SoftBank is seeking a $10 billion margin loan secured by its OpenAI shares at roughly 7.88% interest, sitting on top of a $40 billion bridge loan signed in March 2026, bringing the Japanese conglomerate's total financing against its OpenAI position to around $50 billion and making it the single most leveraged institutional bet in the history of AI investment.
The $10 billion margin loan is a two-year facility with a one-year extension option, priced at approximately 425 basis points above the secured overnight financing rate. The loan is backed by SoftBank's OpenAI equity, not by the conglomerate's broader balance sheet. That structure is meaningful because it creates a direct link between OpenAI's private market valuation and SoftBank's borrowing capacity. If OpenAI's valuation declined significantly, the collateral value would fall, creating margin-call risk. At the current $852 billion post-money valuation, SoftBank's roughly 13% stake is notionally worth around $110 billion, which gives the $10 billion loan a comfortable collateral cushion on paper.
The $40 billion bridge loan signed in March was a different structure entirely. That deal was non-collateralised, underwritten by JPMorgan Chase, Goldman Sachs, Mizuho Bank, Sumitomo Mitsui Banking Corporation, and MUFG Bank, with a 12-month tenor designed to bridge SoftBank's $30 billion follow-on investment in OpenAI while the company sold assets to refinance. The margin loan is an additional layer on top, designed to provide longer-term liquidity without forcing a share sale. Together, the two facilities give SoftBank access to roughly $50 billion in financing tied to a position that cost it approximately $64.6 billion in cumulative investment.
The debt load has attracted rating agency scrutiny that founders and investors should take seriously as a signal. S&P has downgraded SoftBank's credit outlook to negative at BB+, and the company faces an estimated $32 billion funding gap over the next two years. That gap reflects how the financing is structured. The bridge loan matures in 12 months, meaning SoftBank needs to replace $40 billion in short-term borrowing by March 2027 through asset sales, long-term bond issuance, or another refinancing event. If the asset sales are slower than expected or market conditions shift, the refinancing question becomes more acute. The margin loan helps by providing longer-dated liquidity, but it also adds to the debt stack rather than reducing it.
For SF readers, the structure of SoftBank's OpenAI financing shows how frontier AI is increasingly funded through capital markets engineering rather than straightforward venture rounds. OpenAI's $122 billion funding round, the largest private financing in history, closed in March 2026 with $50 billion from Amazon, $30 billion each from Nvidia and SoftBank, and further tranches from other investors. At that scale, no single investor writes a cheque from operating cash flow. They use leverage, bridge facilities, and asset collateral. That is a normal practice in private equity and infrastructure investing. It is less usual in AI, and it imports risks that the venture world has not traditionally had to manage.
The second-order risk is what happens if OpenAI's private market valuation is tested. The $852 billion figure is a post-money valuation set in a round where the same large investors that set the terms also participated. There is no public market for the shares, no daily price discovery, and no short sellers to push back. If revenue growth slows, the competitive environment shifts, or broader AI sentiment reverses, the notional valuation could fall sharply without any single observable trigger. A 30% decline in notional OpenAI valuation would still leave SoftBank's position well above the $10 billion margin loan, but it would tighten the buffer and increase pressure on the $40 billion bridge refinancing. That is the kind of second-order risk that does not appear in the headline AI funding numbers but runs through every leveraged position in the stack.
The broader question for the AI funding boom is whether SoftBank's retreat on the margin loan target, reportedly cut from $10 billion to $6 billion under lender pressure, suggests banks are starting to apply more discipline to AI-linked leverage. If it does, that is actually healthy. The alternative, where lenders extend unlimited credit against private company valuations set by the same investors who benefit from high marks, is the dynamic that produced some of the worst outcomes in the last technology bubble. Lenders asking for smaller positions and higher collateral coverage rates are the market doing its job. The real test will come when the $40 billion bridge needs to be refinanced.
Also read: Nvidia's $2.1 billion bet on IREN shows miners with power are becoming AI infrastructure • The Nvidia chip smuggling network shows Washington's AI containment strategy has a serious leakage problem • TSMC's record quarter is proof that AI infrastructure spending is real, not just projected
