gpu-networking
(4)
Lumilens Bet $50 Million That AI's Next Bottleneck Is the Wire
The two-year-old photonics company has a product inside a hyperscaler's live data centers and a plan to replace short copper links with light. Its harder bet is not merely designing faster optics - it is manufacturing them like semiconductors.
Rochan Sankar
Rochan Sankar is the founder and CEO of Enfabrica, a Silicon Valley startup building networking silicon to move data between the GPUs, CPUs and memory that power large-scale AI. A 26-year semiconductor veteran who ran Broadcom's data center Ethernet switching business before striking out on his own in 2019, he bet the company on a single idea: that AI's real bottleneck is not compute but the plumbing that feeds it. In September 2025, Nvidia paid more than $900 million to license Enfabrica's technology and bring Sankar and much of his team in-house.
Aviz Networks
Aviz Networks is a San Jose-based networking software company that helps enterprises run open, vendor-agnostic networks built on the open-source SONiC operating system and layered with AI. Its stack - ONES, Open Packet Broker, and the GenAI Network Copilot - lets organizations deploy, observe, and operate multi-vendor, multi-ASIC networks across the data center, edge, and AI factory without vendor lock-in, combining 24x7 SONiC support with orchestration, telemetry, and plain-language automation.
nEye.ai
nEye.ai (nEye Systems) is a Santa Clara semiconductor company building optical circuit switches (OCS) on a single chip. By fusing MEMS with silicon photonics and CMOS at wafer scale, its flagship SuperSwitch aims to move data between thousands of GPUs, CPUs and memory units with far less power and space than conventional electrical switching. Founded in 2020 by UC Berkeley photonics researchers and led by telecom-optics veteran Ashish Vengsarkar, the company raised an $80M Series C in April 2026, bringing total funding to about $152.5M as it scales foundry-based manufacturing for AI data centers.