Eridu’s $200M Debut Targets the AI Network Bottleneck

AI workloads are blowing past the limits of today’s network switches. Eridu just surfaced from stealth with a $200 million Series A led by Socratic Partners, John Doerr, Matter Venture Partners, and a roster of strategic investors to rebuild AI networking from the silicon up.

CEO Drew Perkins—best known for PPP, Lightera, Infinera, and Mojo Vision—and co-founder Omar Hassen are collapsing multi-tier optical gear into new switches with heavy on-chip networking so GPUs can communicate at full speed.

Instead of stacking more boxes every time capacity increases, Eridu integrates optics and switching directly on silicon. That reduces hop counts, lowers latency, and cuts power draw while improving reliability, because optics remain the least reliable component in current fabrics.

The gap is widening: GPU compute and memory bandwidth are scaling roughly 10x per year, while traditional switches crawl at 2–3x every two to three years. Perkins says every hyperscale builder he knows is looking for a new playbook so their GPU investments aren’t bottlenecked by the network.

The raise pulled in Hudson River Trading, Capricorn, TDK, Eclipse, Bosch Ventures, and TSMC’s VentureTech Alliance. Eridu’s hundred-person team is now building silicon, software-defined control, and integrated systems that slot into AI data centers where Arista or Cisco hardware sits today.

The takeaway for operators is simple: treat networking as a first-class AI design problem. Budget for fabrics that match GPU roadmaps, monitor power and latency like you monitor FLOPs, and keep an eye on Eridu’s roadmap as a template for the next generation of AI data centers.