SoftBank and Intel Enter the Storage-Memory Arena: Low-Power “HBM” Targeted for Commercialization by Fiscal 2029

AI infrastructure spending is helping to ignite a new “supercycle” in the memory and storage market, with both investors and tech giants racing to secure a foothold. On Tuesday, SoftBank Group said its subsidiary, Saimemory, has reached a partnership with Intel to jointly develop a next-generation memory technology.

The two companies signed a collaboration agreement on February 2 to commercialize ZAM (Z-Angle Memory), a new memory architecture designed to deliver high capacity, high bandwidth, and lower power consumption—features increasingly critical for modern AI workloads.

Under the agreement, Saimemory will leverage Intel’s next-generation DRAM (dynamic random-access memory) assembly technology and aims to complete a prototype by early 2028. More specifically, Saimemory plans to build a prototype in fiscal 2027 and target commercialization in fiscal 2029.

SoftBank said the technology is being developed with AI data centers in mind. As running generative AI models demands enormous compute resources, high-speed memory has become one of the most essential building blocks in AI-focused data centers.

“ZAM, our next-generation memory technology, will provide data centers and other environments that require large-scale AI model training and inference with high-capacity, high-bandwidth data processing, higher performance, and lower power consumption,” SoftBank said in a statement.

As demand for AI training compute grows exponentially, energy use in data centers has become a major pain point across the industry. While high-bandwidth memory (HBM) can significantly improve data throughput, its power draw remains a difficult challenge. Some estimates suggest that in Google’s AI data centers, HBM can account for as much as 35% of total memory power consumption, while cooling costs continue to climb year after year.

SoftBank and Intel’s joint effort is positioned squarely at this bottleneck. The partnership signals that AI hardware may be heading toward a meaningful shift—one centered on energy efficiency as much as raw performance.

Saimemory itself is a relatively new SoftBank subsidiary, established in December 2024 to drive research, development, and real-world deployment of next-generation memory technologies. After expressing interest in working with Intel, the company was renamed in mid-2025. From the start, the project has aimed to deliver performance comparable to today’s mainstream HBM, while cutting power consumption by more than half.

Earlier reports noted that SoftBank, as a key backer of the initiative, has committed an initial investment of ¥3 billion, making it the largest contributor at the early stage. Total project costs are estimated at ¥10 billion (roughly $70 million). If Saimemory successfully brings its product to market, SoftBank will receive priority supply access.

For Intel, the partnership carries strategic weight as well. The announcement comes at a time when the company is working to strengthen its chip portfolio and close the gap with competitors in AI-focused hardware.

It also represents a notable return to the memory space for Intel. In 2022, Intel decided to exit the Optane business and sold its NAND flash unit to SK hynix, though it retained related 3D XPoint technology and patents.

Analysts point out that while Saimemory is focused on DRAM rather than Optane’s non-volatile storage path, Intel’s expertise in advanced packaging and chip stacking could provide meaningful technical support for Saimemory’s development roadmap.