Eridu exits stealth with $230M to fix AI's networking bottleneck
What's the deal? Saratoga, California-based Eridu has emerged from stealth with a $200 million Series A — bringing its total funding to $230 million — to build a new generation of networking infrastructure for AI data centres.
The startup was founded by Drew PerkinsDealroom has a profile for this one. Try Dealroom →, a networking veteran whose prior companies include Lightera NetworksDealroom has a profile for this one. Try Dealroom → (sold to Ciena for over $500 million) and Infinera (IPO'd, then sold to NokiaDealroom has a profile for this one. Try Dealroom → for $2.3 billion in 2025).
The oversubscribed round was led by Socratic PartnersDealroom has a profile for this one. Try Dealroom →, venture capitalist John DoerrDealroom has a profile for this one. Try Dealroom →, Hudson River TradingDealroom has a profile for this one. Try Dealroom →, Capricorn Investment GroupDealroom has a profile for this one. Try Dealroom →, and Matter Venture PartnersDealroom has a profile for this one. Try Dealroom →, among others.
Why now? The problem Eridu is targeting has become acute. AI model training and inference require data to move between thousands — and eventually millions — of GPUs at once, but the networking hardware inside data centres has failed to keep pace with advances in compute.
Perkins describes this as a "network wall": a bottleneck that, he argues, is wasting billions of dollars of AI infrastructure investment. Current solutions, including those from incumbents and newer entrants, continue to make incremental improvements to the same underlying architecture, Eridu says, which is insufficient for the scale hyperscalers and frontier labs now demand.
What could go wrong? Eridu's pitch rests on a clean-sheet hardware design — a high-risk, high-reward approach that requires flawless execution across silicon, packaging, systems, and optics.
Building new semiconductor infrastructure takes years and is notoriously difficult; the company says the funding will be used to complete development, meaning no product is yet shipping.
The AI infrastructure market is also fiercely competitive, with well-capitalised incumbents and several other well-funded startups all vying for the same data centre budgets.
The signal: Eridu's raise is the latest sign that AI infrastructure investment is moving beyond GPUs. The bottleneck is no longer just compute — it is the fabric connecting it. With Eridu targeting a $200 billion AI networking market and claiming its switch delivers up to 40% savings in capital expenditure and 70% savings in networking power, the startup is positioning itself at a critical — and so far underserved — layer of the AI stack.
The involvement of Hudson River Trading, one of the world's largest high-frequency trading firms, is a notable signal: it suggests that ultra-low-latency networking for AI is attracting interest well beyond the usual enterprise IT buyers.
Sources:
Eridu
Eridu's LinkedIn post
TechCrunch
SiliconANGLE
The Tech Buzz
Pulse 2.0
Image credit:
Eridu
J.V.
