A "Comeback" or a "Money Pit"? Intel (INTC.US) Seeks to Return to the Storage Market, Betting on New Architecture to Gain a Foothold in the AI Era

According to Zhitong Finance APP, Intel (INTC.US) CEO Pat Gelsinger has recently signaled that the company may return to the storage chip market. If this move materializes, it would mark a strategic shift for Intel, potentially expanding its product portfolio and impacting its competitive approach and collaborations within the broader semiconductor industry.

Reportedly, on August 11 local time, Gelsinger revealed in a podcast that new types of storage architectures, once considered a "commodity business," have now become a strategically significant field and are one of his core areas of focus. He also stated that the storage industry is experiencing a key moment of innovation and implied that Intel is exploring ways to stack storage above CPUs.

After years of commoditized competition, storage chips have once again become strategic assets in recent quarters—a trend likely to persist for some time. This has led to exceptionally high profits for the world's major storage chip manufacturers. Gelsinger is clearly paying close attention to these dynamics.

Gelsinger stated, "In the past, my thinking was, 'Don't invest in storage chips, because it's a commodity business,' but now it's different." "There's a lot of new technology emerging. So we're focusing on some new storage architectures, which is one of my key focus areas."

Gelsinger also mentioned Lee Seok-hee, the former SK Hynix CEO who joined Intel in June. Lee currently serves as Intel's Executive Vice President of Foundry Services, overseeing advanced packaging, system integration, as well as backend technology development and manufacturing. While discussing this arrangement, Gelsinger remarked that outsiders can "probably guess" the direction he's considering, though he is not ready to disclose more details at this time.

In fact, Intel originally started as a storage chip company. After its founding in 1968, Intel achieved considerable success in the storage chip sector. However, by the 1980s, Japanese companies began to take the lead, inflicting severe losses on Intel and ultimately forcing its complete exit from the storage chip market.

Since then, Intel has made several attempts to re-enter the storage domain. The company ventured into NAND and Optane storage businesses and tried to seek new growth opportunities through emerging storage technologies like RDRAM. However, Intel eventually exited all three storage ventures, and without incurring particularly serious financial losses.

This year, Intel has continued to make moves in storage and advanced packaging. In February, Intel and SoftBank-owned SAIMEMORY announced a partnership to develop ZAM (Z-Angle Memory), a technology aimed at AI and high-performance computing that seeks to address the balance among memory capacity, bandwidth, and power consumption. A prototype is planned for completion in Intel’s 2027 fiscal year, ending March 2028, with commercialization expected in fiscal 2029. Some of ZAM’s technology originates from Intel’s earlier involvement in the NGDB (Next Generation DRAM Bonding) project, which has already completed tests on eight-layer DRAM vertical stacking and explores how improved DRAM stacking and interconnects can reduce the capacity sacrifices needed for high-bandwidth memory.

In July, media reported on Intel’s patent filing for Cross-Batch Memory (XBM). XBM still uses DRAM but replaces HBM’s ultra-wide parallel interface with backend-of-line (BEOL) transistors and serial UCIe links. This design eliminates the silicon interposer required for traditional HBM and allows for a smaller overall package with assembly complexity and costs lower than HBM4, all while occupying about the same package area as HBM4.

Additionally, Intel is continuing to advance the development of cutting-edge packaging and 3D integration technologies, such as EMIB, Foveros, and 18A-PT processes for 3D integration, all of which complement these storage architectures. Together, these moves show that Intel is not simply seeking to revive traditional DRAM or NAND production but rather looking for new entry points in storage architecture.

Historically, CPUs, GPUs, and storage chips have existed as relatively independent chips. However, as large model training and inference require higher data throughput, data transfer has become a new performance and power bottleneck, pushing the industry to explore closer chip stacking and system integration beyond HBM. Gelsinger directly discussed stacking CPUs with storage chips, arguing that new combinations are possible and that changing memory architectures can further shorten the physical distance between CPUs and storage.

For investors, Intel signaling a possible return to the storage market is directly related to its core investment logic of refocusing its business portfolio and rebuilding trust in its role within the AI infrastructure space. Bringing storage back into its product mix may help Intel achieve its goal of offering a more complete platform for emerging AI workloads, moving beyond only CPUs and foundry capacity.

Given the current profitability of 3D NAND and DRAM manufacturers, storage chip production has again become a lucrative business and is likely to remain so for some time. At the same time, this increases a risk within the existing investment thesis. Intel’s complex organizational structure, high operating expenses, and capital expenditures have long been focal concerns for the market. Reentering the storage market would require significant capital—at minimum to build a wafer fab—as well as research and development investment to develop competitive manufacturing processes, and enough time to reach scale. Rejoining capital-intensive businesses like storage will further test how much Intel can truly streamline its operations.

A practical metric for investors to track going forward is how Intel will position its storage business within its overall capital planning and AI product roadmap during future earnings calls and industry events. Whether the company can clarify its planned investment in storage, whether target customers are concentrated in data centers and AI, and whether it will use existing fabs or build new capacity—all these factors will reveal whether this strategic shift supports the “refocusing” narrative or adds new risks around execution and business complexity.

However, multiple media outlets point out that the outlook for Intel’s return to the storage market remains uncertain. Tom's Hardware, a prominent technology hardware media platform, notes that if Intel wants to resume large-scale storage chip manufacturing, it would not only need to reinvest in fabs, but also establish competitive manufacturing processes and endure long construction and validation cycles. Given that Intel must continue to invest in CPU products and foundry businesses, it’s uncertain whether the company is willing to commit much capital to traditional storage chips again. Tech media outlet ZDNet also believes Gelsinger’s comments should not be interpreted as Intel preparing to revive a general DRAM business, but more as a sign that the company is stepping up its investment in next-generation storage technologies.

Even if new storage architectures are commercialized, entering the AI market will not be easy. Some media note that Nvidia currently controls over 80% of the AI accelerator market, and its hardware and software ecosystem is already built around HBM, making it difficult for new storage architectures to fully replace HBM in the short term. Even if ZAM enters mass production, it is more likely to first appear in custom AI chips or inference-focused products, supplementing HBM rather than replacing it.

Although Gelsinger admits that plans to re-enter the storage chip market have not been finalized—there is no timetable, product roadmap, or capital commitment—his comments have opened a new chapter in Intel’s transformation narrative. After missing out on several eras, the one-time Silicon Valley giant is now seeking to launch a belated but crucial “comeback” in the AI era through "new architectures" that merge compute and storage.