Intel (INTC.US) CEO Pat Gelsinger recently signaled a potential return to the memory chip market, marking a significant strategic pivot for the company. This move could expand its product portfolio and reshape its competitive dynamics and partnerships within the semiconductor industry.
During a podcast appearance on August 11, Gelsinger revealed that new memory architectures, previously dismissed as "commodity businesses," have now become strategically important areas and a key focus for him. He noted that the memory industry is in a critical period of innovation and hinted that Intel is exploring ways to stack memory directly on top of CPUs. After years of commoditization, memory chips have re-emerged as a strategic asset in recent quarters, a trend that may persist, generating substantial profits for major memory manufacturers—a development Gelsinger is closely monitoring.
"I used to think, 'Don't invest in memory because it's a commodity business,' but that's no longer the case," Gelsinger stated. "There are a lot of new technologies emerging now. So, we're looking at some new memory architectures, and that's one of my key focus projects." He also referenced the appointment of former SK Hynix CEO Lee Seok-hee, who joined Intel in June as an executive vice president for the foundry business, overseeing advanced packaging, system integration, and back-end technology development. Gelsinger suggested that observers could "probably guess" what direction he is considering, though he is not ready to disclose further details.
Interestingly, Intel began as a memory chip company upon its founding in 1968, achieving notable success in the field. However, by the 1980s, Japanese competitors gained dominance, leading to severe losses that forced Intel to exit the memory market entirely. The company has since attempted multiple comebacks, including ventures into NAND, Optane memory, and RDRAM technology, but each time it ultimately abandoned these efforts without significant financial damage.
This year, Intel has been active in memory and advanced packaging. In February, it partnered with SoftBank-backed SAIMEMORY to develop ZAM (Z-Angle Memory), a technology targeting AI and high-performance computing by balancing memory capacity, bandwidth, and power consumption. Prototypes are expected by the 2027 fiscal year ending March 2028, with commercialization planned for the 2029 fiscal year. ZAM partially builds on Intel's previous NGDB (Next-Generation DRAM Bonding) project, which successfully tested 8-layer DRAM vertical stacking. In July, a patent application for Cross-Batch Memory (XBM) was disclosed. XBM uses DRAM with back-end-of-line (BEOL) transistors and serial UCIe links, replacing HBM's ultra-wide parallel interface. This design could reduce package size, eliminate the need for silicon interposers, and lower assembly complexity and cost. Intel is also advancing complementary technologies like EMIB, Foveros, and the 18A-PT process for 3D integration.
These moves indicate that Intel is not simply reviving traditional DRAM or NAND production but aiming to find a new entry point in memory architecture. As large model training and inference demand higher data throughput, data movement has become a performance and power bottleneck, driving the industry beyond HBM toward tighter chip stacking and system integration. Gelsinger directly discussed stacking CPUs with memory chips, suggesting new combinations that could shorten the distance between CPU and memory through architectural changes.
For investors, Intel's potential return to the memory market is tied to its core investment thesis of refocusing its business portfolio and rebuilding trust in its role in AI infrastructure. Adding memory could help Intel offer a more complete platform for emerging AI workloads, rather than just CPUs and foundry capacity. Given the current profitability of 3D NAND and DRAM manufacturers, producing memory chips is again a lucrative business, likely maintaining high margins for some time. However, this also introduces risk to the existing investment thesis. Intel's complex organizational structure, high operating and capital expenditures, remain core concerns. Re-entering the memory market would require substantial capital investment, including at least one fab, R&D for competitive manufacturing processes, and a long timeline to scale. This capital-intensive endeavor will test how much Intel can simplify its operations.
A key metric for investors will be how Intel positions memory in its broader capital planning and AI product roadmap during future earnings calls and industry events. Clarity on planned investment, target customers (e.g., data center and AI), and whether it will use existing fabs or build new capacity will indicate whether this strategic shift supports the "refocusing" thesis or adds execution and complexity risks.
However, multiple media outlets suggest the outlook for Intel's return to the memory market remains uncertain. Tom's Hardware noted that a full-scale return to manufacturing memory chips would require not only re-investing in fabs but also developing competitive manufacturing processes, with long construction and validation cycles. Given Intel's ongoing commitments to CPU products and foundry business, its willingness to invest heavily in traditional memory chips is questionable. ZDNet also argued that Gelsinger's remarks should not be interpreted as Intel preparing to revive general DRAM operations, but rather as an increased focus on next-generation memory technologies. Even if new architectures are commercialized, entering the AI market is challenging. Nvidia currently holds over 80% of the AI accelerator market, with its hardware and software ecosystem built around HBM, making it difficult for new architectures to replace HBM in the short term. Even if ZAM reaches mass production, it may first appear in custom AI chips or inference-focused products as a complement to HBM.
While Gelsinger acknowledged that plans to return to the memory chip market are not finalized, with no timeline, product roadmap, or capital commitments revealed, his statements have opened new possibilities for Intel's turnaround story. After missing several technological eras, the former Silicon Valley giant is attempting a late but critical "return" in the AI era by merging computing and memory through "new architecture."
Comments