Lumilens raises more than $700 million to connect AI data centers with light

Why it matters

AI spending is moving from GPU procurement into the networks around those chips. Lumilens's round shows investors expect optical interconnects to produce platform-scale winners.

Lumilens founder and CEO Ankur Singla has raised more than $700 million, at a reported $5.5 billion valuation, for a two-year-old hardware startup built around a costly premise: the next constraint on AI infrastructure is moving data between chips.

The San Jose company emerged from stealth on August 6, bringing its total capital raised above $900 million, according to Lumilens. The Wall Street Journal reported that Lumilens has started shipping optical transceivers into the production data centers of a large cloud provider. Lumilens has not named the customer and says the agreement is worth billions of dollars over several years.

Singla is taking on the technical and manufacturing risk with a record that helps explain why investors wrote such a large check before Lumilens disclosed revenue or shipment volumes. He previously founded software-defined networking company Contrail Systems, which Juniper Networks acquired, and Volterra, a distributed cloud company that F5 agreed to acquire for approximately $500 million in 2021. He earned electrical engineering degrees from the University of Southern California and Stanford University, according to a Contrail Systems biography.

Singla is also listed as CEO of cybersecurity startup Exaforce. At Lumilens, he is joined by founders Ritesh Kapahi, Samuel Liu, Dave Friedman and Ted Schmidt. The Lumilens leadership page describes a team drawn from Cisco, Juniper, Meta, Lumentum and Coherent.

Investors are backing a manufacturing ramp

The financing was co-led by Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures and Spark Capital.

Other participants included Addition, Aiconic, Alkeon, EDBI, HarbourVest, J.P. Morgan Private Capital, Lingotto, Mayfield, MVP Ventures, Peak XV, Qualcomm Ventures, Redpoint Ventures, Seifdune and Thomvest Ventures.

Lumilens said the new money will expand its silicon, systems, software, process engineering and high-volume manufacturing operations. That spending plan gets to the central challenge facing photonics startups. A design that moves data efficiently in a lab still has to survive customer qualification, component supply constraints, packaging complexity and mass production at an acceptable cost.

Lumilens is trying to control more of that process. Its LumiCore platform combines silicon photonics, mixed-signal integrated circuits, electrical-optical interposers and optical systems. Lumilens also says it has developed proprietary assembly processes, robotics and automated testing, while using manufacturing partners and its own facilities.

Those capabilities are Lumilens claims, and the financing gives it the capital to prove them at production scale. The disclosed customer agreement offers an early indication that a hyperscaler has qualified at least one product. Its economic value remains difficult to measure from the information released: Lumilens identifies neither shipment volumes nor recognized revenue, and describes the agreement's value across several years.

The network is becoming part of the AI computer

Lumilens is targeting two related connection problems inside AI data centers.

Scale-out networks move data between racks and clusters. Lumilens is developing 800-gigabit and 1.6-terabit pluggable optical transceivers for those links. Scale-up networks connect processors more tightly within a computing system. Lumilens's roadmap includes near-package optics and co-packaged optics that bring optical connections closer to GPUs and other processors.

Copper remains cheaper and well understood, but electrical signals lose effectiveness as bandwidth and distance increase. Singla told the Journal that optical equipment can cost several times as much as copper. Lumilens must therefore deliver enough bandwidth, reach and power savings to outweigh a substantial equipment premium.

That tradeoff is becoming more important as AI systems spread beyond single racks. The useful output of a large GPU cluster depends partly on how quickly processors can exchange model parameters and other data. A slow or power-hungry interconnect leaves expensive chips waiting, cutting the return on the data center's largest capital expense.

Singla's thesis is that AI infrastructure spending has moved past a singular focus on acquiring GPUs. In Lumilens's announcement, he said, "The constraint on AI has shifted from how many GPUs you can buy to how many you can connect."

Photonics has become a capital race

Lumilens is entering a field where startups and established semiconductor suppliers are already spending heavily.

Lightmatter, founded in 2017, had raised $850 million and was valued at $4.4 billion in 2024, according to the Journal. In March, Lightmatter introduced its Passage L20, a 6.4-terabit-per-second optical engine designed for near-package and on-board optics.

Ayar Labs is pursuing optical connections closer to the processor package with its TeraPHY optical I/O chiplet. The company raised $500 million in March 2026, according to the Journal and other reporting cited in the research.

The category has also produced a large strategic exit. Marvell completed its acquisition of Celestial AI on February 2, 2026. The transaction was announced with approximately $3.25 billion in upfront consideration, plus a potential earnout tied to revenue milestones.

That competition helps explain the scale of Lumilens's round. Optical interconnects require years of semiconductor development, packaging work, customer qualification and manufacturing investment. Once a component is designed into a hyperscaler's architecture, however, the supplier can become tied to a multi-year hardware roadmap.

Singla has spent his career building around changes in network architecture. Contrail treated networking as programmable software. Volterra addressed applications distributed across clouds and edge locations. Lumilens takes the same infrastructure instinct into physical hardware, where execution depends on factories, yields and optical packaging alongside software and systems design.

The $5.51 billion valuation prices in a rapid transition from a qualified product at one unnamed hyperscaler to repeatable high-volume production. Lumilens has the capital and founder history to pursue that transition. Its next test is proving that light can beat copper on economics, as well as on physics.