Storage manufacturers used FMS 2026 in Santa Clara to outline flash and memory designs aimed at feeding AI and high-performance computing systems, led by the release of the first High Bandwidth Flash specification through the Open Compute Project. The announcements point to a data-center hardware race centered on moving and storing much larger data sets with less power, though the technologies remain primarily geared toward enterprise and AI workloads rather than offering an immediate upgrade for blockchain networks.
SK hynix and SanDisk introduced the High Bandwidth Flash, or HBF, specification as an open framework for stacked flash memory. The initial standard supports eight-layer and 16-layer configurations, with up to 512GB of capacity per device. Its specified read bandwidth ranges from 0.4TB per second to 3.0TB per second, placing its top-end performance in territory more commonly associated with High Bandwidth Memory products used alongside AI accelerators.
HBF is designed to address a growing mismatch in data centers: processors and accelerators can calculate quickly, but systems often spend substantial time waiting for large data sets to move between storage and compute hardware. NAND flash has traditionally offered far higher capacity than HBM, though at much lower bandwidth. The proposed standard attempts to narrow that gap by stacking flash dies and creating wider, faster connections to host systems.
That approach could give server builders another memory tier between conventional enterprise SSDs and costly high-bandwidth DRAM. It may be particularly useful for AI inference, vector databases, retrieval systems, and other workloads that repeatedly access very large stores of data without requiring every byte to reside in expensive system memory.
Samsung pushes NAND stacking beyond 400 layers
Samsung Electronics presented its V10 BV-NAND design, which exceeds 400 layers and uses wafer bonding to stack memory structures. The company described the application of wafer bonding in this NAND architecture as an industry first.
NAND makers have been increasing layer counts for years to raise storage density and lower the cost per bit. As stacks become taller, manufacturing them becomes more difficult: etching channels through hundreds of microscopic layers, aligning structures, and maintaining production yields all create technical constraints. Wafer bonding divides parts of the memory structure across separate wafers before joining them, a method that could help manufacturers continue expanding layer counts without relying on a single increasingly complex fabrication process.
Samsung also previewed zHBM and zNAND-O as concept products. The company has not said that either product has entered mass production. Their appearance at FMS nevertheless shows that memory suppliers are exploring designs that connect NAND-based storage more closely to high-bandwidth memory architectures.
The distinction between demonstrations and shipping products matters for data-center buyers. A specification can influence future server design and supplier roadmaps well before equipment reaches commercial deployment, but adoption depends on controller development, platform compatibility, manufacturing economics, and demand from major cloud and enterprise operators.
SK hynix and Kioxia target efficiency and thermal limits
SK hynix introduced its 10th-generation 4D NAND, a 375-layer product. The company said the new generation delivers energy efficiency 2.5 times higher than its predecessor. Energy consumption has become a central engineering consideration as flash density, interface speeds, and AI-server rack power levels rise together.
Flash storage typically uses less power than spinning hard drives for many workloads, but the performance requirements of modern enterprise systems are raising the thermal burden on SSDs. Faster interfaces, denser NAND packages, and sustained data transfers can make cooling a limiting factor, particularly in tightly packed server racks.
Kioxia addressed that issue with its CM10 enterprise SSD, which combines PCIe 6.0 connectivity, 332-layer BiCS10 NAND, and direct cold-plate liquid cooling. A cold plate transfers heat from a component into circulating liquid, rather than relying solely on airflow through a server chassis.
The CM10’s design reflects the pressure on conventional data-center cooling systems. PCIe 6.0 doubles the data-transfer rate of PCIe 5.0, creating a path for substantially faster storage connections, but it also increases the challenge of keeping components within operating temperatures during sustained workloads. Liquid-cooled SSDs could become more common in AI clusters and other high-density systems where air cooling is already stretched by GPUs and networking equipment.
Implications for crypto infrastructure remain indirect
The FMS announcements have clearer implications for AI data centers than for public blockchain networks. Most major blockchains do not depend on ultra-fast enterprise NAND in the same way that AI training or inference systems do. Bitcoin mining is dominated by specialized ASIC hardware, while Ethereum and many proof-of-stake networks rely more on distributed validators, networking, and execution-client performance than on the kind of storage bandwidth described in the HBF specification.
Some crypto-adjacent infrastructure could benefit over time. Blockchain analytics platforms, exchange surveillance systems, archival node providers, decentralized storage operators, zero-knowledge proof developers, and firms running large indexers all process growing volumes of historical and real-time chain data. Faster and more efficient storage could reduce infrastructure bottlenecks for those centralized or semi-centralized services.
The effect would depend on commercial availability and pricing. The FMS products range from a newly published specification to early product designs and previews, rather than a single wave of hardware entering every data center at once. Enterprise operators will also weigh their needs against the premium costs of advanced NAND, PCIe 6.0 platforms, liquid cooling equipment, and supporting server architecture.
The more immediate message from Santa Clara is that storage is becoming an active part of the AI infrastructure race rather than a passive repository for data. By pushing flash bandwidth, NAND density, and thermal design forward together, suppliers are positioning storage systems to handle workloads that can no longer rely on conventional SSD designs alone.
To explore how tokenized infrastructure reshapes data and payments, read this deep-dive analysis next.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
