SoftBank and Intel Team Up to Build Z-Angle Memory for AI

The race to scale AI infrastructure just found a new contender. SoftBank Corp. and Intel Corporation have formally teamed up to build and commercialize a next-generation memory technology called Z-Angle Memory (ZAM), a design aimed squarely at the exploding power and cost demands of AI data centers.

The agreement, announced on February 3, 2026, brings Intel back into the memory arena after years on the sidelines and gives SoftBank a direct shot at reshaping the economics of AI hardware.

A new vehicle for a strategic push

The partnership is structured around SAIMEMORY, a Tokyo-based subsidiary created by SoftBank in December 2024 specifically to develop advanced memory IP. SoftBank remains the largest shareholder, having invested roughly ¥3 billion at inception.

The leadership mix reflects how seriously both sides are taking the effort: a former Toshiba executive oversees operations as CEO, Intel controls the technical roadmap through the CTO role, and a University of Tokyo scientist leads fundamental research as Chief Science Officer.

This deal also builds on SoftBank Group’s $2 billion purchase of Intel stock in mid-2025, signaling that the memory project is part of a broader, long-term alignment rather than a one-off experiment.

What makes Z-Angle Memory different

ZAM is positioned as an alternative to today’s High-Bandwidth Memory (HBM), which has become essential for AI accelerators, but also expensive, power-hungry, and difficult to manufacture at scale.

Instead of relying on the complex interposer-heavy HBM process, ZAM uses 3D-stacked DRAM with a proprietary “Z-Angle” wiring and bonding structure. The approach is built on Intel’s Next Generation DRAM Bonding (NGDB) technology, previously validated under the U.S. Department of Energy’s Advanced Memory Technology program.

The targets are aggressive:

• 40–50% lower power consumption compared with current HBM
• More than double the capacity of leading-edge DRAM chips
• Lower production costs, achieved by stacking standard DRAM dies rather than manufacturing specialized HBM stacks

If those numbers hold, ZAM could directly address two of AI’s biggest constraints: energy budgets and memory density.

A long but defined roadmap

The project is still firmly in the R&D phase, but the companies have laid out a multi-year path to market.

• Prototype chips are expected by the fiscal year ending March 31, 2028
• Commercial deployment is targeted for FY2029 (ending March 31, 2030)

SAIMEMORY will retain control over design and intellectual property, while large-scale manufacturing is expected to be outsourced, most likely to foundries in Taiwan or potentially to Japan’s emerging advanced chipmaker Rapidus.

Why investors are paying attention

The announcement triggered an immediate market reaction. Intel shares jumped 5%, while SoftBank Corp. rose 3.13%, reflecting optimism that ZAM could loosen the grip of the current memory heavyweights, Samsung, SK Hynix, and Micron, on AI-grade memory supply.

For Intel, the project marks a symbolic return to memory after exiting NAND and Optane in 2022. For SoftBank, it’s a calculated infrastructure play: if AI models keep growing, memory efficiency may matter as much as raw compute.

The bigger picture

Z-Angle Memory isn’t about incremental gains. It’s a bet that the next phase of AI scaling won’t be unlocked by more GPUs alone, but by rethinking how data is stored, stacked, and moved inside the data center.

If ZAM reaches production on schedule, it could redefine the cost-power equation of AI hardware, and give both SoftBank and Intel a rare chance to shape a foundational layer of the AI era.