Intel CEO Signals Potential Return to Memory Chip Manufacturing

What happened

During a TechSurge podcast appearance, Intel CEO Lip-Bu Tan discussed the company's interest in new memory architectures, specifically highlighting the recruitment of Seok-Hee Lee, former CEO of SK Hynix, as executive vice president of Intel Foundry. Tan described memory innovation as 'one of my pet projects' and noted that memory has shifted from being a commodity business to one with significant strategic value. Intel has filed a patent for XBM (cross-batch memory), a high-bandwidth memory architecture that uses UCIe (Universal Chiplet Interconnect Express) links running at 32 gigatransfers per second instead of the traditional parallel interface requiring costly silicon interposers. The patent describes moving DRAM transistors to back-end-of-line (BEOL) metal layers using thin-film transistors, enabling denser stacking and more compact designs. Intel is also co-developing ZAM (Z-Angle Memory) with SAIMEMORY, a SoftBank subsidiary, which uses fusion bonding to create a nine-layer DRAM stack targeting roughly twice HBM4's bandwidth density. Tan stated Intel is 'not ready to unfold' specific plans but acknowledged the company is evaluating CPU-memory stacking technologies.

Why it matters

Memory has become a critical bottleneck in AI infrastructure, with current projections indicating memory supply will be effectively sold out through 2027 or 2028. This shortage has transformed memory from a commodity into a strategic asset, with memory prices projected to climb approximately 130% by the end of 2026 according to Gartner. For Intel, re-entering memory could provide vertical integration advantages in an AI-dominated semiconductor landscape where performance is increasingly limited by memory bandwidth rather than compute power. The hiring of Seok-Hee Lee, who led SK Hynix during its development of the high-bandwidth memory now used in major AI accelerators, signals Intel's seriousness about advanced memory and packaging technologies. Intel's EMIB-T advanced packaging has already been selected by Google for its ninth-generation TPU, positioning Intel as an alternative to TSMC's oversubscribed CoWoS packaging line. If XBM reaches production, it could allow memory to attach to any UCIe-compatible processor without specialized interposers, reducing dependency on constrained packaging capacity.

Bigger picture

Intel originally founded in 1968 as a memory company and its 1103 DRAM chip became the world's first commercially successful dynamic RAM in 1970, capturing roughly 83% of the DRAM market by 1974. The company exited DRAM in 1985 under pressure from Japanese manufacturers who achieved yields as much as 40% higher than American firms, reducing Intel's market share to 2-3% by 1984. Intel's subsequent memory ventures, including RDRAM, 3D NAND, and Optane, failed to establish durable businesses, with the company selling its last significant memory operation (NAND) to SK Hynix in 2020. Today, SK Hynix, Samsung, and Micron collectively control memory supply and pricing in a market transformed by AI demand. The convergence of a CEO commitment, specific technical patents, and executive talent acquisition represents Intel's most coherent memory strategy in decades, though actual re-entry would require dedicated fabrication facilities, competitive process technology, and years of manufacturing ramp-up while the foundry business lost more than $10 billion in 2025.

What to watch

Watch for any formal announcement of Intel's memory strategy, including capital allocation decisions and facility plans. The commercialization timeline for XBM is not expected before 2030, while ZAM prototypes are targeted for fiscal 2027 with commercialization in fiscal 2029. Monitor Intel's advanced packaging progress, particularly EMIB-T customer wins and SK Hynix's testing of Intel's EMIB for HBM integration. Intel reported Q2 2026 revenue of $16.1 billion, up 25% year-over-year with Data Center and AI Group revenue up 59% to $6.3 billion, and priced a $20 billion equity offering for capital expenditures and working capital. Key signals include whether Intel commits formal resources to memory development, how the company balances foundry recovery priorities against new memory investments, and whether BEOL-transistor DRAM and fusion bonding technologies prove viable at manufacturing scale.