AI Network Chip Startup Eridu Raises $200M Series A, Challenges Broadcom's AI Cluster Switch Monopoly

The AI infrastructure sector has seen another blockbuster funding round as mysterious networking chip startup Eridu emerged from stealth mode, announcing the completion of over $200 million in Series A funding. This previously little-known company is developing revolutionary high-radix switch systems specifically designed for large-scale AI cluster network interconnection, directly challenging Broadcom's long-standing monopoly in AI data center network switching.

Eridu's core technology focuses on solving the most critical bottleneck in current AI training and inference: network communication latency between GPUs. In traditional AI clusters, thousands of GPUs need to exchange massive volumes of data through network switches. The mainstream solution from Broadcom uses relatively low-radix switch architectures (typically 32-64 ports), requiring multi-tier switch topologies for large clusters, with each additional tier adding significant latency and complexity. Eridu's high-radix switch design supports over 512 ports per switch, dramatically reducing the number of network layers needed and cutting communication latency by 40-60%.

The $200 million funding round was led by top-tier infrastructure investment firms, with participation from several major cloud providers' strategic investment arms. Sources indicate the valuation exceeds $1 billion, making Eridu one of the most highly valued stealth-mode AI hardware startups. The investment coincides with a broader trend: NScale also raised $2 billion for AI data center expansion, and overall AI infrastructure investment in Q1 2026 has already surpassed all of 2024.

From a competitive landscape perspective, Broadcom's Tomahawk and Jericho series have dominated AI data center switching for years, holding over 80% market share. Eridu represents the most serious challenge to this dominance, alongside other emerging players in custom AI networking silicon. The key question is whether hyperscale cloud providers will adopt Eridu's technology — if even one major cloud player commits, it could reshape the entire AI networking stack.

Industry analysts note that networking has become the primary bottleneck for AI scaling, surpassing even GPU compute. As models grow larger and distributed training spans more nodes, the network fabric between GPUs matters as much as the GPUs themselves. Eridu's timing capitalizes on this shift, offering a purpose-built solution as the market transitions from general-purpose networking to AI-optimized infrastructure.