Each successive facility that funds SoftBank’s OpenAI position is secured against something one step further from the asset being bought โ and that drift is a structural property of private markets, not a judgement about any party.
Company Disclosure and Press Reporting โ Chronology
Earlier โ Bloomberg reporting (not independently verified by FourWeekMBA)
SoftBank takes a $10 billion margin loan secured against OpenAI shares โ the tightest collateral coupling available.
9 September 2026 โ SoftBank press release
SoftBank announces it will prepay the $25.9 billion outstanding balance on a $40 billion bridge facility (of which $30 billion had been drawn), with prepayment on 15 September. $40 billion is the facility size, not the amount repaid.
14 September 2026 โ Bloomberg reporting (not independently verified by FourWeekMBA)
SoftBank reportedly secures an upsized $11.87 billion loan, above an earlier $10 billion target. Write-ups of that reporting describe it as a two-year facility from roughly twenty banks and as non-collateralised. Apollo, SoftBank and OpenAI have not commented.
17 September 2026 โ Bloomberg reporting (not independently verified by FourWeekMBA)
Apollo Global Management is reported to be in talks with SoftBank about boosting a loan to $9 billion from $5.4 billion, backed by Vision Fund 2 assets. Size not finalised; nothing agreed or closed. Apollo, SoftBank and OpenAI have not commented.
Bloomberg reported on 17 September 2026 that Apollo Global Management is in talks with SoftBank Group about boosting a loan to $9 billion from $5.4 billion, to help SoftBank amplify its bets on OpenAI. FourWeekMBA has not independently verified that reporting, and Apollo, SoftBank and OpenAI have not commented on it. The reported size has not been finalised and nothing has been agreed or closed. The critical structural detail in the Bloomberg report is what would secure the facility: not a pledge of the OpenAI stake, but assets inside Vision Fund 2 โ a portfolio reported to hold more than 150 positions, of which OpenAI is reported to account for roughly $2.2 billion.
That follows Bloomberg’s reporting of 14 September that SoftBank had secured an upsized $11.87 billion loan above an earlier $10 billion target. FourWeekMBA has not independently verified that reporting either, and Apollo, SoftBank and OpenAI have not commented on it. Write-ups of the Bloomberg reporting describe that facility as a two-year arrangement from roughly twenty banks and as non-collateralised โ meaning it is secured by nothing specific, only by SoftBank’s general credit standing. Separately and earlier, Bloomberg reported that SoftBank took a $10 billion margin loan secured against OpenAI shares; FourWeekMBA has not independently verified that report, and no comment has been made on it by the parties involved.
One item in this sequence is not press reporting at all. On 9 September SoftBank announced by its own press release that it would prepay the $25.9 billion outstanding balance on a $40 billion bridge facility, of which $30 billion had been drawn, with prepayment on 15 September. No terms, pricing, maturity, covenants or loan-to-value figures are reported for any of these facilities, and none appears in this analysis.
The key insight: Three distinct credit structures are funding the same underlying exposure to OpenAI, yet each one is secured against something one step further from that asset than the last. That is not an accident of deal-making โ it is a structural consequence of what private markets allow and what they do not.
Neither figure on the right is a completed transaction in the same sense: one is reported as secured, the other is reported as under discussion and explicitly not finalised.
The Structural Read
Read in sequence, these three instruments trace a clean arc of what Business Engineer analysis calls collateral drift: the progressive decoupling of the security behind a financing from the asset the financing is meant to fund.
The margin loan against OpenAI shares is the tightest coupling available. The thing borrowed against is the thing being financed; the lender’s security moves directly with the value of the position. The Vision Fund 2-backed facility, which Bloomberg reported Apollo is discussing at up to $9 billion (though that figure is not finalised and no agreement has been reached), is one step out. OpenAI is reported to represent roughly $2.2 billion of a portfolio with more than 150 positions; the other holdings are effectively doing the underwriting โ what Business Engineer frameworks call portfolio-level underwriting of a single position. The non-collateralised $11.87 billion bank facility, as described in write-ups of Bloomberg’s 14 September reporting, is a step further still: there is no specific security at all, only the borrower’s general credit standing.
Business Engineer โ Collateral Drift
Illiquidity forces capacity from elsewhere on the balance sheet
A listed holding can be trimmed at the margin โ quietly, continuously, in small amounts โ to fund a follow-on position. A large stake in a private company cannot. There is no continuous market; sales are negotiated events rather than transactions; and a partial exit communicates something the holder may have no wish to communicate. So financing capacity has to come from somewhere else on the balance sheet: other holdings, or general credit standing. The result is that each successive facility ends up secured against assets that are increasingly distant from the one being financed. This is a property of the private asset, not a characterisation of the holder.
The upsizing in both of the reported facilities is worth noting without over-reading. Bloomberg’s 14 September reporting described the bank facility closing above its $10 billion target, at $11.87 billion. The Apollo facility under discussion is reported at up to $9 billion against an existing $5.4 billion โ though, again, that size is not finalised and the discussion may not produce a transaction. Taken loosely, the pattern suggests that availability has not been the binding constraint in either case. That observation says nothing about the terms on which the money was or would be offered โ none of which is reported โ and it characterises no lender’s risk appetite, underwriting discipline or judgement.
There is also a wider structural point about how the credit layer is attaching to AI. Much of the credit deployed into AI so far has attached to physical plant: data centres, chips, power contracts โ assets with contracted cash flows and an identifiable resale market. What this sequence describes is different: credit extended against a fund portfolio and against general corporate standing, in order to fund an equity position in a private company. A private minority stake is not marked continuously by a market; it is marked by periodic primary rounds. A fund portfolio containing one is marked the same way. No claim is made here about what any specific collateral is worth, about any valuation of OpenAI, or about any loan-to-value ratio โ none of which is reported.
It is also worth being precise about the distinction between these instruments and an equity issuance, since private capital is often described as supplying what public issuance once did. An issuance raises permanent capital by selling ownership; the buyer’s claim is residual. A borrowing โ whether secured or not โ raises temporary capital that must be repaid, and the lender’s claim ranks ahead of equity. Those are different instruments carrying different obligations. Nothing here argues that either is better, safer, or evidence of anything about the health of the broader market.
Three Implications
IMPLICATION 1 โ The collateral map is the analytical object, not the headline number
When multiple financing structures fund the same exposure, the question that matters is not the aggregate size but what each instrument is secured against and what that security can bear. The Apollo facility under discussion โ reportedly up to $9 billion, not finalised, no agreement reached โ would be backed by Vision Fund 2 assets broadly, with OpenAI representing a reported fraction of that portfolio. The variables that govern the arrangement from there are collateral valuation, refinancing windows, and general credit assessment. None of those is reported, and none is predicted here.
IMPLICATION 2 โ Private-asset illiquidity is a financing constraint with a predictable shape
The drift from asset-backed to portfolio-backed to unsecured is not idiosyncratic to SoftBank or to OpenAI. It is the predictable shape of what happens when an investor wants more of one private asset than any single financing structure will comfortably supply against that asset alone. Understanding that shape โ Business Engineer’s illiquidity forces capacity from elsewhere on the balance sheet โ is more useful than treating each instrument as a standalone event.
IMPLICATION 3 โ The AI credit layer is developing a different collateral profile from infrastructure credit
Credit attached to AI infrastructure โ data centres, chips, power โ has a resale market and contracted cash flows to underwrite against. Credit attached to fund portfolios holding private equity positions does not offer the same continuous price signal. As more credit extends against fund assets and general corporate standing rather than physical plant, the marking mechanism for the underlying collateral โ periodic primary rounds rather than continuous market pricing โ becomes the relevant structural variable. No claim is made here about what any of that collateral is worth.
The collateral drift described here โ from a direct stake to a portfolio to unsecured credit โ only makes sense once you know where in the AI stack the underlying asset sits and why that position makes it structurally illiquid. The Business Engineer Map of AI traces 200+ companies across nine layers and shows how capital formation, credit, and ownership are attaching differently at each layer. The SoftBank-Apollo sequence is a case study in what happens when the equity layer does not offer the same financing handles as the infrastructure layer below it.
SoftBank is building an OpenAI position across three credit structures โ a margin loan against the shares themselves, a reported non-collateralised bank facility, and a Vision Fund 2-backed facility with Apollo reportedly under discussion at up to $9 billion (not finalised, nothing agreed or closed, FourWeekMBA has not independently verified the Bloomberg reporting, and Apollo, SoftBank and OpenAI have not commented). Read individually, each is a financing decision. Read in sequence, they trace exactly the pattern you would expect when a large private stake cannot be financed against itself alone: capacity migrates outward, from the asset to the portfolio to general credit standing, because private markets leave no other route. The structural lens here is not a forecast and predicts no outcome โ it is a description of what private-asset illiquidity looks like when an investor is determined to keep building.
Sources: Bloomberg, 17 September 2026 โ Apollo in talks with SoftBank about boosting loan to $9 billion; Bloomberg, 14 September 2026 โ SoftBank secures upsized $11.87 billion loan; SoftBank Group press release, 9 September 2026 โ prepayment of $25.9 billion outstanding balance on $40 billion bridge facility. FourWeekMBA has not independently verified the Bloomberg reports. Apollo, SoftBank and OpenAI have not commented on those reports. No terms, pricing, maturity, covenants or loan-to-value figures are reported for any of these facilities and none appears above. This is businessanalysis, not investment advice; no view is expressed on any security and no recommendation is made. SoftBank Group and Apollo Global Management are publicly listed; OpenAI is private.
The $9 billion figure is under discussion and has not been finalised. Nothing here should be read as saying that loan has been raised, agreed or closed, and the talks may not produce a transaction. The Apollo talks and the $11.87 billion facility come from press reporting attributed to people familiar with the matter rather than from company announcements, and FourWeekMBA has not independently verified either; Apollo, SoftBank and OpenAI have not commented on them here. The 9 September prepayment is different in kind: it is SoftBank’s own announcement, made by press release, and the $40 billion figure there is the size of the bridge facility rather than the amount repaid, which was the $25.9 billion outstanding balance. No terms, pricing, interest rate, covenant, maturity, loan-to-value ratio or haircut is reported, and none appears above. No valuation for OpenAI and no value for any collateral is stated. Nothing here says or implies that any party is distressed, overextended, imprudent or at risk, and nothing predicts any outcome, margin call, default or refinancing. No lender’s risk appetite, underwriting or judgement is characterised. The contrast drawn between a public issuance and a margin facility describes two different instruments carrying different obligations; it is not an argument that either is better or safer, and no conclusion is drawn about the health of any market. SoftBank Group and Apollo Global Management are publicly listed companies and OpenAI is private. No claim is made about any share price, share-price move, market capitalisation or market reaction, and nothing here suggests how any investor should read this. This is businessanalysis, not investment advice, no view is expressed on any security, and no recommendation is made.
Gennaro Cuofano is a business analyst with an International MBA in corporate finance and business strategy (2012). He leads research at FourWeekMBA, which he created, and at The Business Engineer (businessengineer.ai), his newsletter on AI and business strategy, with over 90,000 subscribers and more than 660 published analyses. His work breaks down how companies make money, read from their own filings: revenue and profit structure, capital allocation and ownership. FourWeekMBA reached about four million business people, including C-level executives, investors, analysts and product managers, in 2022 alone. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
