SoftBank markets a jumbo junk bond to fund its OpenAI stake

SoftBank is borrowing over $11 billion in risky bonds to fund an OpenAI stake². IFR described the deal as an AI-related jumbo high-yield offering⁴. Barron’s reported that SoftBank is seeking a record amount from high-yield investors, and that part of the cash raised will be used to fund a follow-on investment⁵.

Demand has outrun supply. Bloomberg reported that the deal drew over $20 billion of early interest¹. BigGo Finance put the same figures together, reporting that an offering of more than $11 billion had attracted over $20 billion in preliminary orders ahead of pricing, with the longest tranche running 7.5 years⁶.

The Decoder added the balance-sheet detail. SoftBank wants to swap existing borrowings for longer-term debt, a move that could push the company past its own debt ceiling for a while².

Not every AI financing is clearing

The Financial Times reported that SoftBank’s $50 billion data centre group has slowed its initial public offering⁷. The company, also backed by Masayoshi Son, faces market fears over an AI slowdown and lacklustre interest in new debt⁷.

Those two facts sit awkwardly together. A jumbo high-yield deal is oversubscribed at the parent level while a listing at the subsidiary level stalls on the same thesis^6,7. The difference is what each instrument asks investors to underwrite: a coupon from a diversified borrower, or the equity value of data centre capacity.

Why it matters to investors

This is the AI build-out arriving in the credit market rather than the equity market. When the marginal dollar of AI capital spending is raised as sub-investment-grade debt, the funding cost is set by high-yield spreads, and the exposure lands in bond funds and credit indices rather than in technology portfolios^4,5.

The leverage detail matters more than the headline size. A borrower moving toward, and possibly past, its own stated debt ceiling has less room to absorb a slower ramp in the asset it is funding².

Watch where the deal prices against the early interest, because a book that large and a coupon that high do not usually coexist for long^1,6. Watch the length of the curve investors will accept, which the 7.5-year tranche is testing⁶. And watch whether the data centre listing resumes, since that is the cleaner read on whether public equity still wants this risk⁷.