Eridu, a networking startup building infrastructure specifically for AI workloads, has raised more than $200 million in funding to take on the $200 billion AI networking market. The Series A round was oversubscribed, led by Socratic Partners, investor John Doerr, Hudson River Trading, Capricorn Investment Group, and Matter Venture Partners.
Additional participants include Bosch Ventures, Eclipse Capital, Fusion Fund, Osage University Partners, SBVA, TDK Ventures, VentureTech Alliance, and Zelda Ventures, among others. The round adds to a broad base of backers that includes Black Opal Ventures, Catapult Ventures, Chamaeleon, Fathom Fund, Friends and Family Capital, Godfrey Capital, Hyperlink, Leslie Enterprises, MediaTek, Modi Ventures, Ohio Innovation, Open Field Capital, Perkins Enterprises, Pierre Lamond, Rice Management, Struck Capital, and Triple Point Capital, among others.
AI compute has been advancing fast. Networking has not. As hyperscalers, frontier labs, and neoclouds race to build larger AI data centers, the volume of data that must move between systems is growing exponentially. And new AI architectures are making that pressure worse, not better.
What exists today, from both established vendors and newer entrants, amounts to incremental upgrades on the same underlying design. That path leads to more latency, higher power consumption, greater cooling demands, and rising costs. According to Eridu, the gap has grown wide enough that patching the old architecture is no longer a viable answer.
"Billions of dollars of investment in AI data centers are being wasted because of the network wall," said Drew Perkins, CEO and founder of Eridu. "The plodding pace of improvement promised by the existing industry vendors is simply inadequate to solve the problem. Even new companies and solutions promising higher performance are in fact still subscale."
Eridu's approach starts from a blank page. The company has developed a new network switch designed specifically around the demands of AI, rather than adapting general-purpose hardware to the task.
The platform delivers fewer network tiers, lower latency and jitter, single-hop scale-up domains supporting thousands of GPUs, scale-out domains reaching millions of GPUs, and savings of up to 40 percent in capital expenditure and up to 70 percent in networking power consumption.
"Today's cloud networking gear is clearly insufficient for the bandwidth demands of accelerated compute," said Dylan Patel, founder and CEO of SemiAnalysis. "As larger MoE models and trends like disaggregation of prefill and decode emerge, demands for network bandwidth and scale are only accelerating. Eridu is the first company I've seen with the team and vision to deliver the next level of interconnect scale required to meet the insatiable demands of accelerated compute."
The capital raised will be used to complete development of Eridu's solution and bring it to market. The company is targeting hyperscalers first, given the scale demands at that tier, but the platform is designed to serve a broader range of operators including neoclouds, sovereign clouds, and large enterprise AI data centers.
Eridu was founded by Drew Perkins, who brings a background spanning semiconductor and networking companies. The founding team includes experienced operators with prior stints at Lightera and Infinera, both of which produced significant commercial outcomes in the networking space.
The company's thesis is direct: the industry has been adapting old tools for new problems, and that approach has run out of road. Eridu's design addresses constraints across silicon, packaging, systems, and optics simultaneously, rather than optimizing one layer at a time.
The round drew interest from a wide range of investors, reflecting how seriously the AI infrastructure gap is now taken across venture and strategic capital.
"The disruptive demands of AI create an urgent need for completely re-thinking high speed interconnect and packet switching," said Gregory Waters, managing partner at Socratic Partners. "Eridu's novel architecture dramatically improves throughput and is a platform that nearly all next generation AI will depend on."
