SoftBank founder Masayoshi Son approached Wall Street institutions about financing a possible $1 trillion purchase of SK hynix, one of Nvidia's important AI memory partners, according to The Information. The idea would have put a major source of high-bandwidth memory inside his expanding AI business.
The proposal has since been shelved. A SoftBank spokesperson told the publication that Masayoshi was no longer actively considering the transaction, leaving the reported approach as an exploration of financing rather than an announced takeover.
A trillion-dollar idea reaches Wall Street
The report, published November 21, cites people familiar with the matter and describes recent outreach to financial institutions. The $1 trillion figure refers to the possible acquisition, not a disclosed amount of borrowing or a financing package approved by banks.
That distinction leaves the central commercial question unresolved: how a purchase of that scale would have been funded. The publicly available account does not identify lenders that committed money to the proposed deal.
SK hynix would have given SoftBank a business at the center of the AI hardware buildout. High-bandwidth memory, or HBM, stacks memory chips to move large amounts of data quickly, helping powerful processors handle AI workloads efficiently.
The supplier's connection to Nvidia is documented in its own announcements. In July, SK Group said the companies were expanding their partnership, with Nvidia and SK hynix establishing a long-term relationship to secure and jointly develop next-generation AI memory, including HBM.
SK hynix already has customers booking years ahead
The memory company's October results show why its capacity would attract attention. It reported third-quarter revenue of 104.6 trillion won and operating profit of 83.3 trillion won, with a stated operating margin of 80%.
More than half of its production capacity for 2027 and 2028 was already covered by long-term agreements. It said it had signed those agreements with 14 major customers, including leading global cloud providers.
Chief executive Kwak Noh-Jung described the change in the business:
The customer trust built through our technological competitiveness is leading to long-term contracts, and our business structure is transforming into one with earnings stability and mid-to-long term demand visibility.
The company also forecast memory demand exceeding supply through 2030, citing the difficulty of advanced manufacturing and the time needed to build new facilities.
It planned to ramp its M15X facility to maximum capacity by year-end to increase HBM4 production. It expected that generation of memory to overtake HBM3E in both shipments and revenue in the fourth quarter.
SoftBank has an existing route to large bank loans
Separate from the SK hynix idea, SoftBank has disclosed substantial financing for its AI investments. In March, it signed a $40 billion bridge facility, primarily to fund follow-on investment in OpenAI and general corporate purposes.
That facility named JPMorgan Chase, Goldman Sachs, Mizuho Bank, Sumitomo Mitsui Banking Corporation and MUFG Bank as lenders. It was unsecured and carried a March 25, 2027 maturity date.
Those banks belong to the disclosed OpenAI-related financing. Their participation there does not identify them as backers of the reported SK hynix proposal.
SoftBank has also been buying infrastructure businesses. It completed its approximately $3.1 billion acquisition of DigitalBridge on September 30, adding an investment platform focused on data centers, towers, fiber and other digital infrastructure.
Memory suppliers are pushing beyond individual chips
SK hynix's latest technology update, published November 20 after Supercomputing 2026 in Chicago, shows how it is trying to expand its role in AI systems.
The company demonstrated a memory-centered rack system, pooled memory shared across servers, and technologies aimed at reducing the time processors spend waiting for data. It also displayed HBM4 and next-generation HBM4E products.
The pooling demonstrations focused on reusing stored information across AI tasks, reducing duplicate processing and shortening GPU idle time, according to the company.