Data Center News
-
Tower Semiconductor and Semtech are expanding dedicated SiGe manufacturing capacity for AI data-center optical and copper interconnects.
-
The Federal Energy Regulatory Commission has approved PJM Interconnection’s new Large Load Registry while rejecting most of the grid operator’s broader plan for curtailing new data centers and other large electricity users during capacity shortages. FERC’s October 9 order in docket ER26-3515 accepted PJM’s definition of a Large Load as an end-use customer with a…
-
A new open-access SoftwareX study comparing 12 data-center cooling topologies found that a direct-to-chip liquid-cooled design reached a PUE of 1.12 in a full-year Miami simulation, versus 1.35 for its comparable air-cooled configuration—a 17% reduction in PUE. Liquid cooling won on energy and carbon The DCOPTEST study used a physics-based Modelica testbed to run 8,760…
-
SK hynix unveiled MARS, a CXL-based memory-centric AI rack architecture designed to pool memory across CPUs and GPUs and reduce duplicated data movement.
-
Open Compute Project is evaluating high-temperature superconducting power infrastructure for future AI data centers, from grid interfaces to rack-level distribution.
-
BitcoinVersus.Tech reported in September 2025 on TeraWulf’s move from Bitcoin mining toward large-scale AI and HPC hosting. That strategy has now expanded into a direct 20-year lease with Anthropic. In a July 6 SEC filing, TeraWulf said Anthropic will lease about 401 MW of critical IT load at the Justified Data Campus in Hawesville, Kentucky.…
-
NVIDIA and Cisco announced on October 12, 2026, at the Open Compute Project Global Summit plans for Ultra-Dense Small Form Factor Pluggable, or USFP, a new optical-transceiver format designed for 400 Gb/s electrical signaling per lane and up to 12.8 Tb/s of overall capacity. The proposal targets AI-factory Ethernet networks that need more bandwidth without…
-
OCP says AI racks are on a path past 1 MW as the industry moves from 800 VDC toward 1,500 VDC and more tightly co-designed power, cooling and grid infrastructure.
-
Meta’s new IcePack rack uses 102.4 Tb/s IceCube switches, liquid cooling, Flattened Planar Fabric and MetaRoCE to scale open Ethernet networking for AI clusters.
-
Rittal and Bosch are bringing an 800V DC power sidecar, megawatt liquid cooling and OCP racks together for dense AI infrastructure.
-
Cisco’s new Silicon One G302 targets top-of-rack AI networking with 51.2 Tb/s capacity, 200G SerDes, 1.6T optics and direct-to-chip liquid cooling integration.
-
NVIDIA and Cisco plan a new USFP optical format for 400Gb/s-per-lane AI networks, with a path to 12.8Tb/s aggregate capacity and higher bandwidth density.
