Lumilens Raises $700 Million as AI Data Centres Hit Network Limits
Frontier Enterprise coverage shows Lumilens raising more than $700 million at a $5.51 billion valuation after its optical interconnect product moved into a hyperscaler’s production AI data centres.
Lumilens has secured more than $700 million in fresh capital for AI data centre connectivity, lifting its valuation to $5.51 billion as large compute buyers confront a networking problem that more GPUs alone cannot solve, Frontier Enterprise detailed.
Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures and Spark Capital led the Series C round.
The wider syndicate included Peak XV, Qualcomm Ventures, J.P. Morgan Private Capital, Harbourvest, Alkeon, Addition, Mayfield, MVP Ventures, Redpoint Ventures, Seifdune and other backers.
The round pushes Lumilens’ total fundraising beyond $900 million.
The transaction gives a two-year-old company a much larger manufacturing mandate.
Lumilens was formed in early 2024 after its founders identified the connection limits beginning to appear in the biggest AI clusters.
One unnamed hyperscaler was already planning around those limits as it sized the infrastructure required for its AI roadmap.
That customer relationship has moved beyond evaluation.
Lumilens’ first scale-out product has completed qualification and is now being used in the hyperscaler’s production AI data centres through an agreement described as worth multiple billions of dollars.
For the company, the deployment turns the funding story from a pure capacity bet into a customer-backed expansion plan.
The technical target is the web of links inside an AI data centre.
Lumilens builds optical interconnects for scale-up networks, which connect GPUs within a compute system, and for scale-out fabrics, which tie racks and clusters together.
Its portfolio includes near-package optics and co-packaged optics that move optical input and output closer to the GPU, plus pluggable modules at 800G, 1.6T and higher rates for replacing copper connections across rows of equipment.
Ankur Singla, Lumilens’ founder and chief executive, described the bottleneck as a shift from GPU purchasing power to GPU connection capacity.
He also tied hyperscaler demand to two requirements: more optical bandwidth for networks already in operation and a route to joining thousands of GPUs into one cluster for frontier-model training and large-scale inference.
Investors framed the same constraint from the supply-chain side.
Seligman Ventures managing partner Umesh Padval linked AI infrastructure performance to how efficiently thousands of GPUs can behave as a single system.
Atreides managing partner Gavin Baker highlighted the speed of the move from founding to qualified production deployment, and Mayfield managing partner Navin Chaddha argued that copper can no longer meet the industry’s scaling needs.
Lumilens is pitching the round against a market opportunity for photonic interconnects estimated at more than $100 billion.
The company has recruited leaders and engineers with backgrounds at Meta, Cisco, Juniper Networks, Marvell, Coherent and Lumentum.
The new money is designated for expanding manufacturing, process engineering, software, systems and silicon operations as demand for AI cluster networking grows.
