Apollo's Loan to SoftBank Doesn't Back OpenAI — and That's the Point

Reports this week say Apollo Global Management is mulling raising its loan to SoftBank Group, with numbers around $9 billion being discussed. That would be a significant step up from the $5.4 billion Apollo has already lent — a record-size facility for this type of deal. It sounds like one more gear in SoftBank's never-ending borrowing machine to fund its OpenAI bet.

But the loan in question isn't actually what it sounds like. It doesn't lend directly against OpenAI. It doesn't even touch OpenAI's books. The loan is collateralized by something much weirder: the estimated value of more than 150 assets held inside a fund called Vision Fund 2. You can't buy these companies on a market. You can't sell them on short notice. Nobody knows exactly what they're worth today because there is no today price for most of them.

And yet this is the collateral base.

This is what's called a NAV loan — net asset value financing. The name describes the mechanism: you borrow against the manager's own estimate of what a portfolio of private companies is worth. The loan isn't secured by cash flow, earnings, or even a publicly traded share price. It's secured by a spreadsheet.

The basic point is that NAV lending has become one of the quietest plumbing changes in finance over the past few years. A facility that was once a last-resort liquidity bridge for private equity funds is now standard-issue capital — a way for fund managers to borrow cheaply against paper wealth that exists on a mark-up basis. The market for NAV loans is estimated at approximately $150 billion and growing. Apollo is one of the biggest lenders in it, using capital from its insurance subsidiary Athene, which holds vast reserves that need to earn a return.

So when you hear "Apollo is raising a loan to SoftBank for OpenAI bets," the actual contract is more like: Apollo lends money against SoftBank's entire private portfolio, and OpenAI happens to be the reason that portfolio looks bigger right now.

Here's what that means, layer by layer.

The NAV loan is the low-leverage cushion

Apollo's NAV loan to Vision Fund 2 covers only a low-teens percentage of the fund's portfolio value. That's finance speak for roughly 10% to 14% loan-to-value. Vision Fund 2 holds more than 150 assets. Even if OpenAI is the dominant name on the list, the collateral base is diversified. Apollo's loan is rated investment grade and placed within Apollo's insurance funds — which matters because it lets Apollo deploy insurance capital that faces regulatory return requirements. The proceeds from the loan go to the fund's investors, specifically SoftBank and CEO Masayoshi Son, providing liquidity without forcing a sale of underlying assets.

A potential increase to $9 billion would still leave the facility at a modest percentage of portfolio value. The NAV loan doesn't move the risk needle dramatically up or down. It's a steady, low-cost line of credit against a bucket of assets that Apollo is comfortable with.

The real leverage is the margin loan

If the NAV loan is the cushion, the margin loan is the knife. In August 2026, SoftBank closed a $10 billion margin loan backed by its OpenAI stake — later upsized to $11.87 billion. The facility attracted commitments from around 20 banks. The interest rate is SOFR + 425 basis points, roughly 7.88%. That spread is steep, and it tells you something about how the lenders view the collateral: you don't charge that much unless the underlying asset is illiquid, hard to value, and subject to sharp re-pricing.

Unlike the NAV loan, the margin loan is a classic pledge arrangement. OpenAI shares are the collateral. If OpenAI's valuation drops below a certain threshold, the lender can demand more collateral or call the loan. The loan has a two-year term with one-year extension, which is the exact window where everyone hopes OpenAI goes public. If it doesn't, the put/call dynamic gets ugly fast.

The margin loan was not a smooth process. SoftBank faced lender hesitation and was forced to downsize the target by 40% to $6 billion before working back up. It eventually came back and closed the full $10 billion — upsized to $11.87 billion — by September. That push-pull tells you the banks wanted the yield but needed convincing about the collateral quality.

The bridge loan is the elephant

Both the NAV loan and the margin loan sit on top of a $40 billion bridge loan underwritten by JPMorgan Chase, Goldman Sachs, Mizuho, Sumitomo Mitsui, and MUFG. That bridge was meant to fund the $30 billion follow-on OpenAI investment (on top of the roughly $40 billion already committed). Bridge loans are expensive short-term debt designed to be repaid from a longer-term financing. The question for SoftBank is what the "longer-term" looks like.

The company is considering a $10 billion to $20 billion bond sale to help refinance, and it has been raising nearly $90 billion in total debt this year across bonds and loans. Total SoftBank debt has reached roughly $135 billion. The $40 billion bridge alone represents a significant fraction of that stack.

Who actually holds the risk?

This is where the plumbing produces a useful answer. The risk isn't shared evenly across these instruments — it's partitioned by structure.

Apollo's NAV loan is the most insulated. It's a small percentage of a diversified portfolio, rated investment grade, and funded by insurance capital that expects steady yield rather than explosive returns. Apollo earns fees and interest on a relatively stable loan. Even if OpenAI's valuation were to collapse, Vision Fund 2's other 150-plus holdings would still provide collateral value. Apollo's worst case is a re-marking of the portfolio, which would trigger standard collateral adjustments.

The ~20 banks on the margin loan hold the concentrated exposure. Their collateral is specifically the OpenAI stake. If OpenAI gets revalued downward — whether from an IPO pricing below $852 billion, a failed funding round, or a broader AI valuation correction — the margin loan lenders feel it first and most directly. The 425 basis point spread over SOFR is their premium for that concentrated, illiquid risk.

And SoftBank holds all the residual exposure. The company borrowed to buy an illiquid private company, then borrowed again using that same private company as collateral, all while maintaining a $40 billion bridge that needs to be refinanced. The total OpenAI commitment is roughly $65 billion for a roughly 13% stake. At the $852 billion post-money valuation, that stake is notionally worth around $110 billion — which makes the leverage look manageable. But the notionally worth number is the exact problem: it's a private valuation, set in a funding round, not a market price.

The valuation cliff

Every layer of this debt structure depends on OpenAI staying above some implied valuation threshold. The margin loan has explicit collateral calls. The NAV loan has portfolio re-marking. The bridge loan repayment assumes SoftBank can access capital markets or generate liquidity from an OpenAI exit.

A public listing would resolve the collateral question. OpenAI's $852 billion private valuation becomes a public market price, the margin loan lenders can monitor the stock daily, and SoftBank could potentially sell into strength to retire debt. That's obviously the happy-ending assumption built into the structure.

The less convenient scenario is a public offering that prices below expectations, a prolonged private status, or a broader repricing of private AI valuations. In that case, the margin loan is the first thing to get squeezed, and the NAV loan gets re-marked. Apollo's cushion holds, but the whole structure tightens.

So the Apollo headline is worth paying attention to, but not for the reason it suggests. The $9 billion number doesn't represent Apollo making a wild new bet on OpenAI. It represents Apollo continuing to run a business model it's built over years — lending insurance capital against private fund portfolios at investment-grade rates — on a client whose portfolio happens to contain the most expensive private company in the world.

The real risk isn't Apollo's. The real risk is SoftBank's: a company that has turned its OpenAI stake into the central collateral asset of a massive debt stack, with the margin loan holding the most concentrated exposure and a bridge loan that needs refinancing by its March 2027 maturity. The plumbing works as long as the valuation holds. That's the entire trade.