Intel Revives Next-Generation Memory Push for AI Chip Stacking

Artificial-intelligence accelerators increasingly face a “memory wall”: compute engines are advancing faster than data can be moved between processors and memory, making bandwidth, power use and packaging costs critical constraints. Intel, founded as a memory-chip company before retreating from major memory businesses, is returning through XBM and Z-Angle Memory, or ZAM, alongside its EMIB and Foveros packaging platforms. Tighter integration of logic and memory could give Intel a route into an AI supply chain now built largely around high-bandwidth memory.

Intel Chief Executive Lip-Bu Tan said in August that the company was examining new memory architectures and ways to stack memory with CPUs. Intel on June 19 appointed former SK hynix Chief Executive Seok-Hee Lee to oversee advanced packaging. A patent published July 2 outlined XBM, which uses UCIe links and removes HBM’s silicon interposer. Intel also signed a February 2 agreement with SoftBank Corp.’s SAIMEMORY to develop ZAM, targeting a prototype in the fiscal year ending March 2028 and commercialization in fiscal 2029. No investment amount was disclosed.