SoftBank is selling $11 billion of junk bonds to fund its OpenAI stake
Ten billion dollars across three tenors, a billion euros across two, pricing September 24. The proceeds go toward a stake that reaches roughly 13 percent of OpenAI.
The lab behind a large share of production AI features is being funded with high-yield debt.
SoftBank Group launched a high-yield bond sale of more than $11 billion on September 20, 2026, with the proceeds going in part toward its next payment for a stake in OpenAI. Quartz reported the deal alongside Bloomberg, which first put the figure at over $11 billion in junk bonds.
The structure: $10 billion in dollar-denominated notes across three tenors of 3.5, 5.5 and 7.5 years, plus 1 billion euros across 4 and 6 year maturities. Pricing is set for September 24, with settlement on September 29. SoftBank's cumulative investment in OpenAI is expected to reach about $64.6 billion once the third tranche completes, for an ownership share of roughly 13 percent.
If it closes at that size it would be the largest non-financial corporate bond deal ever from Asia Pacific and Japan, passing the $10.93 billion 7-Eleven raised in January 2021, according to LSEG data cited in the coverage. The Decoder, citing the Financial Times, adds that the raise also refinances roughly $40 billion of short-term loans into longer-dated debt, and notes that most other large technology companies have been issuing safer bonds.
Junk is a rating, not an opinion
High-yield means the issuer pays more because the rating agencies rank the credit below investment grade. The word describes a tier. It is a common, legal and frequently sensible way to raise money, and calling a bond junk predicts nothing about whether it gets repaid. The relevant fact is not the label but the pairing: a highly leveraged holding company borrowing at high-yield rates to buy more of a business that is itself consuming cash at a rate the FT has put in the hundreds of billions through 2030.
What is genuinely unknown is the coupon, since pricing had not happened when this was written, and whether the book fills at the full size. Both will be public within days of publication, and both are more informative than any amount of commentary written before them.
Why a build studio cares
Nothing in this story changes an API response today, which is exactly why it is worth writing down. Vendor concentration is the risk clients accept without ever being asked: a product gets built against one provider's models, one provider's pricing page, and one provider's rate limits, and the switching cost compounds quietly for a year. We ask about it in audits under vanish risk, the same heading as domain control and hosting access, because the question is identical. If this vendor's terms changed in a quarter, how long would it take you to be somewhere else? For most of the codebases we open, the honest answer is that nobody has ever tried, and a second provider wired behind a router is a week of work now and a scramble later.
Next step: read Quartz on the bond sale, then go check whether your application can call a second model provider without a deploy. If you want that tested rather than assumed, write to us at hello@gattyworks.com.