SoftBank's $50 Billion Borrowing Machine for OpenAI

SoftBank is about to sit down with U.S. institutional bond investors and pitch something that may be one of the largest junk-bond sales ever from an Asian company: $10 billion to $20 billion of high-yield debt.

That was a lot of borrowing. But the interesting part is not how much. The interesting part is what the borrowing is for, and what the borrowing is for is — to replace the borrowing SoftBank already took on for the same purpose.

It is a refinancing loop, and it is entirely deliberate.

The machine

SoftBank's total investment in OpenAI is projected to reach nearly $65 billion by October, for roughly a 13% stake. OpenAI's last private valuation was $852 billion, set in a February 2026 fundraising round that brought in $110 billion from SoftBank, NVIDIA, and Amazon.

Of SoftBank's $65 billion commitment, only about $35 billion was actual equity — money coming from existing assets, asset sales, and earlier fund capital. The rest was borrowed. Specifically:

• A $40 billion bridge loan from a syndicate of major banks (JPMorgan, Goldman Sachs, Mizuho, Sumitomo Mitsui, MUFG and 21 others), due March 2027. Unsecured means the lenders have no collateral claim on anything. They are lending on the promise that SoftBank will raise the money to pay them back.
• Two $10 billion margin loans, each collateralized by SoftBank's preferred shares in OpenAI. The first closed in early August, the second was announced in late August. These are structured like the margin loan you get at a broker when you want to buy more stock than you can afford in cash — except this is $10 billion, against a private company with no daily market price. The contracts include margin call provisions: if the implied value of the stake drops, SoftBank must post more cash or accept early repayment.

So the total OpenAI-linked borrowing now exceeds $50 billion. And the bridge loan — the $40 billion chunk that has no collateral — is due in six months.

The potential $10–20 billion bond sale is meant to refinance that bridge loan. Replace the short-term, bank-lender money with longer-term, bond-investor money. It is not new capital for new investments. It is paying off one loan with another loan.

That is how a bridge loan works. The bridge is supposed to be temporary. You cross it and build the permanent road behind you. The question everyone is now asking is whether SoftBank is building that road fast enough.

Who is buying, and why does it matter?

The bonds SoftBank is considering are junk-rated — specifically, rated BB+ by S&P Global, which is the highest tier of speculative grade, one notch below investment-grade territory. Barclays strategists flagged that the offering, at the upper end of its range, could be one of the largest high-yield corporate bond sales on record.

SoftBank is also running this under Rule 144A, a Securities and Exchange Commission safe harbor that lets companies sell bonds to U.S. institutional investors without going through the full public registration process. It is the standard mechanism for private placements of junk bonds, and it has not been SoftBank's usual play — the company has not used a 144A format in more than a decade.

This is a shift. SoftBank's earlier bond fundraising this year leaned heavily on Japanese retail investors. In April, June, and early September, it issued three retail bond offerings totaling roughly ¥1.68 trillion (about $10.8 billion), with coupons in the 4.3% to 4.9% range for 7-year maturities. Japanese grandparents buying corporate bonds at a friendly rate. That pool has a ceiling. To raise another $10–20 billion, SoftBank needs to go where the deep capital is: U.S. institutional fixed-income investors who buy junk bonds every day and are used to pricing risk in basis points.

The cost will be higher. Barclays analysts expect a "meaningful pick-up" — a higher yield — over SoftBank's existing bonds. The April 2026 bond sale had coupons running from 7.6% to 8.5% on the dollar tranches. The new bonds will likely be in that range or above, depending on where junk spreads are when they price.

The self-imposed guardrail

Here is the clause that turns this from a large-but-plausible borrowing decision into something worth paying attention to.

SoftBank has a self-imposed financial rule: keep its loan-to-value ratio — net debt divided by the market value of its holdings — below 25% under normal conditions, with a hard cap of 35% in extreme markets. This is not a regulatory requirement. It is a discipline line that Masayoshi Son's team set for themselves, partly to reassure investors after the Vision Fund 1 disaster in the late 2010s, when the fund lost about two-thirds of its capital.

In March, CFO Yoshimitsu Goto told the Financial Times that the company may "temporarily" exceed the 25% threshold to fund the AI investments. The LTV had already climbed from 16.5% to 20.6% by the end of December 2025. Goto projected it could hit 26% or 27% without intervention.

The company plans to get it back down through asset sales, listings, and asset-backed financing. SoftBank cited net assets of more than ¥30 trillion (roughly $195 billion) as "balance sheet ammunition." And the OpenAI stake itself — valued on paper at $852 billion × 13% = roughly $111 billion — is a massive asset that would compress the ratio dramatically if it ever traded publicly at that mark.

But here is the plumbing detail that makes the ratio feel thinner than it sounds: the LTV denominator includes the paper value of an illiquid, privately held stake in a company that has never been tested in a public market. The $852 billion valuation was set in a private round, with no arm's-length buyer setting the price through an auction. It was reaffirmed by an employee tender offer in August at the same price, which is something, but it is not the same as a market. The valuation is an input to the ratio, not an output of it.

Where the risk actually sits

S&P Global kept SoftBank's rating at BB+ in July but lifted the outlook from negative to stable, citing "improved key financial ratios". The upgrade in outlook is real — the company's fiscal year ended March 2026 produced a staggering ¥5 trillion ($31.7 billion) in net profit, driven by roughly $44 billion in unrealized gains on the OpenAI stake alone. The Vision Fund swung from a ¥115 billion loss to a ¥6.4 trillion profit. SoftBank also dethroned Toyota as Japan's largest company by market cap, topping roughly $375 billion in June.

But those accounting profits are paper gains. The OpenAI stake is classified as a fair-value-through-profit-or-loss asset, meaning its quarterly valuation swings flow directly into reported earnings. The money that made SoftBank's fiscal year look spectacular is the same money that would make it look terrible if the valuation reset downward.

The real risks cluster in two places.

First, the refinancing deadline. The $40 billion bridge loan comes due March 25, 2027. The investor meetings start this week. If the bond offering works, SoftBank replaces a time-bomb with a laddered maturity structure. If it does not — if the offering is too large for the current junk market, or if the yield demanded is too steep — SoftBank is back to negotiating with its bank syndicate. That is not a disaster scenario, but it is a constraint.

Second, the margin call mechanic. The two $10 billion margin loans are collateralized by OpenAI preferred shares. The contracts require SoftBank to post additional cash if the implied stake value drops. OpenAI is private, so the "mark" is infrequent and subjective. But if a future valuation comes in flat or down — whether from an actual financing round or from pressure on the secondary market — the lenders can demand cash. And the cash would come from somewhere: more borrowing, more asset sales, or drawing down the pile SoftBank has been building.

Both risks share the same underlying variable: OpenAI's valuation trajectory and liquidity timeline. An IPO would resolve almost everything. It would provide a daily market price, eliminate the margin-call ambiguity, compress the LTV ratio, and give SoftBank a liquid asset it can sell or use as collateral in normal financial markets. Without an IPO, the machine keeps running on private valuations and refinancing promises.

What this means for you

If you are watching SoftBank stock (ticker 9984 on the Tokyo exchange, SFTBY on U.S. OTC markets), you are watching a company that has turned itself into a levered bet on a single private company. That is not inherently a bad structure — it is just a specific one. The question is whether you believe in the bet enough to hold through the refinancing cycle, and whether the stock price already reflects the leverage you are implicitly taking on.

SoftBank's shares rallied hard in May and June on AI momentum and the OpenAI valuation repricing. The stock has been volatile. The market treats SoftBank as a proxy for OpenAI's worth — when OpenAI's implied valuation goes up, SoftBank's stock goes up, and the LTV ratio improves. When it goes down, the ratio deteriorates and the leverage looks worse. The stock is essentially a leveraged option on OpenAI, dressed up as a telecom-and-investment conglomerate.

The bond sale itself does not change the underlying bet. It changes the financing plumbing — extending the maturity, diversifying the creditor base, and locking in a fixed interest cost for the bondholders. That is what companies do when they refinance a bridge loan. It is routine, even at $20 billion, as long as the market absorbs it.

The thing to pay attention to is not the bond offering. It is what happens if the offering goes smoothly and then nothing else changes. The bridge gets replaced, the margin loans stay in place, the LTV ratio hovers near the guardrail, and the whole machine continues running on the assumption that OpenAI eventually goes public at a valuation that supports the $50 billion in secured and unsecured debt sitting on top of a 13% stake.

That is a reasonable assumption. It is also an assumption. And the bonds, the margin calls, and the LTV ratio are all just the instruments that enforce the timeline of that assumption becoming reality.