There is a growing need to scale up data center infrastructure to meet the demands of artificial intelligence (AI). Technology companies across the world are spending record amounts of money on this issue—global investment bank Citigroup, USA, projects that investment in AI infrastructure will surpass US$2.8 trillion by 2029. Into this dynamic environment steps Lumilens, USA, a company developing optical interconnects for AI infrastructure.

On 6 August, Lumilens announced that it is emerging from stealth mode and will begin shipping its first products to AI data centers under a multi-billion US-dollar deal. The firm also revealed that it had received more than US$700 million in new funding, putting the company’s value at US$5.51 billion.

Addressing the AI bottleneck

Lumilens was founded in early 2024 “to solve networking bottlenecks limiting AI scale,” according to the company. The group says that the current issue with scaling AI infrastructure is not the need for more GPUs but rather the need to connect the GPUs. “A single 400,000 GPU data center requires more than 2.4 million transceivers and more than five million fiber strands from a market already unable to keep pace,” Lumilens said. “McKinsey projects that 800G optical transceiver production could fall 40-60% short of industry demand through 2027, with 1.6 terabit transceiver shortfalls of 30-40% persisting through 2029.”

There is a limit to scaling GPUs with classic copper connections—electrical signals can only travel about 1.5 meters over such a setup, which only supports hundreds of GPUs. Hyperscalers now need to support thousands of GPUs, so copper connections are no longer adequate .

 “The constraint on AI has shifted from how many GPUs you can buy to how many you can connect,” said Ankur Singla, founder and CEO of Lumilens. “Whether they are training frontier models or serving them to hundreds of millions of users, hyperscalers told us they need the same two things: far more optical capacity for the networks they run today and a path to directly connecting thousands of GPUs together into a single cluster. We built Lumilens to deliver both—and to manufacture at the volume, quality and pace this buildout demands.”

New funding for scaling up

Lumilens says its products cover all of the scale-up needs of an AI data center: the network that connects GPUs to one another and the fabric that stitches the racks and clusters together. The company’s near-packaged optics (NPO) and co-packaged optics (CPO) enable thousands, potentially tens of thousands, of GPUs to operate as a single, coupled domain. For the racks and clusters, the group produces pluggable transceivers to replace copper links that Lumilens claims are capable of supporting more data (800G per second, 1.6 terabits per second or more) with less power for the largest data center fabrics. The NPOs, CPOs and transceivers are all developed in-house and built on the company’s trademarked unified technology platform, LumiCore.

The latest round of financing, totaling more than US$700 million, was co-led by Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures and Spark Capital, the company said. That brings its total fundraising to over US$900 million since its founding. Lumilens has recently begun shipping its scale-out product into the data centers of “top hyperscaler” (which the company did not name) under a multi-billion-dollar agreement.

“The critical component in scaling AI is shifting from compute to connectivity,” said Gavin Baker, Managing Partner at Atreides Management, USA. “Focusing on that core AI challenge, Lumilens has moved from its founding toward a qualified product deployed inside a hyperscaler’s production data centers within two years, and is building the manufacturing engine to scale with demand while bringing much-needed diversification to AI supply chains.”