Intel’s packaging roadmap targets much larger AI chips

Intel’s advanced-packaging plans point towards chip packages many times larger than today’s, with denser connections and more capable power delivery for future AI systems. The key question is whether its roadmap can turn those ambitions into reliable, high-volume production on schedule.

Intel AI Watch analysis

What happened

Tom’s Hardware examines Intel’s EMIB and Foveros packaging roadmaps. EMIB-T is described as targeting mass production in 2027, with finer-pitch connections and support for packages designed to reach 12 reticles and more than 24 HBM stacks. Intel CEO Lip-Bu Tan told investors that EMIB-T’s yield and reliability are hitting targets and that the company is focused on supporting customer ramps in 2027.

The article also says Intel’s Foveros-R is expected to be production-ready in 2027. Foveros-B, aimed at larger and more complex AI and high-performance computing designs, has no disclosed commercial production date. These are roadmap plans and company statements, not proof that every target will be met.

Why it matters

AI accelerators need more than fast compute dies: connecting them to one another and to high-bandwidth memory, while delivering power across large packages, is a major engineering challenge. Intel’s proposals combine local silicon bridges and other packaging approaches to scale connections without relying on a single, full-size silicon interposer for every design.

If the planned technologies reach production, they could give Intel more ways to compete for advanced AI-chip packaging work. The schedule and manufacturing execution matter more than the most impressive package diagram. Large designs are useful only if they can be made reliably and in volume.

Our read

This is a substantial roadmap, not a shipment announcement. The meaningful signal is Intel’s stated progress on EMIB-T’s yield and reliability, alongside a 2027 production target. The less settled part is how much of the broader roadmap, particularly Foveros-B, will make it into commercial products and when.

What to watch

• Whether EMIB-T enters mass production in 2027 as planned.
• Whether Intel reports customer ramps, yields and reliability beyond its current statements.
• When Intel gives Foveros-B a commercial production timeline.
• What package sizes and memory configurations customers actually adopt.

Discussion spark: Should chip buyers treat a packaging roadmap as a meaningful competitive advantage before the technology reaches volume production?

Independent WittyWires tracker for public updates about Intel AI. Not affiliated with or endorsed by Intel AI; this is not an official account.