SoftBank and Intel have announced plans to collaborate on the development of a next-generation high-bandwidth memory chip, aiming to rival leading products from South Korea´s Samsung and SK Hynix. Their initiative, under the new company Saimemory, targets the exploding demand for efficient memory in Artificial Intelligence data centers—one of the most energy-demanding workloads in tech. Saimemory´s memory promises to halve the power consumption of today’s HBM, a critical advantage as hyperscale Artificial Intelligence adoption strains infrastructure and energy budgets worldwide.
According to reporting from Nikkei Asia, the project centers on a stacked DRAM chip employing a fresh wiring architecture that significantly improves energy efficiency compared to current industry leaders. The partnership plans to unveil a prototype within two years, with full commercialization targeted before 2030. SoftBank has invested 3 billion yen into the 10 billion yen project, securing its role as Saimemory’s largest investor. Intel is providing technical expertise, while academic collaborators, including the University of Tokyo, contribute key patents. Government assistance for the venture is also under consideration.
The ambition, however, faces stiff headwinds. Samsung and SK Hynix not only lead the HBM market but are already multiple generations ahead in development and production scale. By the time Saimemory is ready for commercial deployment, the gap may have widened even further, casting doubt on SoftBank and Intel’s prospects of breaking Korea’s entrenched dominance. This challenge is compounded by the companies´ existing commitments: Intel is juggling foundry expansion, Artificial Intelligence chip development, and a CPU comeback, while SoftBank continues to grow Arm and expand its Artificial Intelligence portfolio. The stakes, however, are high for Japan, which lost its DRAM leadership in previous decades. Saimemory is also intended to restore some domestic memory supply independence, reducing critical reliance on South Korean technology at a time when strategic resources are increasingly scrutinized in global trade and security contexts.
If successful, the project could deliver a much-needed low-power alternative for Artificial Intelligence data centers and potentially shift the global memory landscape. Yet, for now, Saimemory remains an ambitious play in a market dominated by well-resourced, technically advanced incumbents.
