Eliyan Series C backs chiplet interconnects for next-generation AI infrastructure

Eliyan Corporation has raised $145 million in Series C funding at a $1 billion valuation, turning the Santa Clara chiplet interconnect company into the latest AI infrastructure unicorn. Seligman Ventures led the oversubscribed round, with participation from new strategic investors Cisco Investments and Lumentum, alongside other new and existing backers. The financing will support Eliyan’s next-generation electro-optical interconnect technologies, product development, ecosystem partnerships and manufacturing scale-up. The company is targeting one of the least glamorous but most important constraints in artificial intelligence infrastructure: the movement of data between processors, memory, chiplets, packages, racks and optical networks. The strategic question is whether Eliyan can become an independent interconnect platform for custom AI chips, or whether the market will remain dominated by closed ecosystems built by NVIDIA, Broadcom, Marvell and hyperscale cloud providers.

The funding round comes as AI infrastructure spending moves deeper into the physical limits of data centres. Expensive graphics processors and custom accelerators can process information extremely fast, but they lose economic value when data cannot reach them quickly enough. Reuters quoted Eliyan’s chief executive officer and co-founder Ramin Farjadrad as saying that GPU utilisation can be constrained sharply when processors wait for data rather than compute.

That makes Eliyan’s timing important. The first phase of the AI buildout rewarded companies that could supply compute. The next phase is increasingly rewarding companies that can move data more efficiently inside and across those compute systems.

Why does Eliyan’s $145 million Series C matter to the AI infrastructure market?

Eliyan’s funding round matters because artificial intelligence infrastructure is entering a connectivity-constrained phase. For much of the current AI cycle, investor attention focused on chips, especially graphics processors and high-bandwidth memory. That made sense because model training and inference require enormous compute density.

But large AI systems are not single chips. They are clusters of accelerators, memory devices, switches, optical links, chiplets, racks and servers that must exchange data continuously. If any part of that chain slows down, the whole system becomes less efficient.

This is where Eliyan’s story becomes commercially important. A data centre operator can spend billions of dollars on processors, but poor interconnect efficiency can leave part of that investment underutilised. The economic problem is not only performance. It is cost per token, power per workload, rack density, memory access and total cost of ownership.

Eliyan is trying to address that problem through chiplet interconnect technology that supports die-to-die, chip-to-chip, memory expansion and rack-to-rack connectivity. Its platform spans electrical and emerging electro-optical architectures, which positions the company at the boundary between semiconductor design and optical networking.

The $1 billion valuation suggests investors believe interconnect is becoming a value-creation layer rather than a supporting technical feature. If that thesis is correct, the companies that solve AI data movement could capture meaningful economics from the next generation of AI systems.

The risk is that interconnect is a technically demanding market where customers move carefully. Chip companies and hyperscalers do not redesign architectures casually. Eliyan must prove not only technical performance, but reliability, integration ease, manufacturing support and commercial relevance across different customer roadmaps.

How does Eliyan’s NuLink and NuGear technology target the AI data bottleneck?

Eliyan’s core technology is built around high-performance physical-layer interconnects and chiplet products. Its NuLink PHY family supports standards such as Universal Chiplet Interconnect Express and Bunch of Wires, while also enabling proprietary interconnect structures. Its NuGear chiplet products are designed to help customers break through memory and input-output constraints in AI and high-performance computing systems.

The purpose is to let data move faster and more efficiently between multiple dies, packages and system components. This matters because modern AI chips increasingly use multi-die architectures rather than relying only on one monolithic chip. Chiplets can improve yield, design flexibility and cost, but they also create a communication challenge. The pieces must talk to each other as if they were part of a coherent high-performance system.

Eliyan says its technologies support high bandwidth, low latency and power efficiency across both advanced packaging and standard organic substrates. The standard substrate point is commercially important because advanced packaging capacity is expensive and constrained. If customers can achieve strong performance without relying entirely on the most expensive packaging methods, the addressable market becomes wider.

The company is also moving into electro-optical interconnects. This reflects a larger industry transition: electrical links remain critical inside packages and boards, but optical connectivity becomes more attractive as distances, bandwidth needs and power constraints increase.

That transition is still developing. Optical interconnects can improve bandwidth and power efficiency, but they require strong integration between photonics, packaging, electrical interfaces and system design. Eliyan’s challenge is to prove it can bridge those domains without becoming trapped as a narrow component supplier.

Why are Cisco and Lumentum strategically important to Eliyan’s round?

Cisco Investments and Lumentum joining the round gives Eliyan stronger strategic credibility in networking and optics. That matters because the AI interconnect problem extends beyond chip packages into data-centre-scale networks.

Cisco is one of the most important networking companies in global enterprise and data-centre infrastructure. Its participation signals interest in how AI systems will evolve from traditional networking toward denser, lower-latency, higher-bandwidth architectures. Cisco shares traded at $112.48 in the latest available market data, with a market capitalisation of about $447.9 billion.

Lumentum is a major optical and photonics supplier, making its participation particularly relevant to Eliyan’s electro-optical direction. Lumentum shares traded at $602.35, down sharply during the session, with a market capitalisation of about $57.9 billion. The stock remains highly sensitive to AI optical-networking expectations because investors increasingly see optics as a core layer in AI data-centre expansion.

Strategic investors can help Eliyan in ways that ordinary venture capital cannot. Cisco can provide insight into data-centre networking requirements, customer architectures and scale-out challenges. Lumentum can support understanding of optical module economics, photonics integration and bandwidth-density requirements.

The question is whether these relationships become commercial pathways. Strategic investment alone does not guarantee customer adoption. Eliyan must convert investor interest into design wins, customer validation and revenue.

Still, the investor mix is meaningful. It suggests that Eliyan’s market is not limited to chip designers. It reaches networking, optics, hyperscale infrastructure and memory-system architecture.

Can Eliyan challenge NVIDIA’s Mellanox advantage without competing directly with NVIDIA?

NVIDIA’s acquisition of Mellanox in 2019 was one of the most important strategic moves in modern AI infrastructure. It gave NVIDIA a deeper position in networking and interconnects, strengthening its ability to deliver full systems rather than only accelerators.

That matters because AI customers increasingly buy platforms, not isolated components. NVIDIA’s advantage comes from chips, networking, software, systems, developer adoption and supply-chain control working together. Competing directly with that stack is extremely difficult.

Eliyan is not trying to become another NVIDIA. Its opportunity is different. It wants to provide an independent interconnect option for companies designing custom AI accelerators, memory systems and chiplet-based architectures.

That market is growing because hyperscalers, semiconductor companies and cloud providers want alternatives to full dependence on NVIDIA GPUs. Alphabet, Amazon Web Services, Meta, Microsoft and others are developing or using custom silicon strategies. These chips still need high-performance data movement.

Broadcom and Marvell already play important roles in custom silicon and networking for large cloud customers. Broadcom traded at $370.32, with a market capitalisation of about $172.4 billion, while Marvell traded at $163.40, with a market capitalisation of about $146.0 billion. Both stocks declined during the latest session, reflecting pressure across AI semiconductor and networking names.

Eliyan’s potential advantage is independence and focus. It can license technology and provide chiplets that customers integrate into their own architectures. That could appeal to companies that do not want to rely entirely on one large supplier’s proprietary ecosystem.

The risk is that the same customers may prefer working with established suppliers that have deeper manufacturing, support and integration experience. Eliyan’s technology must be not only superior, but trusted enough to enter high-value chip roadmaps.

Why are chiplets becoming central to the next phase of AI hardware design?

Chiplets are becoming important because monolithic chip scaling is increasingly difficult. As chips become larger and more complex, manufacturing yield can decline, cost can rise and design cycles can become more challenging.

A chiplet architecture allows designers to combine multiple smaller dies into one system. These dies may handle compute, memory, input-output, security, networking or specialised acceleration. This can improve flexibility and allow companies to mix technologies or process nodes more efficiently.

For AI systems, chiplets can help address several constraints at once. They can support more memory bandwidth, better yield economics, faster product iteration and custom architecture choices. But chiplets also increase the need for high-performance interconnects.

This is where Eliyan fits. A chiplet-based design is only as good as the communication between chiplets. If data movement is too slow, too power-hungry or too expensive, the architecture loses its advantage.

Universal Chiplet Interconnect Express has emerged as an important standard for die-to-die connectivity, but standards alone do not solve every performance problem. Customers still need physical-layer technology, integration know-how and products that can work across different packaging choices.

Eliyan’s argument is that it can provide that foundational connectivity. If the chiplet economy expands as expected, the company could benefit from a market where many chip designers need interconnect expertise but do not want to develop everything internally.

That is a strong thesis. It is also a demanding one because chiplet adoption requires ecosystem maturity across foundries, packaging houses, IP vendors, memory suppliers and system designers.

What does Eliyan’s revenue ambition reveal about the scale-up challenge?

Reuters reported that Eliyan expects to begin initial chiplet shipments this year and aims for hundreds of millions of dollars in sales by the end of 2027, compared with low millions of dollars of revenue in 2025. That projected ramp is aggressive.

The ambition makes sense if Eliyan’s products enter high-volume AI infrastructure programmes. A single successful design win inside a major customer roadmap can create large revenue because data-centre hardware is deployed at scale.

However, semiconductor revenue ramps can be uneven. Customers may evaluate technology for long periods before committing. Even after a design win, revenue depends on tape-out schedules, qualification, production timing, customer demand and supply-chain readiness.

This means Eliyan’s $1 billion valuation is partly based on forward conversion. Investors are paying for the possibility that early technical validation becomes commercial deployment across major AI infrastructure customers.

The company’s more than 100 patents, strategic investors and commercial momentum help support that case. But public-market investors would eventually want clearer disclosure: signed design wins, customer concentration, gross margin, shipment volumes, recurring licensing revenue and product revenue mix.

Eliyan’s business could include both licensing and chiplet sales. Licensing can provide attractive margins, while chiplet products can increase revenue scale but add manufacturing and working-capital requirements. The balance between the two will shape the company’s eventual valuation framework.

How does Eliyan fit into the broader AI interconnect and optical networking race?

Eliyan’s raise is part of a much larger shift in AI infrastructure. The market is moving from processor scarcity toward system-level optimisation. That means interconnects, optics, switches, memory, packaging and power efficiency are becoming more visible.

Companies such as Broadcom, Marvell, Cisco, Lumentum, Coherent, NVIDIA and several silicon photonics startups are all competing around different layers of the same problem. The goal is to move more data at lower power and lower latency across larger AI systems.

Optical networking has become especially important because AI clusters are scaling across racks and eventually across multiple data halls. Electrical connections remain essential at short distances, but optical links can become more attractive as bandwidth and reach requirements increase.

Eliyan’s electro-optical direction places it in a high-potential but competitive part of the stack. If it can connect chiplets, packages and optical modules efficiently, it may participate in the shift toward co-packaged optics and near-package optics.

The difficulty is that large customers often prefer complete, validated ecosystems. An interconnect startup must work with chip designers, foundries, packaging providers, module vendors and system builders at the same time. That coordination is not easy.

The opportunity is large because AI infrastructure is forcing customers to reconsider older architectures. When data-centre operators spend tens of billions of dollars on compute, even small efficiency gains in interconnect can translate into meaningful savings.

What risks could challenge Eliyan after becoming an AI infrastructure unicorn?

The first risk is design-win timing. Eliyan may have strong technology, but revenue depends on customers incorporating that technology into real products. Delayed design cycles could slow growth.

The second risk is customer concentration. A small number of hyperscalers, AI chip developers or infrastructure companies could drive a large share of revenue. That can accelerate growth but also increase volatility.

The third risk is ecosystem dependence. Eliyan’s products must work with foundries, packaging providers, memory suppliers and optical partners. Failure at any integration point can delay adoption.

The fourth risk is competition from established suppliers. NVIDIA, Broadcom, Marvell and other large semiconductor infrastructure companies have deeper customer relationships and stronger balance sheets.

The fifth risk is standards evolution. Universal Chiplet Interconnect Express and other interconnect standards will continue evolving. Eliyan must stay aligned without losing differentiation.

The sixth risk is manufacturing scale. Chiplet products require production execution, test capacity, quality control and supply assurance. That is a different challenge from licensing IP.

The seventh risk is valuation pressure. A $1 billion valuation is reasonable if Eliyan reaches hundreds of millions of dollars in sales by 2027. It becomes harder to defend if adoption ramps more slowly or remains concentrated in evaluation-stage programmes.

The eighth risk is AI capex cyclicality. If hyperscalers pause or slow data-centre spending, component suppliers across the stack can face order volatility.

What should investors and competitors watch after Eliyan’s Series C?

The first milestone is chiplet shipments. Eliyan expects initial shipments this year, and that will show whether the company is moving from IP validation to product delivery.

The second milestone is customer disclosure. Named deployments, hyperscaler relationships or public design wins would significantly strengthen the commercial case.

The third milestone is revenue ramp. The difference between low millions in 2025 and hundreds of millions by the end of 2027 is large. Investors should watch whether bookings and shipments support that trajectory.

The fourth milestone is electro-optical product progress. Eliyan’s expansion into optical interconnects could widen its market if the technology demonstrates power and bandwidth advantages.

The fifth milestone is strategic partner conversion. Cisco and Lumentum are valuable investors, but commercial collaboration would matter more than cap-table participation.

The sixth milestone is manufacturing scale. Chiplet products must be delivered reliably and at volume, especially if customers depend on them for large AI infrastructure deployments.

The seventh milestone is competitive response. Broadcom, Marvell, NVIDIA and other suppliers will not ignore an independent interconnect player if customers show meaningful interest.

Eliyan has raised capital at the right moment in the AI infrastructure cycle. The industry is discovering that the next bottleneck is not only building more compute. It is making that compute useful by moving data fast enough, cheaply enough and efficiently enough. Eliyan’s challenge is to prove that its interconnect technology can become a standard part of that next architecture.

Key takeaways on what Eliyan’s $145 million Series C means for AI infrastructure

• Eliyan has raised $145 million in Series C funding at a $1 billion valuation.
• Seligman Ventures led the oversubscribed round.
• Cisco Investments and Lumentum joined as new strategic investors.
• Umesh Padval, Managing Partner at Seligman Ventures and a former Mellanox board member, is joining Eliyan’s board.
• Eliyan is developing interconnect technologies for next-generation AI compute, memory and networking systems.
• The company’s NuLink PHY and NuGear chiplet products target die-to-die, chip-to-chip, memory expansion and rack-to-rack connectivity.
• Eliyan expects initial chiplet shipments this year and is targeting a major revenue ramp by the end of 2027.
• The company is positioning itself as an independent option for custom AI chip developers and infrastructure companies.
• Cisco and Lumentum’s participation strengthens the networking and optical infrastructure angle.
• The biggest risks are customer adoption timing, manufacturing scale, competition from larger suppliers, standards evolution and valuation pressure.