Lumilens, a San Jose optical interconnect startup founded in 2024, raised $700M in Series C funding led by Atreides Management, Bain Capital, Meritech, Seligman Ventures, and Spark Capital, valuing the company at $5.51B and bringing total raised to $900M+. The company emerged from stealth after two years with product already shipping into an unnamed hyperscaler's production AI data centers under a multibillion-dollar customer agreement. Lumilens designs and manufactures near-package optics (NPO), co-packaged optics (CPO), and pluggable optical transceivers to address the GPU interconnect bottleneck in AI clusters.
The startup's core thesis: as AI clusters scale to hundreds of thousands of chips, the electrical signals over copper wiring degrade after roughly 1.5 meters at high data rates. Optical interconnects replace copper for both scale-up (within racks) and scale-out (between racks) networking. A single 400,000-GPU data center could need more than 2.4 million transceivers and five million fiber strands. Lumilens claims its silicon-photonics platform (LumiCore) enables faster product cycles from design to production than the legacy optics industry—months vs. years.
The constraint in AI infrastructure has shifted from compute capacity to connectivity. Citigroup projects AI infrastructure spending will exceed $2.8T by 2029, and an increasing share funds optics. This round reflects investor confidence in the company's repeat founder (Ankur Singla previously exited Contrail to Juniper and Volterra to F5) and early hyperscaler traction. Architects budgeting rack-scale deployments should track Lumilens as a key bottleneck solver—optical capacity is becoming as critical as GPU allocation in large training or inference clusters.