Eliyan Raises $145M Series C and Hits Unicorn Status Solving AI’s Biggest Hidden Problem

There is a problem hiding inside every AI data center that almost nobody is talking about. And it is costing the industry billions. The world’s most advanced AI chips are sitting idle. Not because of software bugs or power shortages. They are waiting for data. Ramin Farjadrad, CEO and Co-Founder of Eliyan, has been saying this for years. Now, with a $145 million Series C led by Seligman Ventures and backed by Cisco Systems, Lumentum, and semiconductor veteran Umesh Padval, the market is finally listening. The Eliyan $145M Series C closed on 29 July 2026, valuing the Santa Clara startup at $1 billion and making it one of the most significant AI infrastructure unicorn stories of 2026.

What Is Eliyan?

Eliyan is an AI infrastructure company founded in Santa Clara, California, by Ramin Farjadrad, a semiconductor engineer with deep expertise in high-speed interconnect systems. The company focuses on a specific and increasingly critical slice of the AI infrastructure stack: the physical connections between AI processors. While the industry’s attention has been largely on chips like NVIDIA GPUs and Google TPUs, Eliyan’s thesis is that the connections between those chips are becoming the primary constraint on system performance.

Detail	Information
Headquarters	Santa Clara, California
CEO and Co-Founder	Ramin Farjadrad
Chief Strategy and Business Officer	Patrick Soheili
Series C Amount	$145 million
Post-Money Valuation	$1 billion (unicorn)
Lead Investor	Seligman Ventures (Umesh Padval)
Strategic Investors	Cisco Systems, Lumentum
First Chiplet Shipments	2026
Revenue Target	Hundreds of millions by end of 2027

The Problem: Why GPU Utilization Is Stuck at 30 to 40 Percent

This is the statistic that should stop every AI executive in their tracks. “Today we are at the point that maybe only 30% to 40% of the GPUs are being used, mainly because of how fast they can receive the data to process,” Farjadrad told Reuters. “We’re solving that problem.” The world’s most expensive AI hardware, GPUs costing $30,000 to $40,000 per unit, is sitting idle for 60% to 70% of its operating time. Not waiting for a better algorithm. Waiting for data that cannot arrive fast enough across the chip interconnects designed to carry it.

The math is straightforward. When an AI training run requires thousands of GPUs working in coordination, the performance of the entire system is bounded by the slowest communication link. If GPUs can process data 10x faster than they can send and receive it across chip interconnects, the entire cluster runs at 10x below its theoretical maximum. More compute does not help. The bottleneck is the connection.

Problem	Impact
GPU utilization in large AI clusters	30 to 40% of theoretical maximum
Root cause	Data movement bandwidth cannot match compute speed
Effect on AI training	Wasted compute budget, longer training times
Effect on AI inference	Higher cost per query, reduced throughput
Effect on data center economics	Massive underutilization of $30,000+ GPU hardware

The Solution: High-Speed Interconnect and Optical Chiplets

Eliyan develops two primary technology categories. Its high-speed electrical interconnect technology delivers higher bandwidth between AI processors at lower power consumption and reduced packaging complexity, without requiring proprietary networking ecosystems that lock customers into specific vendor stacks. This is particularly valuable for hyperscalers and semiconductor companies building custom AI accelerators who need a high-performance connectivity option not dependent on any single chip vendor’s proprietary network interface.

Eliyan’s roadmap also extends into electro-optical connectivity, an area attracting significant investment across the AI infrastructure sector. Optical interconnects use light rather than electrical signals to move data, enabling dramatically higher bandwidth and longer effective reach within data centers at lower energy per bit. As AI data centers scale to support models with trillions of parameters requiring thousands of accelerators in coordination, optical connectivity becomes increasingly necessary. Instead of building complete AI processor systems, Eliyan licenses its interconnect technology and provides chiplets that semiconductor companies integrate directly into their own AI chip designs, multiplying its reach without requiring the capital of a full chip manufacturer.

The Eliyan $145M Series C: Who Invested and Why It Matters

The investor syndicate behind Eliyan’s Series C is as telling as the dollar amount. The round was led by Seligman Ventures Managing Partner Umesh Padval, an early investor in Mellanox, whose $6.9 billion acquisition by NVIDIA in 2020 made NVIDIA’s AI networking dominance possible. Padval served on Mellanox’s board before the acquisition. The fact that the person who backed Mellanox early is now backing Eliyan is not coincidental. He understands better than almost anyone how critical interconnect technology is to AI infrastructure economics, and is placing a similar bet on Eliyan’s moment.

Cisco Systems’ participation as a strategic investor signals that one of the world’s largest networking companies sees value in Eliyan’s technology. Strategic investors at this stage bring distribution potential, integration opportunities, and market validation that pure financial investors cannot provide. Lumentum, a leading optical and photonic components manufacturer, joining as an investor in a company building optical chiplets is a strong signal that Eliyan’s electro-optical roadmap is technically credible from the perspective of a company that actually manufactures optical components. As part of the financing, Umesh Padval will join Eliyan’s board of directors.

Revenue Trajectory and Commercialization Plans

Eliyan’s commercial plans are ambitious and specific. According to Chief Strategy and Business Officer Patrick Soheili, the company expects revenue to grow from low millions of dollars in 2025 to hundreds of millions by end of 2027. Initial chiplet shipments are expected to begin in 2026, moving Eliyan from development-stage to commercial deployment within the current fiscal year. The company’s revenue model, based on licensing and chiplet product sales to hyperscalers and semiconductor companies, creates long design cycles with sticky, multi-year revenue streams once design wins are secured.

Period	Revenue
2025	Low millions (development stage)
2026	Transitional (first shipments begin)
End of 2027 (projected)	Hundreds of millions

How Eliyan Compares to the Competition

Eliyan is not the only company targeting the AI connectivity bottleneck. NVIDIA’s acquisition of Mellanox gave it InfiniBand networking technology, the dominant high-performance interconnect in large-scale AI training clusters. NVLink, NVIDIA’s own chip-to-chip interconnect, is built into its latest GPU architectures. Broadcom and Marvell Technology supply networking silicon to cloud providers including Google and Amazon Web Services for their custom AI chips. Ayar Labs is a direct competitor in the optical chiplet space, having raised over $200 million backed by Intel Capital. Eliyan’s key differentiator is vendor-agnostic independence: it works across different chip architectures without proprietary ecosystem lock-in, which is a strategic asset in a market where hyperscalers actively seek to reduce vendor dependency.

Company	Approach	Dependency	Best For
Eliyan	Chiplets, independent	Vendor-agnostic	Custom AI chip designers
NVIDIA/Mellanox	InfiniBand, NVLink	NVIDIA ecosystem	NVIDIA GPU clusters
Broadcom/Marvell	Ethernet silicon	Customer-specific	Large cloud providers
Ayar Labs	Optical chiplets	Intel partnership	Optical I/O integration

Why the Mellanox Parallel Matters

The parallels between Eliyan’s opportunity and the Mellanox story are worth examining carefully. When Mellanox was founded in 1999, it was solving the same structural challenge Eliyan is tackling today: compute performance was outpacing network bandwidth. Mellanox built InfiniBand interconnect technology that became the standard for high-performance computing clusters. When AI training clusters replaced HPC clusters as the defining large-scale computing workload, Mellanox’s technology was already embedded. NVIDIA paid $6.9 billion to own that technology in 2020. Umesh Padval saw that story from the inside. His bet on Eliyan suggests he believes the same structural opportunity is presenting itself again, but at a much larger scale, as AI cluster sizes that once numbered hundreds of GPUs now number tens of thousands.

Industry Impact and Expert Analysis

Eliyan’s $1 billion unicorn valuation reflects a broader market recognition that AI infrastructure investment is moving beyond compute hardware into the connectivity and networking layers that determine how efficiently that compute hardware performs. For U.S. AI startup founders and investors, the Eliyan story highlights an important pattern: some of the most defensible AI infrastructure opportunities are not in building AI models or applications, but in solving the physics-level problems that constrain the performance of AI hardware already deployed.

A company that can materially improve GPU utilization from 35% to 60% or higher is not selling a nice-to-have product. It is selling a solution that directly reduces the operational cost of running frontier AI at scale. The Cisco investment is particularly telling. Cisco does not make many AI infrastructure startup investments. When the world’s largest networking company backs a chiplet startup, it is signaling that the optical and high-speed interconnect market is large enough and strategic enough to warrant attention at the board level.

Future Outlook

Three developments will define Eliyan’s trajectory over the next 24 months: the conversion of chiplet product shipments into confirmed design wins with hyperscaler or semiconductor company customers; progress on the electro-optical connectivity roadmap, where early design wins will position the company best for the larger long-term market; and the competitive response from NVIDIA, whose control of the InfiniBand ecosystem gives it strong incentive to expand its proprietary networking stack in ways that could reduce the available market for independent providers. How Eliyan navigates that dynamic will depend significantly on its ability to secure multi-vendor design wins before NVIDIA consolidates further control.

Conclusion

The Eliyan $145M Series C and $1 billion unicorn valuation represent more than one startup’s funding milestone. They represent the market’s recognition that the AI data center bottleneck is real, measurable, and solvable, and that the company best positioned to solve it at the physical interconnect layer has now attracted the capital and strategic partners needed to execute. When only 30% to 40% of AI GPU capacity is being utilized due to data movement constraints, solving that problem is not incremental. It is transformational for the economics of AI infrastructure at scale.

Follow BestStartup.us for weekly coverage of U.S. AI infrastructure startups, funding rounds, and semiconductor innovation. Subscribe to our newsletter for analysis sent every Thursday.