Lumilens Exits Stealth at $5.5B Valuation to Fix AI Data Centers' Connectivity Crisis

Optical networking firm Lumilens valued at $5.5 billion in latest funding round

Aug 6 (Reuters) - Optical networking startup Lumilens said on Thursday it was valued at $5.51 billion in its latest funding round, as surging investment in AI data centers fuels demand for high-speed connectivity technologies. The San Jose, California-based company designs and manufactures optical and networking products that help data centers move data faster while using less power. It raised more than $700 million in the Series C funding round co-led by Atreides Management, Bain Capital Ventures, Meritech, ⁠Seligman Ventures and Spark Capital, bringing its total funding to more than $900 million. Lumilens said it is already shipping products under a multi-billion-dollar agreement with a customer it declined to identify, but said it was one of the four hyperscalers. The term refers to tech giants Amazon (AMZN.O), opens new tab, Alphabet (GOOGL.O), opens new tab, Microsoft (MSFT.O), opens new tab and Meta (META.O), opens new tab. The funding comes as hyperscalers spend billions of dollars to expand data centers to meet the surging demand for artificial intelligence services. The data ⁠center boom is in turn driving demand for networking technologies that can efficiently connect thousands of GPUs. "The constraint on AI has shifted from how many GPUs you can buy to how many you can connect," Lumilens CEO Ankur Singla said. "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." The company said it will use the fresh capital to scale up its engineering and manufacturing operations. Lumilens also said it has hired ⁠executives and engineers from Cisco (CSCO.O), opens new tab, Juniper Networks, Meta, Marvell (MRVL.O), opens new tab, Lumentum (LITE.O), opens new tab and Coherent (COHR.N), opens new tab, bringing expertise in photonics and large-scale networking. Reporting by Prakhar Srivastava and Anhata Rooprai in Bengaluru; Editing by Vijay Kishore and Leroy Leo Our Standards: The Thomson Reuters Trust Principles., opens new tab

Lumilens leaves stealth at $5.5bn to fix AI's wiring problem

The hard part of building AI is no longer buying the chips. It is wiring them together, and a two-year-old startup called Lumilens has raised more than $700m betting the fix is light, not copper. Lumilens came out of stealth on Thursday with a big number and a bigger claim. It announced more than $700m in fresh funding, taking its total to over $900m and valuing it at $5.51bn. It also says it is already shipping its first product into a hyperscaler's live data centres, under a deal it says is worth billions. The pitch rests on a shift the whole industry is feeling. "The constraint on AI has shifted from how many GPUs you can buy to how many you can connect," said founder and chief executive Ankur Singla. Modern models run across hundreds of thousands of chips that have to behave as one computer. The wiring between them now matters as much as the chips. Copper hit a wall Two problems bite. Across racks, every GPU you add multiplies the optical transceivers needed. A single 400,000-GPU site needs more than 2.4 million of them, and the market cannot keep up. Inside the rack, GPUs are wired to each other over copper, and copper is out of road. At AI speeds an electrical signal survives about a metre and a half over copper. That caps a tightly linked cluster at a few hundred chips. To wire thousands together, you have to swap copper for light. That is Lumilens's business: optical interconnects for both jobs, the short hops inside a rack and the long ones between them. It is the same copper bottleneck that pushed Nvidia into photonics. A crowded, hyped race Investors are treating it as a landmark. The round was co-led by Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures and Spark Capital, with Qualcomm Ventures and JPMorgan among the rest. Mayfield's Navin Chaddha, who has backed Singla three times, said he had never seen growth like it in the firm's 56-year history. There is substance behind the froth. Singla has built and sold two infrastructure companies before, Contrail to Juniper and Volterra to F5. His team comes from Cisco, Meta and Marvell. The spending backdrop is real too. Citigroup expects AI infrastructure outlays to pass $2.8tn by 2029. The caveats are worth stating. The marquee customer is unnamed, and the multi-billion-dollar deal is Lumilens's own figure, not an audited one. It is also not alone. Co-packaged optics is one of the hottest corners of the chip world, contested by Nvidia, Broadcom and a row of startups. A fat private valuation is easier to win than a durable lead. Its Chinese rivals, from Eoptolink down, are chasing the same prize. For now, Lumilens has the thing the market values most: a product already inside a data centre. Moving data has become harder than crunching it.

Optical networking startup Lumilens launches with $900M in funding

Lumilens Inc., a provider of optical networking chips for artificial intelligence clusters, launched today with $900 million in funding. The company raised the bulk of the capital, $700 million, through a Series C round that valued it at $5.51 billion. Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures and Spark Capital jointly led the investment. They were joined by more than a dozen others including Qualcomm Inc. Lumilens is led by co-founder and Chief Executive Officer Ankur Singla. His previous two networking startups, Contrail Systems Inc. and Volterra Inc., were sold for a combined $676 million. Two-year-old Lumilens says that it has already received billions of dollars worth of chip orders. Data center operators historically used copper wires to move data between processors. Lumilens' chips make it possible to replace copper interconnects with faster fiber-optic cabling. The company makes optical chips for two use cases: scale-up and scale-out networking. The former term describes technology that links together chips in a single rack, while scale-out networks connect different racks. Lumilens is targeting the scale-up market with two product families. The first is a co-packaged optics, or CPO, chip series that can be integrated directly into graphics processing units. The company says that it's capable of linking together thousands of GPUs. Lumilens also makes near-packaged optics, or NPO, chips. Such chips are not integrated directly into GPUs but rather sit on the motherboard that hosts them. CPO chips are more efficient than NPO hardware because they're located closer to the associated graphics card. That reduces the distance data must travel. NPO technology, on the other hand, is more readily accessible. "We believe that we're still about two to three years away when it comes to large-scale volume deployment of co-packaged optics and the interim solution is near-packaged optics," Singla explained in an interview on SiliconANGLE Media's theCUBE. "We were kind of pioneers in that. Our first chip taped out in 2024 and we now have our second generation of chips taped out and getting ready for commercial deployments next year." Lumilens also makes chips for scale-out networks, which facilitate data sharing across GPU racks. It ships the silicon as part of devices called pluggable transceivers that can be attached to switches. Lumilens says that its technology supports connections with bandwidth of 800 gigabits per second, 1.6 terabits per second or more. The technology doesn't have the range limitations that make copper difficult to use in scale-out networks. "The distance limitation starts to come in the moment you get to 1.6 Tb/s, you're actually at one meter or so in distance with copper. That basically is within the rack," Singla said. "In order to get outside of the rack, you need to get to optics." Lumilens' chips are based on a common set of building blocks called the LumiCore. It includes photonic components, mixed signal integrated circuits that can manage those components and interposers. The latter modules function as the base layer on which a chip's optical and electronic parts sit. Lumilens says that LumiCore reduces the amount of time required to develop new chips from years to months. Additionally, the interposer technology included in the architecture eases the chip assembly process. The company has developed a custom manufacturing workflow complete with robots and AI software. Lumilens will use the capital to accelerate its chip development and manufacturing efforts.