Intel CEO Lip-Bu Tan is signaling a strategic re-entry into memory chips, describing new memory architectures as a 'pet project' and announcing the hiring of Seok-Hee Lee, former CEO of SK Hynix, as Executive Vice President of Intel Foundry (reporting directly to Tan). Lee led SK Hynix's HBM (high-bandwidth memory) development and oversaw the company's 2020 acquisition of Intel's NAND business—now he's returning to Intel as memory gains strategic urgency in AI.
Supporting the signal: Intel has filed a patent for XBM (cross-batch memory), a new architecture that eliminates the expensive silicon interposer used in standard HBM. XBM builds DRAM transistors directly into the chip's back-end-of-line metal layers and communicates via UCIe serial links, targeting HBM4 parity in footprint at potentially lower cost and power. Intel is also jointly developing Z-Angle Memory (ZAM) with Japan's SoftBank subsidiary SAIMEMORY—an 8-layer stacked-DRAM design aimed at high capacity, high bandwidth, and low power.
For architects: AI demand has transformed memory from commodity to bottleneck. Tan stated publicly 'there's no relief until 2028'—HBM supplies are oligopolized by SK Hynix (~60% share), Samsung, and Micron, with AI data centers consuming ~70% of global DRAM output. If Intel can ship a differentiated memory architecture (XBM earliest 2030+), it would break the incumbents' stranglehold and reshape cost economics for inference and training. The Lee hire signals serious intent, but past Intel memory ventures (Optane, RDRAM) failed—execution risk is high.