An AI cluster runs at the speed of its worst connection. As inference pushes workloads across more sites, the network has moved to the front of the design conversation — and every layer of the stack is a candidate for the weak link. Read article

In the world of AI infrastructure, the network is often relegated to a budget line item to be minimized. However, EPS Global argues that this outdated view is a financial mistake: while the network accounts for only 10–15% of infrastructure spend, it acts as the primary gatekeeper for the revenue-generating capacity of the remaining 85% (the GPUs). Read article

As NeoClouds and sovereign operators race to build AI factories, the real risk isn't sourcing GPUs — it's assembling a fabric, optics, cooling, and software stack that works together from day one. This blog breaks down EPS Global's validated AI infrastructure stack, covering merchant silicon and switching from Edgecore, Celestica, and UfiSpace; network operating systems including SONiC, IP Infusion's OcNOS, Aviz Networks, and Hedgehog; fabric visibility and interconnect from Aria Networks and MaiaEdge; and the physical layer — optics, connectivity, power, and cooling — from Coherent, Amphenol, Proficium, and Vertiv. Learn why sourcing each layer separately creates integration risk, and how a single validated relationship lets operators building AI inference put their capital toward more GPUs and more network instead of more engineers. Read article

The AFBR-S50 is a high-end distance measurement sensor based on the indirect Time-of-Flight (iToF) principle. It packs everything needed for optical ranging into aCompact, laser-based distance sensing that keeps drones flying accurately when GPS can't — from precision altitude hold and indoor navigation to trigger sensing for interceptor platforms. single 12x12x8 mm module. Read article

1.6 Tbps in a single OSFP — and it terminates in four LC connectors. 🚀

Our System Engineer Welly is back in the EPS Global Switch Lab in Dublin with the latest episode of our Unboxing Open Networking series — this time getting hands-on with a state-of-the-art 1.6T OSFP transceiver built for the AI era.

In the blog, we cover: ✅ Up to 2 km reach over single-mode fiber ✅ A first-of-its-kind 4x LC connector design ✅ Full CMIS compliance with DDM telemetry to the host switch ✅ Support for both Ethernet and InfiniBand AI fabrics

Paired with Broadcom Tomahawk 6-based switches delivering 102.4 Tbps, this is one of the practical building blocks of the 1.6T era — and it's shipping today.

The jump to 800G networking is no longer a future concept—it’s a production reality. Powered by Broadcom’s Tomahawk 5 silicon, the Edgecore DCS560 delivers a staggering 51.2 Tbps of switching capacity in a compact 2RU form factor. Designed specifically for the explosive demands of AI/ML clusters and hyperscale data centers, this platform offers the ultra-low latency and cognitive routing required for modern workloads. With full support for SONiC (Software for Open Networking in the Cloud) and flexible port configurations, it provides a high-performance, vendor-neutral path to the next generation of connectivity. Read article