Eliyan raises $145 million at $1 billion valuation for AI interconnects

Why it matters

The round prices Eliyan's chiplet links as critical AI infrastructure, backing Farjadrad's bet that data movement and memory access will determine system performance.

Ramin Farjadrad, the serial semiconductor founder behind Eliyan, has raised $145 million at a $1 billion valuation for the Santa Clara company, Reuters reported on July 29th. The financing puts a unicorn price on technology designed to move data between the processors, memory and smaller dies inside increasingly large AI systems.

Farjadrad has spent his career working on that problem at different scales. He previously co-founded Velio Communications and Aquantia, which went public in 2017 before Marvell acquired it in 2019. His work has included high-speed serial links and Ethernet technologies that became industry standards. He started Eliyan in 2021 with Patrick Soheili and Syrus Ziai, a semiconductor engineering executive whose prior employers included Qualcomm, PsiQuantum and chip designer Nuvia.

The new capital is substantial even by Eliyan's prior funding record. It is nearly as large as Eliyan's three disclosed rounds of $40 million, $60 million and $50 million combined, excluding an additional strategic investment announced in 2024.

A bet on moving data

AI hardware is increasingly limited by the rate at which processors can exchange data with memory and other accelerators. Dividing a large processor into chiplets can improve manufacturing economics and let designers combine components made with different production processes. Those smaller dies still need high-bandwidth connections that do not consume excessive power or require costly packaging.

Eliyan sells the physical-layer technology for those links. Its NuLink and NuGear products are intended to connect dies, memory and systems inside AI infrastructure. NuLink covers die-to-die and related connectivity, while NuGear packages the technology into chiplets for scale-up networks and memory expansion.

The architecture grew from Farjadrad's work on Bunch of Wires, an interconnect method adopted by the Open Compute Project. Eliyan says NuLink can work with the UCIe chiplet standard and can deliver high bandwidth using standard organic substrates, giving chip designers an alternative to relying exclusively on silicon interposers and other advanced packaging methods.

Those performance figures remain Eliyan's measurements. In its 2024 Series B announcement, Eliyan said its technology could provide as much as four times the performance of competing approaches while consuming half the power. In January, Eliyan said its NuLink-X family offered about twice the energy efficiency of alternatives and that NuGear configurations targeted link bandwidths from 1.6 terabits to 12.8 terabits per second.

Farjadrad has framed the technical shift as a move from systems built on one large chip toward systems assembled inside a package. In a 2024 interview with investor Celesta Capital, he argued that chiplet connectivity would determine whether processors could continue gaining memory bandwidth and overall performance as conventional transistor scaling became harder.

Semiconductor companies have financed the roadmap

Eliyan's investor list shows how broadly the connectivity problem cuts across the chip industry. A $40 million Series A in 2022 was led by Tracker Capital Management, with Celesta Capital, Intel Capital and Micron participating. Eliyan used that announcement to disclose a tapeout on a 5-nanometer manufacturing process, an early step toward validating the design in silicon.

A $60 million Series B in March 2024 was co-led by Samsung Catalyst Fund and Tiger Global Management. Intel Capital, SK hynix, Cleveland Avenue and Mesh Ventures also participated. Samsung and SK hynix are major memory suppliers, giving both a direct interest in technologies that can connect accelerators to larger pools of high-bandwidth memory.

VentureTech Alliance later made an additional investment that pushed Eliyan's announced funding beyond $100 million. In January 2026, Eliyan secured another $50 million from AMD, Arm, Coherent and Meta, alongside returning investors Samsung Catalyst Fund and Intel Capital. Eliyan said that money would support manufacturing, qualification and deployment of its interconnect intellectual property and chiplet products.

The $145 million round gives Farjadrad far more capital for the same commercialization push. Semiconductor startups must fund design work, manufacturing runs, packaging, testing and lengthy customer qualification cycles before reaching volume production. Eliyan's valuation assumes its standards work and technical demonstrations can become components deployed across AI accelerators, memory systems and data-center networks.

That opportunity is growing as AI system design shifts beyond the performance of an individual processor. Nvidia, AMD, hyperscalers and a widening group of chip startups are assembling systems from multiple compute, networking and memory components. Each additional component increases the need for fast, power-efficient links. Eliyan is betting that the connection between those chips will command as much attention as the chips themselves.