News ticker – March 6

Jedec has published a brace of new standards covering universal flash storage and memory interfaces. JESD220H Universal Flash Storage 5.0 and JESD223G, UFS Host Controller Interface (UFSHCI) 5.0. UFS 5.0 is aimed at mobile applications, automotive and computing systems that demand high performance with low power consumption, while retaining compatibility with 4.x hardware. JEDEC said it will enable sequential read and write speeds of up to 10.8 GB/s, while a distinct power supply rail enables “noise isolation between PHY and memory subsystem.” Ultimately this should enable higher performance, greater reliability, and enhanced security for AI-enabled and other data-centric applications.

Filestreaming platform LucidLink has launched its LucidLink Connect service, which promises real-time data access and streaming from existing data stores without the need for migration. It said this is achieved while preserving existing native formats and workflows, and without disrupting architectures. Bandwidth usage is minimized through byte range streaming, with only the required portions of a file streamed. The service has been trialed with key partners and customers, and the firm has identified sectors such as media entertainment, creative, architecture, engineering and construction, and enterprise IT and AI and machine learning as key markets. It is initially available via LucidLink’s enterprise plans but will soon be available via AWS Marketplace.

Data silicon vendor Marvell turned in record results this week as it continued to surf the datacenter investment wave. Net revenues for the fourth quarter ending January 31 came in at $2.22 billion, compared to $1.8 billion a year earlier. This spawned net income of $396.1 million, a big jump on the previous year’s $202.2 million. For the full year, revenues were $8.2 billion, up from $5.8 billion, with net income of $2.7 billion, compared to a loss of $885 million for fiscal 2025. Matt Murphy, Marvell’s Chairman and CEO. said “We expect year-over-year revenue growth to accelerate each quarter in fiscal 2027, driven by continued strength in our data center business, with bookings continuing to grow at a record pace.”

Nimbus Data has debuted its FlashMax platform, which it describes as a multiprotocol all-flash platform engineered for modern datacenters. It uses industry standard NVMe SSDs, and builds on the vendor’s existing FlashRack platform, adding PCIe-based expansion, rack-level resiliency, and next generation data reduction “all within a single namespace.” That single namespace can support 20 PB raw capacity (up to 100 PB effective). It supports NVMe-oF (TCP and RoCEv2), Fiber Channel, iSCSI, NFS, SMB, and S3, with up to 400G ethernet and 64G fiber channel connectivity. 

Huawei used MWC in Barcelona this week to unwrap its AI Data Platform, which it is pitching at enterprises looking – and failing - to adopt AI agents at scale. The platform integrates knowledge base, KV cache, and memory bank elements, coordinated by the vendor’s Unified Cache Manager (UCM). Huawei claimed the knowledge platform could deliver retrieval accuracy of more than 95 percent. The vendor also debuted its latest SuperPoD products. Top of the list is the Atlas 950 SuperPoD, which supports 64 NPUs per cabinet and can scale up to 8,192 NPUs. Meanwhile, the Atlas 850E supports flexible deployment in standard air-cooled data centers, scaling from 8 to 1,024 NPUs, and is aimed at cluster-level inference.

Google has announced a partnership with DigitalRoute to help solve telcos’ “broken data” problem and enable them to use AI agents to build “self-healing, self-optimizing networks capable of predictive maintenance.” In a blog post, Google said telcos were mired in a “data swamp" with “massive amounts of unrefined information that AI models cannot use effectively.” The partnership will use DigitalRoute’s Usage Engine Private Edition on Google Kubernetes Engine (GKE), to underpin “a reusable pipeline that scales dynamically across edge and cloud locations.” This in turn will provide the foundation for rapid anomaly detection, automated root cause analysis, and predictive maintenance.