SoftBank and Intel have announced a collaboration to create a new AI-focused high-bandwidth memory (HBM) alternative through a joint venture named Saimemory.

Initial funding for Saimemory is set at approximately $70 million, with potential additional support from Japan's Ministry of Economy, Trade and Industry.

The new memory aims to address the increasing energy consumption and throughput demands in AI data centers, making large-scale AI training more efficient and cost-effective.

Saimemory plans to develop a stacked-DRAM solution that could halve power consumption compared to existing HBM technology, leveraging Intel's expertise and academic patents from institutions like the University of Tokyo.

This partnership seeks to alleviate memory bottlenecks that hinder large-scale AI models, as current memory technologies like DDR4/5 fall short of the bandwidth requirements for AI workloads.

This initiative reflects a broader shift in the tech landscape, recognizing memory as a critical component in AI infrastructure alongside traditional CPU development.

Establishing stronger domestic supply chains could help Japan reduce its reliance on Korean memory suppliers, particularly Samsung and SK Hynix.

The memory market is highly competitive, with South Korean and Taiwanese companies leading; Intel must tackle challenges in cost, efficiency, and supply chain management to succeed.

The timeline for Saimemory's development poses significant challenges, as competitors like Samsung and SK Hynix are several product generations ahead in the HBM market.

A prototype of the new memory is expected to be ready within two years, with commercialization targeted before 2030.

Intel's involvement in Saimemory marks its significant re-entry into the memory chip industry after over 20 years, coinciding with a renewed focus on memory solutions for AI data centers.

The success of this venture could redefine Intel's role in the semiconductor industry and mark a significant step in the evolution of AI memory architecture.