Samsung Unveils zHBM, zNAND-O & 400-Layer V10 BV-NAND at FMS 2026: The Future of AI Memory

At the Future of Memory and Storage (FMS) 2026 conference in Santa Clara, California, Samsung Electronics unveiled its most ambitious memory roadmap yet—centered on a radical shift from horizontal to vertical 3D memory architectures. The company introduced the industry's first concept models of zHBM and zNAND-O, revealed V10 BV-NAND with over 400 layers using wafer bonding technology, and showcased its latest AI memory portfolio including HBM4E, HBM5, LPDDR5X-PIM, and the PM1763 enterprise SSD. These announcements come at a critical moment when DDR5 memory prices have surged over 100% in 2026 due to AI-driven demand, making Samsung's push for next-generation 3D memory not just an innovation milestone, but a necessary evolution for the future of data centers and AI infrastructure.

zHBM: Vertically Stacked Memory for the AI Era

Redefining the Processor-Memory Relationship

Samsung's zHBM (vertical High Bandwidth Memory) represents a fundamental architectural shift. In current AI systems, HBM packages sit beside the processor, connected through silicon interposers and advanced 2.5D packaging. zHBM moves the memory stack directly above the AI accelerator, creating a true 3D integration that minimizes data travel distance.

This vertical stacking is not merely a relocation—it changes the physics of AI chip design. Horizontal packaging is constrained by physical area, interconnect length, and signal integrity. By stacking memory vertically, Samsung can provide more connections within a smaller footprint while substantially shortening the path data must travel between memory and compute cores.

Performance Targets

Samsung projects that a next-generation interface system based on zHBM could deliver:

• ~8x the performance of HBM5
• Over 10x the memory density of HBM5
• 3x greater energy efficiency
• More than 50% reduction in thermal resistance

These figures represent design targets for a future architecture, but they signal Samsung's long-term vision: as large language models continue to grow, simply adding more GPUs and more HBM will lead to unsustainable interconnect costs, power consumption, and system complexity. zHBM aims to consolidate workloads that currently require multiple chips into fewer, more efficient units.

Customizable Interlayer IP

A unique feature of zHBM is its support for customer-specific intellectual property within the interlayer between the memory stack and the AI accelerator. This layer could perform data compression, cache management, model scheduling, access control, and even selected computing tasks before data reaches the processing core. Future AI chips could consist of a computing layer, a data management layer, and a memory layer working together as a vertically integrated system.

zNAND-O: High-Performance NAND for Edge AI

Complementing zHBM, Samsung introduced zNAND-O, a next-generation high-performance NAND concept built on the company's V-NAND technology. The architecture comes in four-layer and eight-layer configurations and is designed to improve space efficiency, I/O performance, and latency.

While zHBM targets data center AI training and inference, zNAND-O is positioned for edge AI systems that process large datasets with minimal delay. By improving local storage performance, the architecture supports real-time applications where data cannot always be transferred to a centralized data center for processing—such as autonomous vehicles, industrial robotics, and smart city infrastructure.

V10 BV-NAND: Breaking the 400-Layer Barrier

Wafer Bonding Technology

Samsung unveiled V10 BV-NAND (Bonding V-NAND), featuring a new wafer bonding technology that the company claims is an industry first. The architecture stacks more than 400 layers of memory cells—the highest layer count Samsung has ever disclosed for its V-NAND line. This comes 13 years after Samsung introduced the industry's first V-NAND technology at the 2013 Flash Memory Summit.

The wafer bonding process allows Samsung to manufacture memory wafers separately and then bond them together, enabling higher layer counts without the yield challenges of traditional single-wafer etching. According to Samsung, V10 BV-NAND increases memory density by approximately 58% compared to V9, while also improving read, write, and I/O performance and reducing energy consumption.

Targeting AI Data Centers

V10 BV-NAND is targeted at high-capacity, high-performance storage for future AI systems. As AI training datasets grow exponentially—from terabytes to petabytes—storage density and speed become as critical as compute power. The PM1763 enterprise SSD, which entered mass production in July 2026 and uses related NAND technology, demonstrates Samsung's commitment to delivering end-to-end storage solutions for AI data centers.

AI Memory Roadmap: HBM4E to HBM5 and Beyond

HBM4 and HBM4E: Already in Production

Samsung's 3D vision is built on a foundation of near-term products already shipping. The company began the industry's first mass production of HBM4 in February 2026, using its 1c DRAM (10nm-class sixth-generation) and 4nm base die technologies. In May 2026, Samsung became the first to ship HBM4E samples to global customers, reinforcing its leadership in next-generation HBM despite earlier yield challenges.

HBM4 represents a major architectural change from HBM3E: the base die is manufactured on an advanced 4nm logic process rather than a DRAM process, enabling higher transistor density, smaller dimensions, and better area efficiency. Samsung's HBM4 achieves operating speeds up to 13 Gb/s per pin with 2,048 I/O pins and 3.3 TB/s bandwidth per 36GB stack.

LPDDR5X-PIM: Processing in Memory

Samsung also demonstrated LPDDR5X-PIM, which it describes as the industry's first LPDDR memory with processing-in-memory (PIM) technology. Instead of moving data between memory and processor for every operation, PIM performs selected data-processing tasks within the memory itself. This reduces data movement, improving efficiency and lowering power consumption for AI inference workloads at the edge.

Enterprise Storage: PM1763 and BM1773

For data centers, Samsung showcased the PM1763 (PCIe Gen6) and BM1773 enterprise SSDs. The PM1763 entered mass production in July 2026 and is designed to meet the massive storage requirements of AI training data, model repositories, and inference workloads. These products complement Samsung's memory roadmap by addressing the high-capacity storage bottleneck that often limits AI scaling.

The DDR5 Pricing Crisis: Why 3D Memory Matters Now

Memory Prices Have Doubled in 2026

Samsung's 3D memory push arrives during a severe supply crisis in conventional DRAM. In 2026, DDR5 memory prices have surged over 100% as manufacturers redirect wafer capacity toward higher-margin HBM for AI accelerators. TrendForce revised its Q1 2026 PC DRAM contract price forecast to a staggering 105–110% quarter-over-quarter increase—the steepest single-quarter surge on record.

Memory Segment	Q1 2026 Price Increase (QoQ)	Source
PC DRAM (DDR4 + DDR5)	+105–110%	TrendForce
Server DRAM	~+90%	TrendForce
Mobile DRAM (LPDDR5X)	+58–63%	TrendForce
NAND Flash (Contract)	+55–60%	TrendForce
Enterprise SSD	+53–58%	TrendForce

The HBM Priority Problem

The root cause is structural: HBM production consumes approximately three times the wafer capacity per gigabyte compared to standard DRAM. As AI data centers devour supply, manufacturers have permanently reallocated capacity toward HBM, leaving consumer and enterprise DDR5 markets severely undersupplied. Samsung, SK Hynix, and Micron have all confirmed that memory constraints will persist through 2026 and into 2027.

In this context, Samsung's 3D memory innovations are not just performance upgrades—they are capacity multipliers. By stacking memory vertically and processing data closer to storage, technologies like zHBM and zNAND-O could eventually reduce the total silicon area required for AI systems, freeing up manufacturing capacity across the industry.

Samsung's IDM Advantage: One-Stop AI Infrastructure

Samsung emphasized its unique position as the world's only integrated device manufacturer (IDM) with leading capabilities spanning memory, foundry, and advanced packaging. This vertical integration allows Samsung to offer one-stop turnkey solutions for AI semiconductor customers—from product design through mass production—shortening development cycles while optimizing performance and power efficiency.

At FMS 2026, Samsung's booth was designed to resemble an AI cloud server, displaying approximately 30 memory and storage technologies. The company won two Best of Show awards: the 3D & NAND Innovation Award for V10 Bonding V-NAND and the Next-Gen Memory Award for LPDDR5X-PIM.

What This Means for Consumers and Enterprises

Short-Term: No Relief for DDR5 Prices

Samsung's 3D memory roadmap is focused on 2027 and beyond. For consumers and enterprises looking to buy DDR5 in 2026, the outlook remains bleak. Analysts project prices will keep climbing through Q3 and Q4 2026, with Gartner forecasting a 47% DRAM price increase across the full year. Samsung's new Pyeongtaek production line will not begin mass production until 2028, meaning no significant capacity relief is imminent.

Long-Term: A Fundamental Shift

By 2028–2029, technologies like zHBM could redefine how AI servers are built. Instead of massive racks of GPUs each with side-mounted HBM, future systems might use vertically integrated compute-memory stacks that deliver higher performance in smaller footprints with lower power draw. For edge AI, zNAND-O could enable local processing of large models that currently require cloud connectivity.

Frequently Asked Questions (FAQ)

• What is Samsung's zHBM?
  • zHBM (vertical High Bandwidth Memory) is Samsung's concept for a next-generation memory architecture that stacks HBM directly above AI accelerators instead of beside them. This vertical integration aims to deliver ~8x the performance of HBM5, 10x the memory density, 3x energy efficiency, and over 50% reduction in thermal resistance.
• What is V10 BV-NAND?
  • V10 BV-NAND is Samsung's new Bonding V-NAND technology featuring over 400 layers—an industry first. Using wafer bonding technology, it increases memory density by approximately 58% over V9 while improving read, write, and I/O performance and reducing energy consumption.
• What is zNAND-O?
  • zNAND-O is a next-generation high-performance NAND concept from Samsung, built on V-NAND technology. It comes in four-layer and eight-layer configurations and is designed for edge AI systems that require high space efficiency, improved I/O performance, and low latency.
• When will zHBM and zNAND-O be available?
  • zHBM and zNAND-O are currently concept models unveiled at FMS 2026. Samsung has not announced specific mass production timelines, but these architectures represent the company's long-term vision for 3D memory beyond HBM5.
• Will Samsung's 3D memory lower DDR5 prices?
  • Not in the short term. Samsung's 3D memory solutions target AI data centers and high-performance computing, not consumer DDR5. In fact, DDR5 prices have surged over 100% in 2026 due to capacity reallocation toward HBM. New Samsung fab capacity is not expected until 2028.
• What is LPDDR5X-PIM?
  • LPDDR5X-PIM is Samsung's processing-in-memory technology that performs selected data-processing operations within the memory itself, rather than moving data back and forth between memory and processor. This reduces power consumption and improves efficiency for AI workloads.
• What is the PM1763?
  • The PM1763 is Samsung's PCIe Gen6 enterprise SSD that entered mass production in July 2026. It is designed to meet the high-capacity storage demands of AI data centers, complementing the company's HBM and DRAM roadmap.
• Why are memory prices so high in 2026?
  • Memory prices have surged because manufacturers are redirecting wafer capacity from consumer DDR5 to high-margin HBM for AI accelerators. HBM uses approximately 3x the wafer capacity per gigabyte compared to standard DRAM. TrendForce projects PC DRAM prices rose 105–110% quarter-over-quarter in Q1 2026, with continued increases expected through 2026 and into 2027.