SoftBank and Intel Join Forces to Develop New Low-Power Memory for AI
SoftBank and Intel are teaming up to create a novel type of high-bandwidth memory (HBM) tailored for artificial intelligence applications. Their goal? To deliver a product that can stand up against the offerings from South Korea’s major players, Samsung and SK Hynix.
A recent report from Nikkei Asia reveals that this ambitious project will be spearheaded by a new venture named Saimemory. The plan includes the development of stacked DRAM chips that promise to cut power consumption by 50% compared to current HBM options. A prototype is expected within the next two years, with commercialization targeted before 2030.
A Challenging Landscape
While the technical goals are impressive, the timeline presents a significant hurdle. Samsung and SK Hynix have already secured a substantial lead in the global HBM market, launching several product generations ahead of the newcomers. By the time Saimemory’s offerings hit the shelves, existing competitors will likely have strengthened their dominance even further.
A SoftBank official mentioned that success in this initiative would be crucial: “If the project is successful, we want priority for the supply,” highlighting how early this venture remains in its development.
Targeting AI Data Centers
The new memory chips are designed with AI data centers in mind, addressing growing concerns over energy consumption and the demand for high throughput. SoftBank envisions these chips enabling large-scale AI training to be conducted more efficiently and cost-effectively. Additionally, boosting local supply chains could help Japan reduce its reliance on South Korean manufacturers.
Historically, Japan dominated the global DRAM market, capturing over 70% at its peak. However, as competitors in South Korea and Taiwan gained ground, Japan’s involvement waned. Elpida, Japan’s last DRAM producer, went bankrupt in 2012 and was acquired by Micron, which is now eyeing a larger share of the booming HBM market.
Key Players and Investment
SoftBank leads the investment in Saimemory with a 3 billion yen stake in a total project valued at 10 billion yen. Intel is contributing its technological know-how, while universities like the University of Tokyo are providing essential patents. Additionally, government support is being explored to further bolster the initiative.
Despite the promise this project holds, both SoftBank and Intel must juggle multiple priorities. Intel is currently focused on expanding its foundry capabilities, advancing AI chip development, and regaining its footing in the CPU market. Meanwhile, SoftBank is working on growing its Arm assets and diversifying its AI investments.
As this collaboration unfolds, industry watchers will be keen to see how Saimemory’s ambitions materialize amidst a rapidly evolving tech landscape.
