The networking stories enterprise IT leaders actually read.
Volume 342, Issue 1 · 7 articles from 2 issues
NVSwitch echoes InfiniBand's HPC dominance, but Ethernet-based UALink and ESUN are circling the scale-up fabric.
The analysis positions NVIDIA's NVSwitch as playing the role in AI scale-up networking that InfiniBand played in high-performance computing. NVIDIA currently holds roughly 95 percent of GPU revenues and reinforces that position through proprietary NVSwitch licensing and lock-in.
Competing efforts including UALink and ESUN/SUE-T aim to open scale-up networking on commodity Ethernet infrastructure.
The piece leans on the old networking adage that Ethernet always wins because it steals every good idea from any fabric that can be broadly applied, suggesting the proprietary advantage is durable but not permanent.
Fiber availability is joining power and water as a hard constraint on where data centers can actually be built.
Fiber optic connectivity is emerging as a siting constraint on par with power and water availability.
Data center bandwidth demand rose roughly 330 percent in the early 2020s, driven by AI workloads that move enormous volumes between training and inference tiers.
Established tech hubs enjoy dense, competitive fiber routes, while emerging markets chasing data center investment often lack the connectivity to support it.
The article advises operators to weight fiber availability heavily during site selection, coordinate early with multiple carriers, and evaluate alternative connectivity paths offered through cloud providers such as AWS and Azure.
Zeus Kerravala argues Forrester's data center networking Wave scores AI fabrics against the wrong criteria.
Analyst Zeus Kerravala criticizes Forrester's latest Data Center Networking Wave for placing NVIDIA in the Contenders band, arguing the scoring framework applies legacy enterprise benchmarks to purpose-built AI infrastructure.
AI infrastructure support carried only a 2 percent weighting while traditional enterprise features NVIDIA never set out to deliver dominated the score.
He notes NVIDIA received a 1.0 in enterprise management types, a result driven by criteria designed for conventional data centers rather than high-performance AI fabrics.
His conclusion is that buyers need separate evaluation frameworks for traditional enterprise IT versus AI-optimized infrastructure.
Bluetooth, Wi-Fi, and UWB solve different problems; the right pick depends on power, data, and security needs.
Bluetooth Low Energy, Wi-Fi, and UWB each address distinct connectivity problems. Bluetooth suits low-power sensors that need months of battery life, Wi-Fi handles high-bandwidth streaming and cloud access, and UWB provides precise distance measurement for secure proximity applications.
The article's framing is that the real question is not which technology wins but which fits the job, requiring teams to weigh power consumption, data requirements, positioning accuracy, and security risk.
Many products layer all three, with UWB gaining ground in automotive digital keys and asset tracking.
Network management systems can now correlate siloed signals and auto-resolve problems, raising questions about how autonomous networks should become.
The article explores how AI-driven network management systems now correlate previously siloed signals, helping IT teams identify critical problems and, in some cases, resolve them automatically.
Drawing on analysts from Omdia and EY, it examines the open debate over how much of network management AI should eventually handle with less human intervention, and whether that makes networks themselves smarter or simply makes management tools smarter.
The piece frames autonomous network management as a spectrum rather than a single destination.
CommScope experts on how fiber infrastructure design determines AI deployment speed, resilience, and flexibility.
On a diagram an AI network is clean lines; on the data center floor it is tens of thousands of fibers that must reach the right GPU, work first time, and stay accessible.
CommScope's Ken Hall, Scott Eischens, and Rudy Musschebroeck break growth into scale-up within the rack, scale-out across the hall, and scale-across between buildings.
They argue for structured cabling and offsite integration and testing so cabinets arrive ready to click into place, since power and liquid cooling now crowd the rear of GPU cabinets.
Multiplane architectures, replaceable shuffle modules, and self-cleaning connectors address resilience, while co-packaged optics push fiber deeper inside switches. Sponsored content.
A buyer's guide for testing whether a network governance platform enforces rule lifecycle or merely reports on it.
Network security governance platforms should do more than inventory firewall rules; the real test is whether they help teams manage rules over time.
The guide sets out criteria across rule lifecycle automation, exception governance with enforced expiry, change simulation and conflict detection, exportable compliance evidence, and segmentation validation.
For every rule a platform should answer who owns it, when it was last reviewed, and what business purpose it serves. It recommends designing a proof of concept seeded with deliberate stale rules, expired exceptions, and shadow rules, and framing vendor questions to force demonstration rather than feature description.