TOKYO, 27 AUG 2026 — SoftBank is in talks with banks to sell between US$10bn and US$20bn of bonds, in dollars and euros, to partly repay the US$40bn bridge loan it took to fund its investment in OpenAI. A launch could come as early as September.
At the top of that range it would be the largest such issue by any Asian company this year. It is also the fourth financing instrument stacked on one investment.
Counting the layers
The investment itself is a commitment of close to US$65bn in OpenAI by October. Around it sit four separate pieces of financing.
It began with a US$40bn bridge loan taken earlier this year. The proposed bond sale of US$10bn to US$20bn would repay part of that bridge, while a record ¥1 trillion retail issue in Japan, roughly US$6.3bn, is being prepared alongside it. Underneath all of that sits a US$10bn margin loan already secured against the OpenAI stake.
While each instrument is ordinary, together they describe a position funded almost entirely with borrowed money, with each tranche refinanced as it comes due.
The margin loan is the piece to watch
A bridge loan and a bond issue are unremarkable corporate finance. Borrowing against the asset you have just bought is a different arrangement, and it deserves attention because of what the collateral is.
OpenAI shares are not publicly traded. There is no closing price, no daily mark, no liquid market a lender can sell into if things go wrong. The collateral is valued by reference to the last private round, which is a negotiated figure agreed among a small group of investors.
Loans against listed shares carry maintenance requirements: if the price falls past a threshold, the borrower posts more collateral or the lender sells. Applying that logic to an unlisted holding requires agreeing in advance what a decline even looks like, and the terms of this facility have not been published.
The exposure this creates is worth setting out. If the private mark on OpenAI were to fall, SoftBank would face a collateral call on an asset it cannot readily sell, at precisely the moment the rest of its borrowing looked least comfortable.
Selling AI risk to Japanese savers
The ¥1 trillion retail bond is the least discussed part and arguably the most consequential domestically.
Retail bonds in Japan are bought in large numbers by individual savers, often older, in an economy where deposit rates have been negligible for a generation. A SoftBank retail issue is a familiar product to that audience and has generally paid them well.
What is being funded has changed, though. Money raised at this scale now supports an artificial intelligence position, which carries different risks from a telecommunications operator's cash flows. The bondholder is a creditor of SoftBank rather than an investor in OpenAI, so the exposure is indirect. It is still exposure.
Why borrow rather than sell something
SoftBank holds substantial listed assets, and the obvious alternative to four layers of debt is to sell some of them. The choice not to is deliberate and revealing.
Selling crystallises a price and gives up the future. Borrowing against a portfolio keeps the upside while converting it into spending power now, and for an investor who believes his holdings are worth more later than today, that is the rational instrument. It is also how the same investor has funded previous phases, and the group's history with concentrated, leveraged positions is well documented in both directions — enormous returns from the early Alibaba stake, and heavy losses in the first Vision Fund.
What debt removes is the ability to be wrong slowly. An equity investor who misjudges the timing waits. A borrower has coupon dates, maturities and, in the case of a margin facility, a collateral test that does not care about conviction.
The structure is therefore a statement of confidence in the investment's future value rather than a reflection of limited capital. SoftBank could raise money by selling. It is choosing the option that pays more if the thesis holds and constrains it considerably if the timing slips.
This is the same pattern, one layer up
The financing structures around artificial intelligence keep rhyming, and the rhyme is that operating cash flow is not paying for any of it.
Nvidia has guaranteed up to US$105bn of OpenAI's Ohio data centre lease through residual-value support rather than cash. Compute-backed lending has arrived at scale. Global data centre capital expenditure is on course to pass a trillion dollars.
Each instrument is individually defensible and they share a dependency. All of them are underwritten by an expectation about future revenue from a technology whose largest customers are, so far, mostly each other. This is not a prediction of failure but an observation. The same assumption about future revenue is now load-bearing in several connected places at once.
Why bonds now, and in two currencies
Replacing a bridge loan with bonds is the intended life cycle of a bridge — the name says so. Bridges are expensive, short and arranged quickly; bonds are cheaper, longer and take time to market.
Issuing in both dollars and euros widens the buyer base, which matters at this size. A single-currency deal of US$20bn would strain the appetite of any one investor pool, and splitting it lets the issuer reach American and European institutions separately.
The timing carries its own signal. Coming to market in September, at the top of an enthusiasm cycle for artificial intelligence, is a decision to raise while conditions are favourable rather than when the bridge finally matures. This is prudent treasury management. The price achieved will also be read as a verdict from bond investors, who are paid to be pessimistic.
What it means from here
For readers in this region, the interesting figure is not the size but the price. A US$10bn to US$20bn issue from a major Asian borrower, tied to an AI investment, will produce a yield — the first independent market judgement on this class of exposure.
Equity investors have been enthusiastic about artificial intelligence for two years. Credit investors have not had to price it directly at this scale. What they demand in September will say more about perceived risk than any valuation headline, because a bondholder makes nothing if the thesis works and loses everything if it does not.
The second consequence is regional and practical. A deal of this size absorbs a substantial share of the appetite for Asian corporate credit, and other issuers planning autumn transactions will find the window more crowded and the pricing less generous than they expected.
