Eliyan, a Silicon Valley semiconductor startup developing advanced chiplet interconnect technology, announced on January 29, 2026, that it has raised $50 million in strategic financing led by Samsung Catalyst Fund and Intel Capital. The round included participation from AMD, ARM, Coherent Corporation, and Meta, signaling broad industry recognition that chiplet architectures represent critical pathways for scaling AI computing capabilities.
The investment reflects semiconductor industry consensus that traditional chip designs face fundamental constraints as transistors approach atomic dimensions and manufacturing costs skyrocket at advanced nodes. Chiplet architectures—where multiple smaller chips interconnect to function as unified systems—offer alternative pathways to increase performance, reduce costs, and improve manufacturing yields.
Eliyan’s core products, NuLink and NuGear, target critical bottlenecks in scale-up AI servers: memory bandwidth and input/output limitations. Modern AI workloads are increasingly constrained by data movement speed between processing units and memory rather than raw computational throughput. Chiplet architectures potentially address these bottlenecks by enabling closer integration of memory and compute, reducing latency and power consumption while increasing system bandwidth.
The strategic investor composition is particularly notable. Samsung, as the world’s largest memory chip manufacturer, has direct interest in enabling architectures that efficiently utilize high-bandwidth memory products. Intel sees interconnect technology as essential infrastructure for future product roadmaps transitioning from integrated device manufacturing toward diversified models including foundry services and chiplet-based designs.
AMD has already commercialized chiplet designs in EPYC server processors and Instinct AI accelerators, providing market validation for the approach and potential customer relationships for suppliers like Eliyan. ARM’s involvement reflects the architecture’s expansion beyond mobile devices into data centers where chiplet designs may enable more flexible server CPU configurations.
Meta’s participation reflects the company’s massive AI infrastructure buildout announced concurrently. As Meta deploys hundreds of billions of dollars constructing data centers and procuring AI accelerators, any technology improving performance per watt or reducing infrastructure costs at scale could generate enormous value. If Eliyan’s interconnect technology enables more capable AI systems within existing power envelopes, it directly supports Meta’s ability to train larger models and serve more inference requests without proportionally scaling energy consumption.
The technical challenge Eliyan addresses is substantial. When chips are manufactured as single pieces of silicon, components communicate at extremely high speeds with low latency through substrate connections. When multiple chips must communicate through package-level connections, interfaces introduce latency, power consumption, and bandwidth limitations. Eliyan’s value proposition is that its interconnect technology minimizes these penalties, approaching monolithic integration performance while preserving chiplet architecture benefits.
The $50 million fundraise follows a $40 million Series A in 2024, bringing total funding to $90 million. For hardware-focused startups addressing complex semiconductor industry problems, this represents substantial capital enabling multi-year development cycles, extensive validation and testing, engagement with manufacturing partners, and customer design-ins that may take years from initial discussions to production deployment.
The competitive landscape includes Intel’s Universal Chiplet Interconnect Express (UCIe) representing industry standardization efforts, AMD’s Infinity Fabric connecting current chiplets, and proprietary technologies from Nvidia and Google serving internal needs. Eliyan must differentiate through superior performance, better economics, easier integration, or combinations making its solution preferable to alternatives.
Industry analysts project chiplet adoption could reach 50% or higher in high-performance computing by 2030, driven by economic pressures, manufacturing yields, and ability to mix different process nodes for optimal cost-performance.
