SoftBank Subsidiary to Co-Develop ZAM Memory with Intel, Secures Japanese Government Subsidy

A Revolution in Memory for the AI Era: The Full Picture Behind ZAM Memory by SoftBank and Intel

On April 24, 2026, a quiet stir swept through the tech industry. SaiMemory, a subsidiary of SoftBank Group specializing in memory technology, announced that its next-generation DRAM architecture, “Z-Angle Memory (ZAM),” which is being developed in collaboration with U.S.-based Intel, had been awarded a subsidy from Japan’s New Energy and Industrial Technology Development Organization (NEDO). The backdrop to this announcement is the tech sector’s growing urgency to address the skyrocketing energy consumption and heat management challenges of existing high-bandwidth memory (HBM) in the face of explosive growth in AI computing needs.

The Limitations of HBM and the Future Promised by ZAM

Currently, HBM is the “fuel” powering GPU accelerators that drive AI training and inference. This technology, primarily led by SK Hynix and Samsung Electronics, stacks multiple DRAM dies vertically to achieve high-bandwidth data transfer. However, its success has also become a double-edged sword. As AI models continue to expand, HBM’s power consumption has surged, creating severe challenges for data center cooling costs and power supply. Industry insiders now believe HBM is nearing its performance limits.

Enter ZAM. The “Z-Angle” in its name hints at a novel approach distinct from traditional stacking methods. While detailed technical specifications have not yet been disclosed, sources suggest that ZAM is designed to drastically reduce power consumption compared to HBM while maintaining comparable bandwidth performance. As the industry envisions a future where AI workloads can operate without energy constraints, ZAM holds the potential to be the key to unlocking that future.

The Unlikely Partnership Between SoftBank and Intel

The development of ZAM is the product of an unlikely partnership between SoftBank and Intel. While SoftBank is better known as an investment firm with interests in AI and robotics, CEO Masayoshi Son has consistently emphasized the critical importance of semiconductors. Establishing SaiMemory and forming a technical partnership with Intel represents a strategic move to secure a hardware foundation for its AI ecosystem.

For Intel, this collaboration offers a prime opportunity to leverage its semiconductor manufacturing capabilities and memory technology to strengthen its presence in the AI market. While Intel has long dominated the CPU market, it has lagged behind NVIDIA in the AI-focused GPU and accelerator space. If Intel can mass-produce a groundbreaking memory technology like ZAM in its foundries, it could overcome the “memory wall” that influences AI chip performance.

The Strategy Behind Japan’s NEDO Subsidy

Japan’s decision to subsidize this development reflects a larger national strategy to revive its semiconductor industry. Once a global leader in semiconductors, Japan lost its footing in the DRAM production race and exited the market. However, the advent of the AI era has reignited interest in semiconductors as a strategic asset.

NEDO’s subsidy is more than just financial support for technology development. It is a national investment in rebuilding Japan’s semiconductor supply chain and securing AI-era infrastructure domestically. If ZAM succeeds, Japan could re-establish itself as a global hub for next-generation memory technology. Moreover, domestic companies like SoftBank and Sony Group could collaborate to create an ecosystem covering everything from AI chips to memory and system-level solutions.

Industry Implications: Redefining Power Consumption and Costs

The potential impact of ZAM’s development is immense. Firstly, it directly addresses the power consumption challenges faced by data centers. Today, the electricity consumed by AI computing accounts for several percent of global power generation, raising significant environmental concerns. If ZAM can reduce power consumption by 30–50% compared to HBM, it could dramatically lower data center operating costs while also contributing to the sustainability of AI technologies.

Secondly, it would broaden the design flexibility of AI hardware. Traditional HBM requires close proximity to GPU chips, creating numerous design constraints. If ZAM achieves high performance with low power consumption, it could enable more flexible designs for chip placement and cooling systems, potentially leading to smaller, more efficient AI accelerators.

Lastly, it could shift the competitive dynamics of the memory market. Currently, the HBM market is dominated by SK Hynix and Samsung. If ZAM emerges as a viable alternative, it could intensify price competition, ultimately benefiting AI developers.

Challenges and Future Prospects

Of course, ZAM development comes with significant challenges. First, the technical feasibility of achieving both low power consumption and high performance through its architecture is no small task. Integrating it with Intel’s cutting-edge manufacturing processes could also pose difficulties. Furthermore, reducing production costs will be crucial for competing against HBM.

Another hurdle is market adoption. The current AI hardware landscape is heavily optimized for existing HBM designs. Adopting ZAM will require developing compatible software stacks and tools. SoftBank and Intel will need to focus on building an ecosystem that secures support from AI developers.

Looking forward, prototype chip evaluations are expected to begin between 2027 and 2028, with mass production potentially commencing around 2030. If NEDO’s subsidy accelerates development, commercialization could occur even sooner.

Conclusion: A Step Toward the Future of Memory

The development of ZAM memory is more than just a technical story; it is a potential answer to one of the industry’s biggest challenges—how to design the infrastructure for the AI era. The collaboration between SoftBank and Intel is a fascinating intersection of investment and engineering, with Japan’s industrial policy intricately woven into the narrative.

While it remains uncertain whether ZAM will replace HBM as the new standard, its ability to address AI computing’s energy challenges is undeniably significant. The evolution of memory technology represents an “invisible revolution” that will shape our digital future and expand the capabilities of AI.