BLOG & NEWS - Thinking in public.
Posts on AI infrastructure, wireless networking, and what modern compute environments actually need from the physical layer. Press coverage and announcements as they happen.
The Physical Cost of Connecting an AI Rack
An AI rack like NVIDIA’s NVL72 now carries about 5,000 copper cables behind its trays. Why interconnect has become part of the machine, what a connection really costs, and where contactless links could earn a place.
Why is east-west traffic the real AI cluster bottleneck
MFU looks compelling on paper, then real inference traffic hits the cluster. Why all-to-all patterns on fixed cable topologies drive up p99 latency, and why the fix starts at the physical layer.