Eliyan, a company developing interconnects for AI semiconductors, has completed a $145 million Series C round. The valuation reached $1 billion, making the company a unicorn. The funds will be used to expand from improving within-package wiring that connects chiplets to each other, toward rack-to-rack optical connectivity. Eliyan is attempting to cover everything from AI chips to optical modules to cross-rack connections with a single product family. In AI infrastructure where increasing the number of compute units drains more power to data movement, the key to understanding this funding round lies in how the company plans to unify PHYs that have traditionally been separate for each connection distance.
Expanding Connectivity Range with $145 Million
The Series C round, completed on July 29, was led by Seligman Ventures, with Cisco Investments and optical component maker Lumentum joining as new strategic investors. Eliyan will direct the funds toward developing next-generation interconnects and expanding partnerships, while also scaling up manufacturing. Umesh Padval, Managing Partner at Seligman Ventures, will also join Eliyan's board of directors.
This round's investor composition reflects the direction Eliyan is heading. Cisco focuses on networking equipment and Lumentum on optical components—both companies outside the semiconductor packaging space where Eliyan has traditionally excelled. However, an investment does not imply product adoption or a purchase agreement. What was announced is capital participation; no joint products or supply agreements with either company have been disclosed.
Eliyan had also announced in January 2026 that it received $50 million in strategic investment from AMD, Arm, Coherent, Meta, and others. While the July materials describe the Series C as "completed at a total of $145 million," they do not address whether the January amount is included within that figure. Therefore, one cannot simply add the two figures together to arrive at a total funding amount for 2026.
Funding rounds and product progress have long gone hand in hand at Eliyan. The 2022 $40 million Series A coincided with the tape-out announcement of the 5nm version of NuLink, while the 2024 $60 million Series B set a goal of boosting the 3nm version to up to 64Gbps per link. This round coincides with both a shift in valuation and a pivot in the company's center of gravity—from purely electrical, within-package connectivity toward system-level integration in coordination with optical components.
Not the Component That Emits Light, But the Electrical Segment Leading to the Optical Engine
Five days before the funding announcement, Eliyan unveiled "NuLink-XD." This is a 224G PAM4 SerDes built on TSMC's 3nm process, serving as an electrical interface connecting AI chips to each other or chips to optical modules. It is designed for configurations such as Near-Package Optics (NPO), OSFP, and XPO, handling the segment immediately before signals are handed off to optical communication.
It would be a mistake to interpret NuLink-XD as an optical transceiver—doing so misreads the product's role. The SerDes converts data inside the chip into high-speed serial electrical signals and carries them to the optical engine. Converting the electrical signal into light beyond that point is handled by a separate component. Eliyan's term "electro-optical interconnect" refers to a design that bridges the electrical-optical boundary at low power; it is not an announcement that the company has completed its own products encompassing lasers or optical modulators.
NuLink-XD adopts a configuration that adjusts power consumption according to reach distance. Compared to general-purpose SerDes designs that carry extra margin to drive long board traces, this design narrows its focus to the short electrical segment from the chip to a nearby optical module, cutting unnecessary power in the process. Eliyan claims up to 40% lower power consumption compared to alternative approaches. However, the comparison target and wiring distance are not specified, and the announcement does not clarify measurement conditions or whether actual measured silicon exists. This figure should currently be understood as the company's design target or internal evaluation, not an independent performance comparison.
Moreover, the up-to-40% figure pertains specifically to the SerDes. It does not follow that overall link power—including lasers, optical modulators, and optical fiber—drops by the same percentage. Still, in configurations that route numerous links to hundreds of accelerators, the effect of reducing per-chip I/O power accumulates. Eliyan's aim is either to redirect the freed-up power budget toward computation, or to fit more high-speed links within the same power envelope.
A Product Roadmap Extending from D2D to Rack-to-Rack
Eliyan's starting point was NuLink D2D, which connects dies mounted on the same package substrate. The company has pursued a vision of providing a PHY usable even with standard organic substrates, reducing dependence on expensive silicon interposers. NuLink-XS extends connectivity to chips in separate packages, offering a reach of up to 100mm on the same board. NuLink-XD, meanwhile, creates a pathway reaching multiple racks by handing signals off to optical modules.
NuGear is a chiplet product family handling memory connectivity and I/O connectivity. As of January, Eliyan had set a target link bandwidth of 1.6Tbps to 12.8Tbps for scale-up connectivity combined with optical engines. Beyond providing PHY IP, the vision includes productizing chiplets that carry data outside the package. However, shipping specifications for each product and how they integrate with D2D and C2C have not yet been disclosed.
For rack-to-rack connectivity, power becomes an issue in addition to distance. In AI clusters, adding accelerators also increases the number of connections, causing power consumption by SerDes and optical modules to accumulate. What NuLink-XD aims for is designing the electrical segment before optical conversion to be as short as the use case allows, fitting more links within the same power budget. Even as compute capacity increases, accelerators are left waiting if data cannot reach them in time. Eliyan treats this I/O constraint as a continuous design problem spanning from the package to the rack.
According to the company's explanation, both scale-up—densely connecting numerous accelerators as a single computing system—and scale-out—horizontally expanding units such as servers or clusters—are targets. The former involves low-latency connections that bundle compute units together; the latter involves connections that expand the computing system itself. Eliyan has set a policy of extending the reach of its electrical PHY and chiplet offerings across both types of networks. However, neither is completed solely with Eliyan's own products—both presuppose integration with partners who supply optical engines and network switches.
Between Evaluation/Deployment and Mass-Production Adoption
Eliyan states that hyperscalers, AI accelerator companies, memory companies, and infrastructure companies are "evaluating or deploying" its technology. However, it does not disclose customer names or target products. Unit volumes and regions also remain unclear. Since this explanation applies to the company's entire technology portfolio, it cannot be read as confirmation that NuLink-XD—announced just five days prior—has already been deployed.
What can be officially confirmed about NuLink-XD is limited to the stage of having announced the architecture and being in the process of integrating it into next-generation systems with key customers and partners. There is no schedule for tape-out, sample shipment, or mass production start. While the $145 million will also be used to scale up manufacturing, the size of the funding and the maturity of the product are separate metrics.
Whether Eliyan can transition from a chiplet-connectivity IP company into a supplier unifying rack-scale electrical and optical connectivity will be determined by NuLink-XD's measured power performance and its first customer silicon. Under what wiring conditions will the up-to-40% claim be reproduced, and on which products will NuGear's 1.6Tbps-to-12.8Tbps target actually ship? The results of the manufacturing expansion can only be truly gauged once the language of "evaluation in progress" transforms into named, mass-production adoption.