AI Networking Startup Eridu Raises $200M Series A

Drew Perkins, a veteran architect of internet infrastructure, is re-entering the spotlight with the launch of Eridu, an AI networking startup that has officially exited stealth mode with an oversubscribed $200 million Series A funding round. The company, which has now secured $230 million in total capital, aims to solve the growing communication bottlenecks within massive AI data centers.

The investment round was backed by notable industry figures and firms, including Socratic Partners, John Doerr, and Matter Venture Partners. Other participants included Hudson River Trading, Capricorn Investment Group, SBVA, MediaTek, Bosch Ventures, TDK Ventures, Eclipse, and VentureTech Alliance, the investment arm of TSMC.

Rethinking Infrastructure for AI

Perkins, whose career spans from co-creating the Point-to-Point Protocol (PPP) in the 1980s to building successful ventures like Lightera Networks, Infinera, and Mojo Vision, found his latest inspiration during a 2023 conversation with OpenAI CEO Sam Altman. The discussion centered on the massive compute requirements driving the AI revolution.

“Sam told me that what enabled AI and ChatGPT was just enormous amounts of compute. At which time, I think he meant 4,000 GPUs, but now we’re talking about millions of GPUs,” Perkins explained. He concluded that the primary challenge for future AI progress is not just the availability of chips, but how these systems communicate with one another.

The Eridu Approach

Founded in 2024 alongside co-founder Omar Hassen—who brings deep expertise in chip design from companies like Broadcom and Marvell—Eridu is looking to overhaul the networking stack. The company is developing new silicon designed specifically to integrate networking functionality directly into AI chips.

The technical focus of Eridu involves:

• Reducing Latency: By minimizing the number of hops data must travel between chips.
• Efficiency: Replacing traditional tiered optical connections with advanced on-chip communication.
• Reliability: Integrating more functions into the chip itself to reduce reliance on external optics, which are often the most fragile components in a network.

Perkins notes that while GPU compute and memory bandwidth are scaling at a rate of roughly 10x per year, traditional data center switches from established players have historically improved at a much slower pace. Eridu plans to offer complete systems that serve as a modern alternative to the gear currently provided by companies like Arista Networks.

Market Momentum

The company’s funding process saw significant interest from investors, a trend Perkins attributed to the urgent need for infrastructure capable of supporting the largest data center expansion in history. While Perkins declined to comment on the company’s current valuation or whether it has reached unicorn status, he emphasized that the focus remains on delivering a high-performance, AI-friendly networking architecture.

With roughly 100 employees currently on staff, Eridu is positioning itself as a foundational player in the next generation of data center design, leveraging the team’s extensive background in networking and silicon manufacturing to address the scaling demands of the AI era.