Credit desks rarely get excited about a Japanese holding company. The five-year contracts that insure SoftBank Group’s debt against default trade in a narrow band most weeks, and the people who watch them are the ones paid to read footnotes. On Nov 6 the spread widened to about 466 basis points at one point, the highest in four years, according to Reuters. The equity did something louder: the stock closed down 11% in Tokyo at 3,549 yen.
The two numbers describe the same event from opposite ends. A 466 basis point spread means a buyer of protection on $10 million of SoftBank debt pays roughly $466,000 a year for the insurance, a price that implies a meaningful probability of restructuring inside five years or a meaningful fear of one. An 11% one-day decline means holders of the equity decided the company’s assets were worth more than a tenth less than they had thought the day before. Since early September the market capitalization has nearly halved.
SoftBank is a holding company, and that structure explains why the two markets moved together. Its assets are stakes in other companies, the largest of which is its controlling position in Arm, and its liabilities are borrowings raised against those stakes. The equity is the difference. When the assets are marked down the difference narrows, the debt becomes a larger share of the whole, and lenders who accepted a given level of borrowing on the way up want more collateral or a higher rate on the way down.
The mechanism runs both ways, which is what makes it dangerous. A wider spread raises the cost of refinancing the debt that comes due, and SoftBank refinances constantly because its borrowings are large relative to the cash its holdings distribute. A higher cost of refinancing reduces the value of the equity, which narrows the collateral cushion further, which widens the spread again. Nothing in that loop requires a single asset to fail. It requires only that the assets stop appreciating.
Reuters attributed the selloff to concerns about AI profitability and reported that the selling moved from stocks into Asian credit markets. That distinction matters. A stock price can fall for a year while a company keeps paying its bills. A spread that reaches 466 basis points is a statement by people who are paid to be right about default, and the statement is that the assets behind this balance sheet have become harder to value with confidence.
The AI connection is not incidental to that valuation. SoftBank’s recent strategy has been to take proceeds from its listed holdings and put them into AI infrastructure and model developers, ventures that produce little cash today and whose worth rests on assumptions about demand late in the decade. That is the kind of asset a credit committee discounts hardest, because the collateral is a claim on future cash flow rather than a building or a dividend. Equity investors can hold a view on the assumption and wait. Bondholders have to price the case where the assumption is wrong.
The counterargument is that this company has run the loop before and come out of it. In 2020 and again in 2022 the shares fell by more than half, spreads widened, and management responded by selling assets, pledging more collateral and buying back stock. The playbook works as long as there are assets left to sell and buyers for them. It works less well when the largest holding is already the thing the market is trying to price, because selling it to support the credit removes the support.
There is a technical reading that is less alarming. SoftBank’s bonds are not numerous, and a spread that appears in a thin market can travel further than the underlying risk justifies. Part of the 466 basis points may be the price of finding a seller rather than the price of default, and Japanese credit has a history of overshooting in both directions around quarter-end. The equity decline is harder to explain away, since it happened on heavy volume in one of the world’s most liquid markets, but the two instruments are not measuring the same thing.
What the pairing does establish is that the company’s cost of capital has repriced. SoftBank’s model depends on borrowing against appreciated assets and deploying the proceeds into the next asset that will appreciate, and the first step of that model just became more expensive. If the holdings resume climbing, the spread narrows and the arithmetic works. If they stall, the mathematics of a holding company with this much debt does the work by itself, and the equity, not the credit, absorbs the difference.
The next data points are dull ones. The quarterly disclosure of the loan-to-value ratio will show how much cushion remains. The pricing on the next bond will show what the market charges for new money rather than for old paper. And management’s choice between an asset sale and a buyback will show which side of the balance sheet it is defending. Anyone looking for a verdict on artificial intelligence from the credit market will not find one in a single day of Tokyo trading. What they will find is a lending market that has started charging SoftBank a rate for the risk it carries, and a company whose next move has to be read as an answer to that price.
