC to build factories in the United States.
• European Union: The European Chips Act aims to increase European chip production capacity to 20% of global output by 2030.
• Japan: Government subsidies support Kioxia and Western Digital to expand production in Japan, and TSMC has been invited to build wafer fabs in Japan.
• India: Actively attracting investment, with Micron building storage and packaging factories in India.

Supply chain security for memory and storage has become a core issue in the technology strategies of major nations.

5. Data Center Dimension: Paradigm Shift in Storage Infrastructure

5.1 Storage Revolution in Hyperscale Data Centers

Amazon Web Services, Microsoft Azure, Google Cloud, Alibaba Cloud, Tencent Cloud, and other hyperscale cloud service providers are the largest single consumer group for global memory and storage.

With the explosion of AI workloads, the storage architecture of data centers is undergoing profound changes:

• Full migration from HDD to SSD: NVMe SSDs have read and write speeds dozens of times faster than traditional HDDs, and the penetration rate of SSDs in data centers continues to increase.
• Refined storage tiering: HBM and DRAM for hot data, NVMe SSD for warm data, and high-capacity HDD or tape for cold data.
• Compute-storage separation architecture: Storage resource pooling and flexible scheduling are achieved through high-speed networks such as NVMe over Fabrics, or NVMe-oF.

5.2 CXL: A Revolutionary Interconnect Standard Breaking Memory Silos

CXL, or Compute Express Link, is one of the most important technical standards in the data center field in recent years. Based on the PCIe physical layer, it achieves high-speed and low-latency interconnection between CPUs, memory, and accelerators, enabling flexible sharing of memory resources among multiple processors.

The significance of CXL lies in:

• Breaking the limitation of one-to-one binding between memory and CPU in traditional servers.
• Enabling elastic expansion of memory capacity, with a single server supporting TB-level memory pools.
• Reducing the total cost of ownership, or TCO, of memory and improving resource utilization.

In the next decade, CXL will become the mainstream standard for data center memory architecture, profoundly changing the paradigm of server design.

5.3 Persistent Memory: Bridging the Gap Between DRAM and SSD

Although Intel Optane persistent memory has been discontinued, the concept of Storage-Class Memory, or SCM, that it pioneered continues to evolve. By combining non-volatile storage media such as PCM and ReRAM with DRAM interfaces, SCM can retain data after a power failure while providing access speeds close to DRAM.

This technology roadmap has revolutionary significance for applications such as databases, real-time analytics, and financial transactions that are extremely sensitive to latency.

6. Sustainability Dimension: The Urgent Proposition of Green Storage

6.1 Energy Crisis in Data Centers

Global data center power consumption accounts for approximately 1% to 2% of total global electricity consumption, and this proportion is increasing rapidly with the explosion of AI workloads. The International Energy Agency predicts that data center electricity consumption may double by 2030.

The energy consumption of memory and storage devices is an important component of data center power usage. In an AI training server, for example, the power consumption of HBM memory can account for 15% to 20% of overall system power consumption.

6.2 Strategic Value of Low-Power Memory Technology

In this context, the strategic value of low-power memory technology is becoming increasingly prominent:

• LPDDR5X: Compared to DDR5, it reduces power consumption by approximately 30% and is the preferred choice for mobile devices and edge AI applications.
• MRAM: Extremely low standby power consumption, suitable for IoT devices.
• 3D NAND energy efficiency improvement: Higher layer counts mean lower energy consumption per bit.

In the next decade, the energy efficiency ratio, measured as performance per watt, of memory and storage will become an equally important competitive dimension as raw performance.

6.3 Data Lifecycle Management: Storage for Sustainable Development

Data storage not only consumes energy but also involves the extraction and use of scarce mineral resources such as cobalt, lithium, and tantalum. As ESG principles deepen, the memory and storage industry faces growing pressure from regulators, investors, and consumers.

Extending the lifespan of storage devices, improving data compression rates, and optimizing data lifecycle management will become important means for enterprises to reduce their carbon footprint.

7. Social Dimension: How Memory and Storage Reshape Civilization

7.1 Personal Digital Sovereignty: Data Storage as Power

In the digital age, personal data is one of the most important assets. Whoever controls the storage of data holds digital power.

With the development of personal AI assistants, digital twins, brain-computer interfaces, and other technologies, the volume of data generated by individuals will grow exponentially. How to securely and privately store this data will become a major issue concerning individual freedom and dignity.

Decentralized storage technologies such as IPFS and Filecoin, as well as end-to-end AI storage solutions, are providing the technological foundation for personal digital sovereignty.

7.2 Healthcare: The Storage Revolution That Saves Lives

Genomics, medical imaging, wearable health monitoring, and other healthcare applications are generating massive amounts of data. A person's whole genome data is approximately 200GB, and full-body MRI images can reach tens of GB. Continuous health monitoring data can generate several GB per day.

Breakthroughs in storage technology directly determine the accessibility of precision medicine. When every person's complete health data can be securely stored and quickly analyzed, early detection and personalized treatment of diseases will become a reality.

7.3 Cultural Heritage Protection: Digital Preservation of Human Memory

Libraries, museums, and archives around the world are digitizing thousands of years of human cultural heritage. These digital assets require long-term and reliable storage media.

Long-life storage technologies such as DNA storage and quartz glass storage are providing possibilities for the permanent preservation of human civilization. Microsoft has successfully stored the code for the Super Mario game in DNA, which theoretically can be preserved for thousands of years.

7.4 Democratization of Education: Bridging the Information Divide

When large language models can run on local devices and massive educational resources can be stored on a small storage card, knowledge acquisition will no longer be limited by network connections or geographical location.

The widespread adoption of memory and storage technology is an important technical foundation for achieving educational democratization and bridging the digital divide.

8. Future Outlook: Key Trends for the Next Decade

8.1 Technology Convergence: Dissolving the Boundaries of Computing, Memory, and Storage

In the next decade, the boundary between computing and storage will become increasingly blurred. Technologies such as processing-in-memory, near-memory computing, and neuromorphic chips are breaking down the barriers of traditional architectures and moving toward deep integration of computing, memory, and storage.

8.2 Quantum Storage: The Future of Next-Generation Information Storage

The development of quantum computing has introduced new requirements for storage technology. Quantum memory can store quantum state information and is a key node in quantum communication networks. Although quantum storage is still in the laboratory stage, its potential value in cryptography and communication security is immeasurable.

8.3 Spatial Computing and XR: Storage Challenges for Immersive Experiences

The emergence of spatial computing devices such as Apple Vision Pro signals the arrival of the next generation of human-computer interaction paradigms. High-resolution and low-latency XR experiences require the integration of high-bandwidth, low-power memory and storage within very compact device spaces. This will become an important driving force for memory and storage technology innovation in the next decade.

Conclusion

Memory and storage are the cornerstone of digital civilization, the lifeblood of artificial intelligence, the strategic high ground of national competition, and the material foundation of human social progress.

From HBM to processing-in-memory, from 3D NAND to DNA storage, from data centers to edge devices, every breakthrough in memory and storage technology is redefining the relationship between humans and information, redefining the boundaries of technology, and rewriting the possibilities of civilization.

The next decade will not be defined solely by algorithms or computing power. It will be defined by memory and storage. Because no matter how technology evolves, data always needs to be stored and information always needs to be remembered. That is the eternal mission and enduring value of memory and storage.

In this silent revolution, every byte carries the weight of human civilization's progress.

Related Forum FAQ