NextFin News - Apollo Global Management is in talks to enlarge a loan to SoftBank Group to $9 billion, up from $5.4 billion, deepening the private-credit giant's role as the hidden financier of the artificial-intelligence boom. The proposed increase, backed by assets in SoftBank's Vision Fund 2, has not been finalized and the discussions remain private, with representatives for both firms declining to comment.
The talks matter because they expose the plumbing behind the AI investment frenzy: SoftBank is not paying for its OpenAI stake with cash on hand, but by borrowing against a portfolio of private assets, and Apollo is the lender willing to stretch the facility to a new record. The arrangement lets SoftBank keep writing checks into OpenAI while pushing the credit risk onto a private-credit balance sheet that marks its collateral by model rather than by market.
That is the central tension of the deal. The loan signals a structural shift in how frontier AI gets funded - private credit replacing public markets as the lender of first resort - but the specific exposure is a cyclical credit bet whose safety depends on an initial public offering that OpenAI has already signaled may not happen in 2026.
The Deal: A Record NAV Loan Gets Bigger
Apollo is discussing lifting the size of its net-asset-value loan to SoftBank's Vision Fund 2 to $9 billion from $5.4 billion. NAV loans are secured by the reported value of a fund's underlying portfolio rather than by hard collateral, and they have become essential liquidity tools for private capital funds and their backers. The $5.4 billion facility, set in August 2025 with a $900 million top-up, was already a record for the asset class and represented a low-teens percentage of the value of the portfolio backing it - a pool of more than 150 assets.
For Apollo, the enlarged facility is a fee-generating use of the balance sheet at a moment when the firm is selling investors on the reliability of its credit engine. Apollo reported $1.026 trillion in assets under management at the end of the first quarter of 2026, crossing the $1 trillion threshold for the first time, while fee-related earnings reached a record $728 million, up 30% from a year earlier.
"Our first quarter results set a strong tone for the year, with record fee-related earnings and assets under management surpassing $1 trillion," Apollo Chief Executive Marc Rowan said.
Yet the same quarter showed the fragility beneath the fee story: on a GAAP basis Apollo swung to a net loss of $1.93 billion, or $3.27 a share, against net income of $418 million a year earlier, driven by a one-time $1.7 billion tax expense tied to its Bermuda-based insurance unit and a $2.1 billion unrealized loss on investments in that unit. Shares of Apollo are down 17.3% year to date, versus a 6.9% gain for the S&P 500, and trade well below their 52-week high of $153.29. The SoftBank loan is precisely the kind of large, relationship-heavy credit that private-credit managers point to when arguing that public markets have ceded ground to private lenders - but it also concentrates counterparty risk in a single borrower whose fortunes hinge on one startup.
How SoftBank Is Paying for OpenAI Without Cash
The $9 billion loan does not fund OpenAI directly. Its function is balance-sheet relief: by borrowing against Vision Fund 2 assets, SoftBank frees up cash and credit capacity to keep funding its OpenAI commitment. That commitment is enormous. In February 2026, SoftBank agreed to invest $30 billion more in OpenAI in three $10 billion tranches, taking its cumulative investment to $64.6 billion and an ownership interest of approximately 13%. The round priced OpenAI at $730 billion pre-money, about $840 billion after the money - the most valuable private technology company in history.
SoftBank has been methodical about funding it. It closed the first tranche of $10 billion on April 1, 2026, and the second tranche of $10 billion on July 1, both drawn under a $40 billion bridge facility the company signed on March 27. That bridge loan - unsecured, maturing in 12 months, and underwritten by JPMorgan Chase, Goldman Sachs, Mizuho Bank, Sumitomo Mitsui Banking Corp. and MUFG Bank - was the largest borrowing the company had ever taken denominated solely in dollars. The company said it would be repaid partly through asset sales.
Since then, SoftBank has layered on more debt. In late July, its $40 billion bridge attracted 21 new lenders in a syndication phase, with around $7 billion allocated - First Abu Dhabi Bank, GIC and Standard Chartered each taking nearly $1 billion. On August 6, SoftBank arranged a $10 billion margin loan backed directly by its OpenAI stake, with Goldman Sachs Bank USA, JPMorgan Chase Bank NA, Mizuho Securities USA LLC, Apollo Global Funding LLC and Sumitomo Mitsui Banking Corp. as arrangers. That facility carries a protection lenders always want but rarely admit is necessary: if the value of OpenAI's preferred shares falls substantially, SoftBank must provide cash or repay early. On August 28, the company was seeking another $10 billion two-year loan at a margin of around 275 basis points over SOFR, with Mizuho Bank as mandated lead arranger. On September 14, it secured an upsized $11.87 billion loan from around 20 banks. And in September it moved to retail investors, filing to issue 1 trillion yen, about $6.3 billion, of seven-year bonds with an indicative coupon of 4.3% to 4.9%.
"We are entering a new phase where frontier AI moves from research into daily use at global scale," OpenAI said in its February announcement.
The market has treated that phase as an invitation to leverage. The question is who ultimately carries the risk when the phase turns.
The Collateral Question: What Backs a $9 Billion Loan?
The safety of Apollo's loan rests on two propositions, and both deserve scrutiny. First, the collateral is not OpenAI itself but a diversified pool of more than 150 Vision Fund 2 assets, with the loan sized at a low-teens percentage of that pool's value. Second, SoftBank says its financial policies are unchanged: it manages its group loan-to-value ratio below 25% under normal market conditions, with an upper threshold of 35% in emergencies, and holds enough cash to cover bond redemptions for at least two years.
Those guardrails are real, but they are calibrated to a world in which the collateral keeps its marks. SoftBank's net worth is increasingly a function of a single private valuation. TD Cowen analyst Krish Sankar estimated in May that SoftBank's roughly 11% stake in OpenAI was worth about $80 billion at the end of March, up from $54.4 billion at the end of December - a gain of $25.6 billion in three months on paper. Reverse the trade and the same mechanism works backward: a de-rating in OpenAI's private valuation would compress the value of the Vision Fund 2 portfolio that backs Apollo's loan, even though OpenAI is not the direct collateral.
The IPO delay makes this concrete. In a magazine interview, OpenAI Chief Executive Sam Altman said the company would not seek a public listing in 2026. SoftBank shares fell more than 11% to 5,795 yen on the news, and the Nikkei 225 slid 2%. The third tranche of SoftBank's OpenAI investment - $10 billion due on October 1, 2026 - can be accelerated in the event of a public listing. No listing means no acceleration, which means SoftBank still needs the cash on schedule. More debt is the obvious answer. More debt against a portfolio whose largest marks are private and illiquid is the answer that keeps credit managers awake.
Research outlet CreditSights has estimated that SoftBank faces a funding shortfall of about $32 billion, including bond maturities over the next two years and other agreed deals such as a $5.4 billion acquisition of ABB's industrial-robotics unit. The company has already sold its entire Nvidia stake to help fund the OpenAI buildout - liquidating the asset that arguably gave it the most certain upside in the very boom it is trying to finance. That is the asymmetry investors should note: SoftBank sold the stock that went up to buy the stake that now has to keep going up.
Second-Order Effect: Private Credit Is Underwriting AI Without Owning It
The first-order read of the Apollo talks is simple: Apollo earns fees, SoftBank gets liquidity. The second-order effect is more consequential and less discussed. Private-credit lenders are becoming the indirect underwriters of AI capital expenditure without owning any AI equity. Apollo does not hold OpenAI shares. Its claim is on a diversified fund portfolio. But the marginal dollar of credit it extends increases the system's aggregate exposure to the one variable that ties the whole structure together - the continued re-rating of frontier AI.
Apollo's own 2026 credit outlook acknowledges the scale of the buildout it is financing: hyperscaler capital expenditure has tripled since 2023, and sell-side forecasts point to more than $2.7 trillion of cumulative AI-related spending from 2025 to 2029. Public bond markets cannot absorb that alone at acceptable yields, so private credit fills the gap - with less transparent pricing, less liquidity, and covenants that are negotiated rather than disclosed. The Apollo-SoftBank loan is the purest example: the collateral is marked by the fund's own valuations, the terms are private, and the ultimate exit depends on a public market that has already shown it will punish the borrower for bad timing.
This is a structural shift, not a cyclical blip. The regime change is durable: once private credit proves it can intermediate trillion-dollar technology buildouts, the flow will not revert to public syndication on its own. But the specific Apollo-SoftBank exposure is cyclical. It is a levered bet on the timing of an IPO and the path of a private valuation - both mean-reverting variables. Getting this distinction wrong flips the conclusion. The structure of AI finance has changed permanently; the profitability of this particular loan depends on a window that may already be closing.
The Counter-Thesis: The Risk Is Contained, and Apollo Knows It
The strongest argument against alarm is straightforward and comes from the numbers themselves. Apollo's loan is not a direct loan to OpenAI. It is secured by a diversified pool of 150-plus assets at a low-teens loan-to-value ratio, and SoftBank's group-level LTV policy caps leverage at 25% under normal conditions. Apollo is one of the world's largest alternative asset managers, and it priced this relationship over multiple facilities dating to 2025. If the risk were truly asymmetric, the margin would not be the only thing growing - the collateral coverage would be shrinking, and there is no evidence of that.
There is force to this view. A 25% LTV cap with a 35% emergency ceiling and two years of redemption coverage is not a reckless balance sheet. And Apollo's exposure is to SoftBank's creditworthiness, which still includes Arm Holdings, a listed semiconductor design company that has ridden the AI chip wave, and a portfolio of listed and private technology assets. The loan is large, but it is not unhedged by structure.
The rebuttal is that structure only works while marks hold. NAV loans are notoriously difficult to enforce in stress because the collateral cannot be sold quickly without realizing the very losses the margin call is meant to prevent. And SoftBank's diversification is thinner than the 150-asset count suggests: a handful of large technology positions dominate Vision Fund 2's marks, and OpenAI is the swing factor for all of them. If OpenAI's valuation resets, the correlation across the pool rises toward one, and the low-teens LTV becomes a backward-looking statistic. The falsifying signal is specific: if SoftBank's group LTV breaches 25%, or if the October 1, 2026 third tranche is delayed beyond the calendar year while OpenAI's private valuation is marked down, the "contained risk" thesis is wrong. Watch the next SoftBank financial report for the LTV disclosure and any revision to the OpenAI tranche timetable.
What Comes Next: Three Horizons
Short term (sentiment and liquidity): The deal, if completed, reinforces the narrative that Apollo's credit platform can place very large, bespoke facilities - a positive for fee income and for the firm's standing with relationship borrowers. Apollo's shares, down 17.3% year to date, could find support from evidence that its private-credit engine remains in demand even as public-market peers struggle. SoftBank avoids a near-term funding squeeze ahead of the October tranche.
Medium term (fundamentals and refinancing): The $40 billion bridge loan matures roughly 12 months after March 27, 2026 - around March 2027. That is the first real test. If OpenAI has not listed by then, SoftBank must refinance into whatever credit conditions prevail, potentially at wider margins than the 275 basis points over SOFR it was seeking in August. A delay in the IPO combined with tighter credit spreads would force asset sales into a weaker market - the exact dynamic the margin-loan covenants were designed to prevent.
Long term (structural): The durable change is the normalization of private credit as the financier of AI infrastructure. Apollo, Blackstone and their peers will keep intermediating capital into the buildout because public markets price the risk differently and disclose more. The beneficiaries are the asset managers with scale, balance sheets and borrower relationships; the exposed are the lenders who underwrite single-borrower concentration on the strength of private marks, and the retail investors buying the 4.3%-4.9% yen bonds without a transparent view of the collateral chain.
Base case: the $9 billion facility closes, the October tranche funds on schedule, and OpenAI lists in 2027 at a valuation that keeps the marks intact. Upside case: an earlier-than-expected listing accelerates the third tranche, SoftBank deleverages into strength, and Apollo books a large fee on a loan that never stresses. Downside case: the IPO slips past 2027, OpenAI's private mark resets lower, SoftBank's LTV approaches its ceiling, and Apollo is asked to extend or restructure a facility whose collateral is worth less than the model says.
Data as of September 17, 2026.
Apollo's $9 billion loan is not a bet on OpenAI - it is a bet that everyone else will keep believing in OpenAI long enough for SoftBank to cash out. That is a profitable position until it is the only position left.
Explore more exclusive insights at nextfin.ai.
