TOKYO — Masayoshi Son’s current bet is being framed as far larger and more concentrated than his earlier Vision Fund investments, whose exposure to companies such as WeWork and Uber contributed to losses exceeding $70 billion.
After years of rebuilding, Son is now pursuing a strategy that makes those earlier wagers appear comparatively cautious. The line captures both the scale of his renewed ambition and the risk of staking so much on a single idea.
SoftBank Group launched a record bond offering on Monday, seeking $10 billion in dollar-denominated notes and €1 billion in euro-denominated securities to fund the third and final tranche of its $30 billion follow-on investment in OpenAI. The deal, structured across five tranches from 3.5 to 7.5 years in maturity, is expected to price on September 24 and settle September 29. If completed at its intended size, it would be the largest non-financial corporate bond ever issued from Asia-Pacific and Japan, surpassing the $10.93 billion raised by 7-Eleven in January 2021, Nikkei Asia reported.
The October 1 closing would bring SoftBank’s cumulative investment in OpenAI to $64.6 billion, approximately 13% ownership. No institutional investor anywhere in the world holds a stake in a single company at comparable scale. The closest parallel in AI infrastructure is Microsoft, which committed $190 billion in capital expenditure for 2026, financed through operating cash flows rather than high-yield debt secured against a single private equity position.
Fitch Ratings assigned the proposed notes a BB+ designation, the highest speculative-grade rating and one notch below investment grade. SoftBank has sold nearly $15 billion in bonds across currencies already in 2026, the largest junk-rated corporate borrower in public bond markets this year. Its credit default swaps hovered near a three-year high as the offering launched, Seeking Alpha reported, signaling that fixed-income investors believe the leverage has moved into uncomfortable territory.
The new bonds cancel a $10 billion bridge loan that SoftBank secured for this tranche while permanent financing was arranged. But the full debt structure is far larger: stacked above the bonds sits an $11.87 billion two-year syndicated loan from approximately 20 banks, finalized last week, and a $10 billion margin loan against the OpenAI position itself, with Apollo Global Management in talks to raise its lending commitment to $9 billion. Unlike the equity-based financing deployed by other AI investors, Alibaba raised $10.2 billion in Hong Kong’s largest share sale for its AI expansion; SoftBank is financing a private equity stake entirely on borrowed capital.
Son took the first decisive step toward this position in November 2025, when he sold SoftBank’s entire $5.8 billion Nvidia stake, reportedly telling reporters he cried over the decision, to free capital for OpenAI. Nvidia has roughly doubled in value since that sale. Whether Son made the wrong trade or the right one at the wrong price is a question the OpenAI exit will eventually answer.
The broader AI funding wave is moving at comparable velocity but through different mechanisms. Mistral AI closed a €3 billion Samsung-led round this month, the largest equity raise in European technology history, dilutive to shareholders but carrying no default risk. SoftBank’s debt-on-a-private-equity-stake model has no direct precedent at this scale in technology investing.
The thesis requires OpenAI to reach the public markets at a valuation that covers the leverage. Latest reports put OpenAI’s internal fundraising discussions at up to $1.2 trillion. At 13% ownership, SoftBank’s stake would be worth roughly $156 billion against a $64.6 billion cost, a return sufficient to service every instrument stacked to fund the position. The timing is the difficulty: the OpenAI IPO is not expected before 2027, and the shortest bonds in this deal mature in 3.5 years. Anthropic, targeting a November IPO at a $2 trillion valuation, will reach the public markets ahead of OpenAI, and its pricing will set a benchmark that either validates or complicates Son’s math before his own exit window opens.
What happens September 24 will carry its own reading. If investors absorb the full $11 billion, the bond market is endorsing Son’s arithmetic. If the book covers light, the credit default swap signal was the honest one. Son has been wrong at scale before, and the Vision Fund stands as a cautionary case study. What is harder to argue this time is that the outcome is modest either way.
