Eridu Raises $200M to Build Next-Gen AI Networking

A new startup called Eridu is stepping out of stealth mode with a massive funding boost and a big vision for the future of AI infrastructure. The company, led by veteran tech entrepreneur Drew Perkins, announced Tuesday that it has secured an oversubscribed $200 million Series A funding round, pushing its total funding to $230 million.

The round was led by Socratic Partners, with participation from venture capital legend John Doerr, Matter Venture Partners, and several other investors. The strong interest from investors highlights how critical networking technology has become in the rapidly growing AI ecosystem.

A Veteran Returns With a New Idea

Perkins is no stranger to building successful technology companies. His career in networking dates back to the early days of the internet. In the 1980s, he helped create the Point-to-Point Protocol, commonly known as PPP, which became an essential component of the TCP/IP stack that powers the modern internet.

Over the decades, Perkins has co-founded several influential startups. One of them, Lightera Networks, was acquired by Ciena for more than $500 million in 1999. Later, he launched Infinera, which eventually went public and was acquired by Nokia in 2025 for $2.3 billion. His other ventures include Gainspeed and augmented-reality startup Mojo Vision.

But the idea for Eridu came from a conversation about artificial intelligence.

The ChatGPT Moment

In early 2023, Perkins attended a small conference where he spoke alongside Sam Altman, CEO of OpenAI. The discussion quickly turned to ChatGPT and the massive computing power required to run advanced AI models.

According to Perkins, Altman explained that systems like ChatGPT relied on huge clusters of GPUs — thousands at the time, and potentially millions in the future.

That insight sparked an important realization: the next big limitation in AI wouldn’t just be computing chips themselves, but how those chips communicate with each other.

Rethinking AI Networking

Soon after, Perkins teamed up with networking chip expert Omar Hassen, who previously worked on chip design for major industry players like Broadcom and Marvell Technology. Together, they founded Eridu in 2024.

Their mission is ambitious: rebuild networking infrastructure specifically for AI workloads.

Today’s data centers rely on multiple layers of networking hardware. When more capacity is needed, companies simply add more boxes. While this approach works, it also increases the number of “hops” that data must travel, which slows things down and raises power consumption.

Eridu plans to change that by designing new AI-focused chips that handle more networking tasks directly on the silicon itself. The company ultimately aims to sell complete systems that could replace traditional networking hardware from providers such as Arista Networks.

By shifting more functionality onto the chip, Eridu hopes to cut power usage, reduce costs, and improve reliability — particularly by minimizing optical connections, which are often the weakest point in large networks.

Riding the AI Data Center Boom

Perkins points out that AI hardware is evolving at a dramatic pace. GPU performance and memory bandwidth are improving roughly 10 times each year, while traditional networking equipment from companies like Cisco, Broadcom, and Marvell tends to improve only two to three times every few years.

That growing gap is exactly what Eridu wants to address.

The startup already has around 100 employees, and investor demand for the funding round was reportedly intense. Perkins declined to reveal the company’s valuation but hinted that it’s in line with other startups raising large Series A rounds. He also wouldn’t confirm whether Eridu has reached unicorn status.

Still, if the company succeeds in delivering a new generation of AI-optimized networking chips and systems, it could play a key role in what many believe will be the largest data center expansion in history.

And unlike many young startups, Eridu is built by founders with decades of experience in networking — something that could give it a major edge in the fast-moving AI infrastructure race.