Intel’s 14A/18A Super Chip: The AI “System Foundry” That Challenges TSMC

Intel’s 14A/18A “super chip” concept, unveiled in late 2025, is pitched as a full system foundry solution capable of rivaling TSMC’s CoWoS-led approaches. The design combines the 14A-E compute tiles with RibbonFET 2 and PowerDirect on the 14A era, alongside an 18A-PT base die that introduces PowerVia backside power delivery, all bound by Foveros Direct 3D stacking and EMIB-T interconnects.

In its most extreme configuration, the package can house 16 high‑performance compute dies and 24 HBM stacks, supported by a twelvefold reticle scalability that implies substantially larger chips than today’s single‑reticle limits. Intel emphasises this heterogeneous, multi-die approach as essential for data centres and AI model training, aiming to deliver an end‑to‑end system foundry encompassing packaging, stacking and testing.

The video release underscored a strategic intent to demonstrate mass‑production capability for 18A and 14A chips and to offer a complete service beyond manufacturing nodes alone. As industry players including NVIDIA, AMD and major cloud providers pursue in‑house AI silicon, Intel argues it can meet demand for advanced packaging technologies and high-bandwidth memory interconnects. 25 December 2025.