Samsung Unveils Gen 10 V-NAND and 256TB QLC Flagship at GMIF2026

Samsung has presented the industry's first V-NAND with more than 400 layers, delivering 58% higher bit density, 33% faster I/O and 25% lower power consumption than the previous V9 generation. Kevin Yoon, CVP and CTO of Samsung Memory China at Samsung Electronics, showed the stack at GMIF2026 alongside the PM1763 SSD, the new BM1773, and zNAND-O. Gen 10 uses advanced bonding: memory cells and peripheral circuit wafers are manufactured separately, then bonded vertically for high-density stacking.

The BM1773 is new and serves as Samsung's capacity flagship. Built on V9 QLC, it spans 16TB to 256TB and is billed as "the world's smallest 256TB SSD." zNAND-O, a next-generation NAND architecture for on-device AI, is in development in 4-layer and 8-layer versions. It is co-packaged with the MPU and connected through the UCIe interface, delivering 10x the bit density of DRAM and 7x the read bandwidth and read energy efficiency of NAND flash. In a simulation using Moonshot AI's Kimi K2-1T model, an NPU-optimized architecture with zNAND-O cut memory cost to one quarter at equal performance and fit a model nearly twice as large.

Can Kioxia pass 1,000 stacked word lines?

Kioxia says generative AI data growth demands higher 3D flash capacity and density, but stacking more word lines worsens cell current degradation, wafer warpage and block size, making more than 1,000 stacked word lines difficult. Its answer is the Multi-Stacked cell Array CMOS directly Bonded to Array (MSA-CBA) architecture, which sequentially stacks two or more memory cell array wafers on a CMOS wafer using wafer-to-wafer Cu direct bonding. Limiting word lines per array wafer structurally mitigates degradation and warpage, while sharing the word line switch circuit among array wafers reduces block size and power. Kioxia demonstrated the world's first QLC operation in MSA-CBA on prototype devices, with clearly separated threshold voltage distributions and sufficient voltage margins across arrays and the overall structure.

What is new at IBM FlashSystem?

At IBM TechXchange 2026, IBM announced FlashSystem 5600, 7600 and 9600 can now alert administrators when data theft starts, before attackers encrypt data for ransom. FlashSystem already alerts within a minute of encryption; the new detection runs validation tests to confirm suspected exfiltration in under 30 minutes, with a false-positive rate below 1%. Attackers typically spend 11-14 days inside a network before deploying ransomware, so exfiltration alerts arrive days ahead of a typical encryption-based attack.

FlashSystem.ai, introduced in February as part of the refreshed portfolio, adds advisory capability so routine provisioning takes up to 90% less effort. Admins can ask questions, receive contextual answers with next-best-action recommendations, and still approve decisions within policies they set.

Where is cooling capacity coming from?

Hitachi Vantara is piloting one of the world's largest Phase Change Material installations at its Netherlands data centre with Dutch thermal specialist Duraflow. The deployment delivers 240 kW of cooling using 5-8 kW of electricity, against an estimated 60-80 kW for conventional cooling, and requires no water. Duraflow's technology absorbs, stores and releases heat, and can cut thermal management electricity by up to 90% depending on application, freeing power for AI within the existing connection.

What changed in enterprise data and AI software?

• Chelsio launched its Seventh-Generation AI Interconnect Platform on the Unified Wire architecture: S7 SmartNICs, T7 Storage Controllers offloading NVMe/TCP, NVMe-oF, iSCSI and RDMA, and T7 DPUs with native 400GbE and Unified RDMA (iWARP and RoCEv2), compatible with prior generations.
• CloudCasa by Catalogic added Dell PowerStore to its DR replication providers alongside HPE Alletra Storage MP B10000, HPE Alletra 9000 and SUSE Storage; it also shipped MCP server support and a preview of Changed Block Tracking for VM backups.
• Adeptia released Automate 5.3, adding native MCP capabilities, a redesigned AI Mapping experience that cuts partner onboarding from days or weeks to minutes, Bring Your Own LLM support and Observe monitoring.
• Pinecone released pg_pinecone, an Apache 2.0 open-source Postgres extension that indexes existing vector columns, applies filters during search and keeps data inside Postgres.
• Qdrant 1.20 ranks text with BM25 directly on the full-text index, combining keyword and semantic search in a single query with no sparse encoding pipeline.
• Starburst announced limited GA of Torin, a GPU-native, Trino-based engine built on Nvidia cuDF, delivering up to 7x faster TPC-H queries on Blackwell GPUs, with one query at 11.4x; on ClickBench an L4 GPU delivered 3.4x CPU throughput at similar hourly cost.
• Zadara achieved Nvidia AI Cloud Ready validation for its sovereign multi-tenant AI cloud, tested across HGX H100 and GB200 NVL72 systems with Quantum InfiniBand and Spectrum-X Ethernet.
• Supermicro published a solution brief pairing IBM Storage Ceph with GrandTwin servers, NVMe storage, 100GbE networking and Intel Xeon 6740E processors.
• N-able expanded its SentinelOne partnership beyond EDR to AI, cloud, identities and managed response, added FedRAMP delivery, and launched AI-accelerated Vulnerability and Patch Management that cuts patching windows from days to minutes.
• Nasuni engaged AHEAD as its first partner for Professional Services, covering architecture design, data migration and cutover.
• Smartbird partnered with WEKA, combining WEKA NeuralMesh with Smartbird's dedicated AI clusters, launched in June 2026 as managed services on-premises or in colocation.
• ActiveViam is collaborating with Databricks on validated, real-time risk analytics built on the Databricks Data + AI Platform for financial institutions deploying AI agents on governed data.

UNSW and the people moves

UNSW Sydney will replace 11 legacy storage systems after an 18-month evaluation, combining Arcitecta MediaFlux with DDN's EXAScaler Lustre parallel file system under its Research Data Experience program. Arcitecta built an EXAScaler connector capturing file-system metadata for Mediaflux, with tiering between on-campus flash caches, lower-cost HDDs off-campus, and cloud support. The project began September 1, 2026, equipment arrives early 2027, and the system goes live later in 2027.

Hammerspace hired John McKillop as VP of Americas Sales after senior roles including Quantum Corporation; predecessor Tom Whaley left in May for NetApp. Chief Business Officer Tony Asaro said: "We've experienced phenomenal growth this year. We are excited to bring John onto the team to help us increase our momentum in the market." McKillop said his focus is scaling the sales organization and expanding partnerships to help customers realize the value of AI anywhere.

Danish startup KineDB raised an €850K pre-seed round led by Superhero Capital to build a general-purpose database running in the cloud or offline on local devices, targeting sectors without reliable connectivity. Seagate, meanwhile, claimed its Mozaic HAMR drives are the first and only SMR drives on Broadcom's approved HBA vendor list, though a Broadcom Tech Doc lists both WD and Seagate drives; Seagate confirmed "Microsoft Azure Storage has completed qualification of the Broadcom HBA with Seagate HAMR + SMR technology" at hyperscale, and Chip Dispatch has asked Seagate to reconcile the contradiction. The coming quarters will show whether bonded-array NAND and 400-layer stacking translate into shipped capacity before rival HAMR and QLC roadmaps converge on the same density targets.