Optical Networking Startup Lumilens Valued at $5.5 Billion in $700 Million Series C

The technology industry has spent the last three years obsessing over the supply of artificial intelligence chips, but the actual bottleneck constraining next-generation AI has quietly shifted to the cables connecting them. Traditional copper wiring is hitting physical and thermal limits inside massive GPU clusters, throttling the speed at which data can move. To resolve this tension, silicon photonics startup Lumilens has raised $700 million in a Series C round to replace those copper links with light. The massive capital injection underscores how the AI infrastructure race is expanding beyond processors to the physical networks that tie them together.[1][4]

The financing values the San Jose-based company at $5.51 billion just two years after its founding. Co-led by Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures, and Spark Capital, the mega-round brings Lumilens's total capital raised to over $900 million. The rapid accumulation of capital reflects the urgent demand from the world's largest technology companies, which are spending historic sums to build data centers capable of training increasingly complex artificial intelligence models without being constrained by legacy hardware.[2][6]

Modern artificial intelligence models are trained across hundreds of thousands of processors that must function seamlessly as a single, unified computer. The overall performance of that massive supercomputer is now determined just as much by the network linking the chips as by the compute capabilities of the chips themselves. Lumilens designs advanced optical interconnects—using silicon photonics to transmit data via pulses of light rather than traditional electrical signals—that allow these massive clusters to communicate exponentially faster. By moving data optically, the technology enables data centers to transmit information over longer distances and with significantly less power consumption than thick, heat-generating copper cables.[1][3]

The company's portfolio targets two distinct networking domains inside the modern AI data center. For "scale-out" architecture, which connects different racks and rows of servers together across a massive facility, Lumilens produces pluggable optical transceivers capable of 800G and 1.6T speeds. These high-bandwidth modules replace copper links across the broader data center fabric, carrying significantly more data at lower power levels to support the largest and most demanding artificial intelligence deployments currently being built by major tech firms.[3][6]

For "scale-up" architecture, which links graphics processing units directly to one another within a single compute system, the startup is developing near-package and co-packaged optics. This cutting-edge technology brings optical input and output directly to the GPU itself, bypassing the severe bandwidth limitations of electrical traces on a standard motherboard. Ultimately, Lumilens aims to use this co-packaged technology to enable tens of thousands of GPUs to operate as a single, tightly coupled domain without being bottlenecked by internal data transfer speeds.[3][6]

Unlike many deep-tech hardware startups that spend years in research and development before seeing commercial traction, Lumilens is already generating significant revenue. The company disclosed that its first scale-out products have completed rigorous qualification processes and are actively shipping into production AI data centers. These shipments are part of a multi-billion-dollar customer agreement with an unnamed hyperscale cloud provider, validating the immediate market need for the technology and proving that the startup can deliver at enterprise scale.[3][5]

"The constraint on AI has shifted from how many GPUs you can buy to how many you can connect," said Lumilens founder and chief executive officer Ankur Singla. By engineering its manufacturing processes alongside its product development, the company aims to scale its proprietary photonic assembly technologies to meet the surging demand. This integrated approach includes automated testing and dedicated robotic manufacturing processes designed specifically to accelerate high-volume production and maintain tight control over the supply chain.[3][6]

As artificial intelligence labs push toward frontier models requiring unprecedented amounts of compute, the industry's reliance on optical networking will only deepen. With financial analysts projecting that technology giants will spend trillions on AI infrastructure by the end of the decade, the physical limits of electricity and copper have become a hard ceiling. Lumilens's rapid ascent from stealth to a multi-billion-dollar valuation signals that the next massive battleground in AI will not just be about the processors themselves, but the light-speed highways that link them together.[1][4][5]