The Data Center Engineer @DataCenterEngnr

OmniOn ships QODN Osprey 48V-to-12V bus converters up to 2.6 kW. More here: thedatacenterengineer.com/news/omnion-sh… OmniOn Power has added four second-generation Osprey bus converters aimed at AI data center and high-density compute platforms that use 48 V distributed power with a regulated 12 V intermediate bus. The new QODN modules are designed for 48/54 V-to-12 V conversion and keep the same modified DOSA quarter-brick footprint as the previous-generation QODE parts, while pushing higher power and higher stated efficiency.

7 benefits of ultrasonic welding for data center wire-to-terminal and busbar applications. More here: thedatacenterengineer.com/whitepapers/en… As AI data centers demand ever-higher power densities, electrical connections are becoming a critical factor for efficiency, thermal management, and system reliability. Ultrasonic welding helps manufacturers create high-strength, low-resistance wire-to-terminal and busbar connections while supporting scalable, high-volume production.

LiquidJet Diamond coldplate claims 35% more AI tokens per watt. More here: thedatacenterengineer.com/news/liquidjet… Frore Systems has introduced LiquidJet Diamond, a liquid-cooling coldplate for AI-focused data centers that adds an integrated diamond spreader aimed at GPU hotspot cooling. The company ties the design directly to higher “AI Tokens/Watt,” claiming up to a 35% increase versus traditional skived coldplates. LiquidJet Diamond builds on Frore’s LiquidJet coldplate, which uses what the company describes as a 3D “ultra short-loop” multi-stage coldplate architecture that can be customized to a GPU’s power map, including localized hotspots. Frore reports that LiquidJet (without the diamond spreader) delivers a 12 °C reduction in GPU die temperature, which it says translates to 25% higher Tokens/Watt. With the added diamond spreader, Frore reports an additional 10 °C reduction in hotspot die temperature, taking the claimed Tokens/Watt improvement to 35%.

Kioxia samples E1.L QLC SSDs for 1U hyperscale servers up to 30.72 TB. More here: thedatacenterengineer.com/news/kioxia-sa… Kioxia has introduced the KIOXIA LD4 Series, its first SSD in the E1.L form factor built with BiCS FLASH generation 8 QLC NAND. The drive targets hyperscale deployments looking to push more flash capacity into 1U server footprints, particularly for read-intensive storage tiers. The LD4 Series is NVMe 2.0e compliant and uses a PCIe 5.0 interface at x4 lanes, with stated compliance up to 16 GT/s per lane. It also supports the Open Compute Project (OCP) Datacenter NVMe SSD Specification v2.6, aligning the drive with the mechanical, thermal, and management expectations common in OCP-influenced server and storage designs.

PCIe 7 signal conditioning portfolio spans copper and optical links for AI racks. More here: thedatacenterengineer.com/news/pcie-7-si… Astera Labs has rolled out its Aries 7 PCIe 7 signal-conditioning portfolio, spanning copper and optical interconnect options aimed at high-bandwidth, multi-rack AI systems. The lineup covers retimers, redrivers, cable modules, optical drivers, and optical transimpedance amplifiers (TIAs), with management and diagnostics tied together under the company’s COSMOS software suite. Astera Labs says the Aries 7 portfolio is built to support PCIe 7 links “from chip-to-chip, box-to-box, and rack-to-rack,” including chip-down designs as well as smart cable modules and optics approaches such as near package optics (NPO) and linear pluggable optics (LPO). The company also frames Aries 7 as a single “toolbox” that can be applied where copper is practical and where optical reach is required.

Midea debuts power-cooling hyperconverged architecture for AI data centers. More here: thedatacenterengineer.com/news/midea-deb… Midea Building Technologies brought a set of AI data center cooling products and a power-cooling “hyperconverged” architecture to Data Center World Asia 2026 in Singapore, targeting high-density AI racks, liquid cooling deployments, tropical climates, and lower-carbon operation. The company’s “power-cooling hyperconverged architecture” combines power systems, energy storage, magnetic bearing cooling sources, and liquid cooling CDUs into a single infrastructure stack. MBT describes it as connecting “power supply—energy storage—cooling—liquid cooling distribution” from early-stage data center construction, with an aim of tighter coordination between electrical and thermal systems as rack power density rises. MBT also published a technical white paper tied to the architecture that covers planning, construction, and operations for AI data centers.

Infineon on protecting the 800 V rack: fast enough to save it, slow enough not to trip it. More here: thedatacenterengineer.com/interviews/pro… Short a bus in an industrial plant and the long cable feeding it buys you time. Its inductance slows the rise of the fault current, and a mechanical breaker that needs milliseconds to open gets its milliseconds. An AI rack offers no such favor. The cables are short, because every extra meter is parasitic inductance and I²R loss, and the bus sits at 400 V or 800 V DC. Put high voltage across almost no inductance and the current heads for a few kiloamps fast enough that customers ask for protection that reacts in microseconds. Give it milliseconds, the way a relay does, and Giuseppe Bernacchia’s estimate is hundreds of kiloamps. “And then you destroy everything.” Bernacchia is a distinguished engineer at Infineon and the system architect for its AI data center work, a brief that runs from the power entry of the IT rack down to the voltage regulator controllers feeding the GPUs. He has spent 25 years at the company. The last couple have gone to high-voltage systems, protection and DC/DC conversion alike, and what he has found is that the big end customers want the protection behavior they have always had at 12 V or 50 V, scaled up to 800 V. The physics does not cooperate.

Vertical power delivery: Infineon’s VRM density roadmap and what it does to your thermals. More here: thedatacenterengineer.com/interviews/ver… Put a voltage regulator module directly under an AI accelerator and it inherits the accelerator’s problems before it switches a single pulse. The motherboard beneath a multi-kilowatt ASIC already sits at 80° to 85° C with the cold plate running. The silicon inside the module is rated to 125° C, but customers want it held under 110° C, ideally closer to 100° C. That leaves about 20 degrees of thermal headroom to absorb the module’s own losses, which double when a 90 mm² footprint carries four phases instead of two. That is the bargain behind vertical power delivery (VPD), and Surendhar Somasundaram knows both sides of it. He is a principal engineer at Infineon in San Jose, working on the concept, definition, and development of the VRM power modules that sit in the shadow of the ASIC. The case for moving them there is strong: a vertical path through the motherboard cuts the power distribution network impedance to a quarter or a fifth of a lateral layout, which trims I²R loss and tightens the voltage gradient across a GPU die that can measure roughly 100 mm on a side.

Soft magnetic cores with 12–14 µm nanocrystalline ribbon target AI data centers. More here: thedatacenterengineer.com/news/soft-magn… Transmart is highlighting its soft magnetic core portfolio—nanocrystalline, amorphous, and grain-oriented silicon steel—for high-frequency power conversion and filtering use cases that show up in power infrastructure and AI data center power architectures. The company also said it is placing more emphasis on serving customers in North America. Transmart’s manufacturing setup includes two production facilities in China: one focused on silicon steel cores, current sensor cores, and instrument transformer cores, and another dedicated to nanocrystalline cores, amorphous cores, and common mode chokes. The facilities are certified to ISO 9001, ISO 14001, ISO 45001, and IATF 16949.

100 Gbps dedicated servers now available from 19 US, EU, and APAC locations. More here: thedatacenterengineer.com/news/100-gbps-… HostColor has expanded its 100 Gbps dedicated server hosting footprint, adding more data center locations where customers can deploy dedicated servers and cloud infrastructure with a 100 Gbps internet connection. The update broadens where operators can place high-bandwidth bare metal, especially for latency-sensitive deployments that need to sit close to users or upstream networks. In the US, HostColor now offers 100 Gbps-connected dedicated servers from Ashburn, Atlanta, Chicago, Dallas, Denver, Los Angeles, Miami, New York, and Seattle. Internationally, the company lists 100 Gbps server availability in Europe from Amsterdam, Frankfurt, London, Madrid, Paris, Prague, and Vienna, and in Asia from Hong Kong, Singapore, and Tokyo. In South America, São Paulo is listed as the company’s only point of presence for 100 Gbps servers.

Renesas samples 650V GaN in dual-side-cooled 8x8mm PQFN for 800V AI data centers. More here: thedatacenterengineer.com/news/renesas-s… Renesas has introduced the TP65H020G4PLSGBD, a 650 V GaN device in an 8 x 8 mm dual-side-cooled (DSC) PQFN package aimed at high-density power conversion for 800 V high-voltage DC (HVDC) AI data center architectures. The part targets space- and thermally constrained power stages in megawatt-scale racks, where intermediate bus converters and sidecar power shelves are getting harder to fit and harder to cool as power rises. The TP65H020G4PLSGBD is a D-mode device built on Renesas’ Gen IV Plus GaN architecture. Renesas lists 20 mΩ on-resistance for the device, and frames that as “one of the lowest” in the 650 V class. The company is sampling the part to major AI data center OEMs and ODMs.

Avery Dennison launches PSA tapes for BESS thermal and HV insulation. More here: thedatacenterengineer.com/news/avery-den… Avery Dennison has introduced a portfolio of pressure-sensitive adhesive (PSA) tapes designed for battery energy storage systems (BESS), targeting thermal runaway protection, high-voltage insulation, and no-cure assembly. The lineup includes functional bonding and protection tapes intended for use inside BESS modules, packs, and containerized units. The tapes are aimed at applications where PSAs bond and insulate cells, thermal barriers, and electrical connections. Avery Dennison describes the products as adhesives engineered for thermal runaway protection, high-voltage insulation, and instant bonding without cure time.

LG to build Virginia factory for AI data center air-cooled chillers. More here: thedatacenterengineer.com/news/lg-to-bui… LG Electronics is building its first US factory to manufacture chillers for AI data center cooling, putting new air-cooled chiller capacity in Virginia as demand rises across North America. The company said the facility is intended to improve supply responsiveness for regional customers while also serving as a base for further expansion. The Virginia site is planned as a 350,000-square-foot factory on a 43-acre site in Isle of Wight County. It will manufacture air-cooled chillers for AI data centers, with production scheduled to begin in the first half of 2027. Virginia Governor Abigail Spanberger publicly disclosed the project, and LG Electronics North America President and CEO Don Kwack said the site is meant to support a “stable, responsive supply system” for North America.

The hidden measurements that determine data center efficiency, cooling performance, and uptime. More here: thedatacenterengineer.com/whitepapers/da… AI workloads are pushing data center infrastructure to new limits. Cooling systems are becoming more complex, rack densities are increasing, and even small performance issues can impact efficiency, reliability, and uptime. Building, operating, and scaling modern data centers requires accurate insight into the systems that keep facilities running. Whether monitoring cooling water flow, tracking temperatures across critical assets, maintaining fluid quality, or validating backup power system readiness, measurement instrumentation provides the visibility needed to make informed decisions and respond quickly to changing conditions.

Microchip launches 65V energy-aware digital power monitors for 48V systems. More here: thedatacenterengineer.com/news/microchip… Microchip has introduced the PAC1761 and PAC1861 families of 65 V “energy-aware” digital power monitors aimed at 48 V power architectures used in AI/data center, networking, automotive, and industrial systems. The devices are designed to measure instantaneous voltage and current while also tracking accumulated energy over time, a combination intended to help 48 V platforms respond to changing load conditions. The PAC1761 and PAC1861 families support up to 65 V measurement headroom and include 75 V spike protection for transient survivability in 48 V environments. Microchip is offering the parts in two resolution tiers: the 12-bit PAC1761 and the 16-bit PAC1861.

Supermicro ships NVIDIA Vera Rubin NVL72 racks with liquid cooling stack. More here: thedatacenterengineer.com/news/supermicr… Supermicro is now shipping NVIDIA Vera Rubin NVL72 racks integrated with its Data Center Building Block Solutions (DCBBS) and direct liquid-cooling stack, DLC-2. For operators building out Rubin-class clusters, the immediate impact is schedule risk: pre-integrated, validated racks can reduce the amount of field-level bring-up work required to get a liquid-cooled AI pod online. The company’s NVIDIA Vera Rubin NVL72 rack is a single liquid-cooled rack built around 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs. Internally, it uses eighteen 1U compute trays, each configured with four Rubin GPUs and two Vera CPUs, connected through nine sixth-generation NVIDIA NVLink switch trays. Supermicro lists 216 TB/s of scale-up bandwidth for that NVLink fabric, plus 20.7 TB of HBM4 and up to 54 TB of LPDDR5X per rack.

Grid to core: the case for shifting to DC-grids. More here: thedatacenterengineer.com/interviews/gri… 50 V bus was fine when a rack drew a few tens of kilowatts. Push that rack into the hundreds of kilowatts and the arithmetic turns hostile: 5,000 to 10,000 A at 50 V, more current than any practical busbar can carry. The power supplies that used to sit inside the rack no longer fit there either, so they’ve been pushed out into a sidecar, which eats floor space that could hold revenue-generating racks. Fanny Björk has spent 23 years at Infineon and now runs application marketing for grid to load, or what the company brands grid to core: every conversion stage between the utility feed and the processor. She sees two shifts happening at once. Distribution voltage is climbing to 800 V DC, and the AC/DC conversion is leaving the white space altogether, moving into the gray space as a centralized transformer-rectifier unit or, eventually, a solid-state transformer that folds the line-frequency transformer and the rectifier into one box. Delete one galvanically isolated stage and you pick up roughly 1% in grid-to-load efficiency. Delete the sidecar and you get the floor back for racks that generate tokens.

Liquid leak detection flex assembly targets AI and HPC liquid cooling. More here: thedatacenterengineer.com/news/liquid-le… Flexible Circuit Technologies has developed a Liquid Leak Detection Flex Assembly, a custom sensing solution designed to detect coolant leaks in liquid-cooled AI servers and high-performance computing (HPC) systems. The flex circuit assembly is aimed at monitoring leak-prone points in server- and rack-level liquid-cooling hardware, including cold plates, manifolds, quick disconnects, coolant distribution units (CDUs), and related electronics. The Liquid Leak Detection Flex Assembly is built around a thin, lightweight flex-based form factor intended to fit around the geometry of a given cooling system. For operators and OEMs pushing direct-to-chip liquid cooling, the practical value is placement: leak sensing can be brought closer to potential failure points, even in space-constrained trays and rack environments where bulky sensors are hard to route.

AEWIN debuts AMD EPYC 9006 servers with two-phase direct liquid cooling. More here: thedatacenterengineer.com/news/aewin-deb… AEWIN detailed two new AMD EPYC 9006-based AI servers, the 1U BG18-A10710 and 5U BG58-A20810, both built around an integrated two-phase direct liquid cooling (2P DLC) design aimed at higher compute density and tighter thermal control in AI and HPC deployments. The 1U BG18-A10710 is built for high-density heterogeneous compute in a compact chassis. It supports a single AMD EPYC “Venice” SP7 CPU with up to 600 W TDP, two PCIe Gen6 SW FHFL accelerators, 16 DDR5 RDIMM or MRDIMM slots with memory speeds up to 12,800 MT/s, and eight E1.S Gen6 NVMe SSD bays. For management, it includes OCP DC-SCM2 system management via an ASPEED AST2700 BMC, and it’s configured with redundant 2,700 W 80 PLUS Titanium power supplies.

800G/1.6T Dual LC OptiShare switches let one tester serve up to 32 DUTs. More here: thedatacenterengineer.com/news/800g-1-6t… Echola Systems has introduced two OptiShare Dual LC switches aimed at AI network testing for data center GPU cluster fabrics: the OPX4D-SM and OPX8D-SM. The idea is straightforward—share a single optical test resource across multiple devices under test (DUTs) and switch connections remotely, instead of dedicating expensive gear per DUT or constantly repatching fiber by hand. The new OptiShare models support 800G and 1.6T optical connectivity using paired duplex LC interfaces in 2×FR4 configurations. Echola says both models provide plug-and-play connectivity to compatible transceivers over standard single-mode LC fiber cables, with no breakout cables required.