SoftBank's Bridge Loan Deadline and the Classification That Lets It Borrow

SoftBank is borrowing more money than most countries do. And the money is coming from places that reveal exactly what the debt markets think of the borrower.

The most concrete number: SoftBank will repay the remaining $25.9 billion of a $40 billion bridge loan on September 15. That loan was arranged in March to fund investments in OpenAI — a private company that may not go public for another year. SoftBank can't just walk away from a six-month bridge loan because it made a multi-year bet. The bridge was supposed to be a bridge. Now it's a deadline.

To replace it, SoftBank is lining up at least three financing lines at once. A potential $10 billion to $20 billion junk bond sale, with investor meetings in New York this week. A record ¥1 trillion ($6.3 billion) retail bond just priced at 4.75% interest — the highest coupon the company has paid in 17 years, back to the Lehman Shock era. And a $10 billion margin loan secured by its OpenAI stake, with talks underway for a second $10 billion margin loan against the same shares.

That was a lot of debt. The basic point is that each of these tranches exists because someone else won't — or can't — lend at the terms SoftBank needs, and the classification of SoftBank's credit is where the plumbing splits.

The rating boundary that makes this possible

SoftBank has two credit ratings, and they disagree sharply. S&P Global rates the company BB+ — one notch below investment grade, which is the formal term for "junk." Japan Credit Rating Agency (JCR) rates it A — solidly investment grade.

That five-notch gap isn't an academic dispute. It determines who is legally allowed to lend.

Japanese pension funds, insurance companies, and many institutional mandates have a hard rule: buy investment-grade only. Under JCR's A rating, SoftBank is investment grade. Under S&P's BB+, it isn't. And Japanese institutions that follow international ratings are barred from touching it.

S&P's rating is the one that reflects stress conditions. S&P treats SoftBank as an investment holding company, where the company's ability to pay depends on the market value of its portfolio. When stock prices drop, the borrower's cushion shrinks. JCR's A rating emphasizes SoftBank's track record of paying back 5 trillion yen in retail bonds over 20 years — which is a real historical fact and not exactly evidence that stress conditions won't break the pattern. JCR's outlook on that A rating, though, is negative. The agency knows the margin is thin.

So the institutional capital that would normally fund a company this large passes through the S&P gate and stops. The remaining capital that can touch this paper includes Japanese retail investors — people in their 60s with surplus retirement savings who have no institutional mandate blocking them. They are, effectively, the lender of last resort for a company that international institutions classify as speculative grade.

The 4.75% coupon on the retail bond tells you something about how the market prices that risk. That's more than 2 percentage points above a Japanese government bond of similar maturity. It's higher than the subordinated debt of major Japanese banks, which average 2.6% to 2.9%. The market is asking a lot of these bondholders for their patience.

Borrowing against the collateral

The margin loans add a second layer of structure on top of the bond refinancing. A margin loan is a loan secured by securities. You pledge the shares; if the shares drop, you post more cash or repay. It's the same plumbing as a margin call at your brokerage, just at a $10 billion scale.

SoftBank closed a $10 billion margin loan in August, secured by its OpenAI preferred shares. Three weeks later, it was seeking another $10 billion against the same shares. The lenders are the usual syndicate — Goldman Sachs, JPMorgan, Mizuho, Apollo, Sumitomo Mitsui.

The collateral is valued at $852 billion for OpenAI, set in a private funding round in March. That's the number the margin loan relies on. It's also a number that hasn't been tested by a public market — OpenAI submitted confidential IPO paperwork in June, but the listing may not come until 2027, and could be delayed further if tech volatility discourages investors. SoftBank's stock dropped 12% on one day in June just on the rumor of an IPO delay.

Here's the structural consequence that matters. SoftBank is spending money to buy preferred shares in a private company. Then it's borrowing against those same shares. The shares can only be liquidated if someone buys them. If OpenAI's valuation drops materially, the lenders can demand cash — and SoftBank would need to find that cash while its primary asset is becoming less valuable. The circular reference is obvious, and it's the standard margin loan risk at a concentration most companies never approach.

Lenders required a corporate guarantee from SoftBank on the margin loan — meaning if the OpenAI shares aren't enough, SoftBank itself is on the hook. This was a concession. SoftBank originally wanted the loan backed only by the shares, which would have left lenders with no recourse if the collateral lost value. The banks said no.

What the debt actually does to the balance sheet

SoftBank's gross debt was about $80 billion as of March 2025. With the pending bond sales, margin loans, and refinancing, it could exceed $130 billion. Against shareholder equity of roughly $139 billion (¥21.4 trillion), the debt-to-equity ratio sits at about 125% — already stretched before the new borrowing comes in.

The company's own financial discipline threshold is a Loan-to-Value (LTV) ratio below 25%. As of June 2026, that ratio was 13% — cushioned heavily by Arm Holdings' stock price surge. But a 50% drop in the value of SoftBank's portfolio shares would push LTV to about 26%, breaching the threshold. The NAV (Net Asset Value) already swung from ¥72.3 trillion to ¥58.3 trillion in just over a month in the summer, a roughly 19% drop.

The LTV is the gatekeeper number. It's the one metric that connects the quality of SoftBank's assets to its ability to borrow. When the assets are Arm stock, the LTV is transparent — Arm trades daily on a public market. When the asset is a 13% stake in a private company, the LTV is based on a last-round valuation that may be months old and never stress-tested. That's not a criticism of the valuation methodology so much as a description of what kind of risk the LTV is hiding.

SoftBank's latest quarterly results were strong — EPS of $0.19, versus an expected $0.07, and revenue of $12.82 billion. The stock surged 18.8% on September 8. The market cap sits around $235 billion. But those earnings are driven by mark-to-market gains on existing holdings (including Intel), not by OpenAI, which hasn't been marked up yet and carries no revenue contribution. The profit the company reports and the profit the debt service assumes are not exactly the same thing.

What the investor actually holds

This isn't a story about whether AI is worth investing in. SoftBank has already decided that. The story is about what kind of financial arrangement SoftBank has built around that decision, and what classification lets it work.

The machine looks like this: SoftBank uses short-term bridge loans to buy long-term, illiquid equity stakes. It then refinances the bridge loans with longer-term junk bonds, which institutional investors can't hold due to rating constraints, so the bonds go to Japanese retail investors who rate them as "A." In parallel, it borrows against the equity stakes themselves, layering margin debt on top of bond debt on top of the original investment. The whole structure requires OpenAI to go public, trade at or above its private valuation, and generate enough mark-to-market gains to service the borrowing without triggering margin calls or breaching LTV thresholds.

Each piece makes sense on its own. Together they create a concentrated bet where the downside is mechanical — valuation drops, LTV breaches, margin calls fire, and the liquidity chain tightens fast — while the upside requires a smooth sequence of events that SoftBank doesn't control. The IPO timing is OpenAI's decision. The valuation at IPO is the market's decision. The creditworthiness of SoftBank during the transition is S&P's decision.

The classification boundary is where the risk relocates. International institutional money, which could have demanded more favorable terms or more protective covenants, is excluded by the BB+ rating. The capital that fills the gap — Japanese retail savings — gets a 4.75% coupon and an A rating from the domestic agency, but the same underlying stress risk that S&P flagged. The bonds are unsecured. If SoftBank can't pay, there's no claim on the OpenAI shares — those are already pledged to the margin lenders.

For an investor watching from the sidelines, the useful takeaway is structural rather than directional. SoftBank is not running out of money. It's moving money from one bucket to another, from short-term to long-term, from institutional to retail, and from unencumbered to pledged. The question isn't whether the refinancing happens — the machinery is already turning. The question is whether the asset that secures the entire operation justifies the leverage layered on top of it. That's not a question about AI. It's a question about what $852 billion means when it's a private number, and what happens when the market gets to price it.