Eliyan raises $145M at $1B valuation to tackle AI chip data bottlenecks in data centers

Eliyan raises $145 million at $1 billion valuation to ease AI chip data bottlenecks

SAN FRANCISCO, July 29 (Reuters) - Eliyan, a startup that aims to ease the bottlenecks in transferring data between AI chips in data centers, said on Wednesday it had raised $145 million in venture funding at a $1 billion valuation. Cisco Systems (CSCO.O), opens new tab, Lumentum (LITE.O), opens new tab and an early investor in networking pioneer Mellanox before its sale to Nvidia backed the raising, according to Santa Clara, California-based Eliyan. The startup is developing technology aimed at the growing list of companies that are developing their own custom AI computing chips to compete with graphics processing units from Nvidia (NVDA.O), opens new tab and Advanced Micro Devices (AMD.O), opens new tab. Regardless of who makes them, most ⁠AI chips today face the same basic problem: They can process data much faster than they can send or receive it, which leaves pricey chips waiting and idle. "People have tried to build faster compute, faster processors and GPUs," Eliyan CEO and co-founder Ramin Farjadrad told Reuters in an interview. "Today we are at the point that we are maybe 30-40% of the GPUs is being used, and is being mainly limited by how fast they can receive the data to process. We're solving that problem." Nvidia overcame the connectivity issue in part by buying Mellanox for $6.8 billion ⁠in 2019. Companies like Alphabet's Google and Amazon.com's cloud unit solve the problem by working with firms such as Broadcom (AVGO.O), opens new tab and Marvell Technology (MRVL.O), opens new tab for access to those firms' networking technology. Eliyan aims to offer an independent option, both licensing its technology and providing pieces of a chip called "chiplets" that companies making custom processors can ⁠integrate into their own designs. Patrick Soheili, Eliyan's chief strategy and business officer, said the company expects to start initial shipments of chiplets this year and forecasts hundreds of millions of dollars in sales by ⁠the end of 2027, up from "low millions" of dollars in revenue in 2025. The Series C funding round was led by Seligman Ventures, with both existing and new investors taking part in ⁠addition to Cisco and Lumentum. As part of the funding, Seligman Managing Partner Umesh Padval, who was on Mellanox's board when it was sold to Nvidia, will take a seat on Eliyan's board. Reporting by Stephen Nellis in San Francisco; Editing by Jamie Freed Our Standards: The Thomson Reuters Trust Principles., opens new tab

Optical interconnect startup Eliyan raises $145M in funding to solve the AI chip cluster connectivity crunch

Silicon Valley-based chip interconnect startup Eliyan Corp.'s valuation has increased to $1 billion after it closed on a $145 million Series C funding round today. Seligman Ventures led the round, with Cisco Ventures and Lumentum also backing the newly minted unicorn. Eliyan says the round validates that its technology is poised to address a growing problem in the artificial intelligence industry. Namely, it's targeting the bottleneck that occurs in data centers as companies build ever-larger clusters of AI accelerators to power more sophisticated large language models and "agentic" applications. As data center operators scale up their chip clusters to try and support these increasingly powerful workloads, they're facing some tricky problems when it comes to managing their power demands and bandwidth requirements. Simply put, existing chip interconnects and power systems lack the bandwidth needed to feed these enormous clusters with enough data and energy to keep them running. As a result, their performance is often throttled as they sit idle, waiting for the data they need to process their calculations. AI training and inference workloads are continuously stalling, because the infrastructure can't keep up. The startup has developed a new breed of electro-optical interconnects that are designed to get around these challenges by facilitating rapid, high-speed communication across compute, memory and networking and eliminate these bandwidth constraints. They include its NuLink PHYs and NuGear chiplets, which provide die-to-die, chip-to-chip and rack-to-rack connectivity that can drastically improve both the bandwidth capacity and power efficiency of AI clusters. Because its chiplets support the full spectrum of AI infrastructure connectivity, it means data centers can support highly distributed, co-packaged computing fabrics that enable chips to run at their full compute capacity, without massively increasing power consumption. Co-founder and Chief Executive Ramin Farjadrad said AI is driving a significant architectural transition in the data center industry. "As compute, memory, packaging and networking become increasingly interconnected, traditional approaches to system connectivity are reaching their limits," he said. "Eliyan was founded to address these challenges at the architectural level. This financing positions us to accelerate product development, customer adoption, and ecosystem expansion as we help enable the next generation of AI infrastructure." Going forward, the startup will use the funds from today's round to finance its ongoing research and development program and develop even faster interconnect technologies. At the same time, it wants to scale its manufacturing capabilities and grow its partner ecosystem to expand the compatibility of its chiplets. Farjadrad says the company has already made good progress on that front, with its technology currently being evaluated and even deployed by some hyperscalers, AI accelerators and memory chip suppliers. He didn't name any of those partners, however. Seligman Ventures' managing partner Umesh Padval praised the startup's "technical depth" and its portfolio of more than 100 patents. He said optical interconnects might not be glamorous, but they are becoming a performance-defining technology in AI data centers. "Eliyan has developed a differentiated architecture to address this challenge, combining deep semiconductor expertise and a scalable approach to the evolving needs of next-generation AI systems," he added.