SoftBank is officially entering the next phase of artificial intelligence (AI) memory development, signaling a potential shakeup in the global semiconductor landscape. Rather than manufacturing the chips itself, the Japanese tech giant is adopting an IP-centric approach—focusing on memory architecture design and seeking hardware partners for production. Samsung Electronics is widely seen as a leading candidate.
Speaking during SoftBank’s FY2024 earnings briefing on May 8, CEO Junichi Miyakawa announced the company will invest approximately $20.41 million over the next two years to develop high-performance, low-power memory optimized for AI inference. “As AI shifts toward inference at scale, we are preparing next-generation memory technology that meets the demands of performance and efficiency,” he said.
Unlike traditional chipmakers, SoftBank’s strategy revolves around securing memory IP and collaborating with external partners for fabrication. Industry analysts suggest Samsung is best positioned to support this strategy, given its broad semiconductor portfolio and recent moves signaling closer ties between the two companies.
SoftBank’s deepening AI push follows increased cooperation with OpenAI and Samsung through a joint project called “Stargate,” a massive AI infrastructure initiative centered in Osaka, Japan. The trio—SoftBank Chairman Masayoshi Son, Samsung Chairman Lee Jae-yong, and OpenAI CEO Sam Altman—reportedly met in February to discuss the $493.41 billion initiative, which includes the construction of a hyperscale AI data center.
To support Stargate, SoftBank acquired a Sharp LCD plant in March for approximately ¥100 billion and is actively recruiting global partners for AI-specific hardware design and manufacturing. Many in the industry believe Samsung could play a pivotal role in memory and chip development for the project.
Samsung stands out not only for its manufacturing scale but also for its vertical integration across the semiconductor supply chain. From high-bandwidth memory (HBM) for AI acceleration to advanced system-on-chip (SoC) design and sub-5nm foundry capabilities, few companies globally can match Samsung’s breadth.
Samsung is also ramping up production of next-gen AI memory solutions such as HBM3E and HBM5, with customers including AMD, Broadcom, and Marvell. These capabilities make it a natural fit for SoftBank’s IP-driven strategy.
Lee Jae-yong’s recent visits to Japan lend further weight to speculation about a deepening partnership. Last month, he spent a week inspecting supply chain operations and meeting with key figures in Japan’s tech and business sectors. He is now returning to Japan to attend the Korea Day event at the Osaka-Kansai Expo—Osaka being the core site of the Stargate project. A meeting with Masayoshi Son during this visit could further solidify AI collaboration plans.
Meanwhile, SoftBank reported operating profit of $6.73 billion in FY2023, up 13% from the previous year. With AI now at the core of its future growth strategy, the company has raised its FY2024 profit target to $6.80 billion.
