Eliyan, which develops NuLink to connect chiplet components more efficiently, raised a $40M Series A from Intel, Micron, and others
Context & Ripple Effects
Eliyan's $40M Series A is the opening move in what became one of the cleaner chiplet-interconnect funding arcs on record: the round brought Intel and Micron in as strategic backers of NuLink, a die-to-die interconnect aimed at moving data between chiplets more efficiently than standard paths allow. The strategic logic is visible on both sides — Eliyan needed foundry-grade credibility, while Intel and Micron sit exactly where the bottleneck bites, on compute and memory respectively.
The follow-on coverage confirms the thesis held: a $60M Series B co-led by Samsung Catalyst Fund and Tiger Global in 2024, then a $145M Series C at a $1B valuation in 2026, by which point Eliyan had shifted from pure interconnect IP toward licensing technology and selling physical chiplets to ease AI data-transfer bottlenecks.
First-order effects
- Intel and Micron convert capex into optionality: their checks buy early visibility into — and influence over — an interconnect layer that determines how fast their own chips and memory can be composed into larger AI systems.
- Eliyan exits stealth-stage scarcity with both capital and two anchor customers' ecosystems attached, letting it fund NuLink development without depending solely on neutral venture money.
Second-order effects
- Samsung's arrival as a Series B co-lead via Samsung Catalyst signals that strategic chip investors treat interconnect startups as contested ground — memory and foundry players will bid against each other for positions rather than cede the layer to rivals.
- Rivals building monolithic AI accelerators face a pricing argument they must answer: if chiplet-based designs close the performance gap while using cheaper process nodes, the interconnect becomes the differentiator rather than the silicon itself.
Third-order effects
- Eliyan's trajectory from interconnect IP to licensed technology and physical chiplets sketches an emerging industry structure: a merchant layer between compute and memory, analogous to how interface IP became a standalone business — with the risk that large foundry-backed players internalize it instead.
- If data movement, not compute, remains the binding constraint on AI chips, capital will keep flowing to companies attacking bandwidth and power per bit transferred — nEye's optical data-center chips being the same pattern in a different medium.
The trend: AI systems are pulling a dedicated interconnect layer out of the chip design stack, and the compute and memory giants are buying positions in it through startup equity rather than building it alone.