$NVDA and Cisco just dropped something big for AI infrastructure—pluggable optics that can scale AI factories.
The USFP module is the first to hit 400Gb/s per lane. That's 4x the bandwidth density of current optics.
This matters because AI training clusters are bottlenecked by data movement. More bandwidth per lane means tighter, faster networks without ripping out entire racks.
Nvidia keeps building the picks and shovels. While everyone watches GPU sales, the real moat is in the ecosystem—networking, software, and now optical interconnects that make massive compute actually work at scale.
If you're long $NVDA, this is the kind of quiet infrastructure win that compounds over years.
The USFP module is the first to hit 400Gb/s per lane. That's 4x the bandwidth density of current optics.
This matters because AI training clusters are bottlenecked by data movement. More bandwidth per lane means tighter, faster networks without ripping out entire racks.
Nvidia keeps building the picks and shovels. While everyone watches GPU sales, the real moat is in the ecosystem—networking, software, and now optical interconnects that make massive compute actually work at scale.
If you're long $NVDA, this is the kind of quiet infrastructure win that compounds over years.