SoftBank Group is back in the bond market with more than $11 billion of junk-rated debt, and the price tag is rising fast. The company is marketing $10 billion of dollar notes and €1 billion, or about $1.1 billion, of euro bonds as it pushes deeper into artificial intelligence, including a follow-on investment in OpenAI tied to commitments nearing $65 billion. Investors have already signaled that the AI build-out is no longer just about scale. It is also about how much balance-sheet pain the market is willing to absorb.
The deal comes with record-high yield guidance for SoftBank’s dollar tranches, underscoring how costly the AI race has become for one of its most aggressive backers. The company is offering $1 billion of 3.5-year notes at 8.75% to 8.875%, $4.5 billion of 5.5-year notes at 9.375% to 9.5%, and another $4.5 billion of 7.5-year notes at 9.75% to 9.875%. Those terms are on the high end for a borrower rated BB+ by S&P, the highest speculative grade, and they reinforce a broader market theme: capital for AI is still available, but it is getting more expensive.
SoftBank’s move lands as credit costs are climbing, not easing. Its 2031 dollar bond yield rose to 8.2% this month from a 6.7% January low, while the cost of default insurance hit a three-year high. That backdrop helps explain why investors are demanding hefty compensation for the new notes. The offering is also being watched as a signal for how public debt markets are pricing AI ambition when the proceeds are effectively tied to private-market exposure rather than immediate operating cash flow.
That tension is at the heart of the trade. SoftBank is not issuing debt to fund a conventional expansion with quick payback. The financing is partly aimed at supporting its OpenAI follow-on investment, a bet on a company still in buildout mode. In that sense, the bond buyers are lending against a story about future AI value, not a near-term earnings engine. The market is making that distinction clear in the yields it is asking for.
The deal also says something bigger about the state of AI financing. Global AI-related debt issuance has topped $575 billion in 2026, according to Goldman Sachs credit strategists. That number shows how the AI arms race has moved well beyond venture capital and equity funding. Data centers, chips, networking gear and power systems all require huge sums, and borrowers across the stack are turning to debt markets to keep pace. SoftBank’s jumbo issue is simply the most visible reminder that the funding chain now reaches deep into junk territory.
For SoftBank, the scale matters as much as the cost. Bloomberg, via The Star Malaysia, said the company has sold nearly $15 billion of notes across currencies in 2026, making it the largest junk-rated borrower this year. That is a striking position for a company that already carries a speculative-grade label. It shows both the appetite SoftBank has for funding large bets and the willingness of the market to keep funding it, even as the borrowing cost climbs.
Still, the market is not handing out cheap money. The spread of likely pricing, especially on the 7.5-year tranche, points to how investors are demanding more for duration and credit risk at the same time. Bloomberg, via The Edge Singapore, said the 7.5-year notes were being marketed at 9.75% to 9.875%, a level described as a record high for SoftBank dollar bonds of that maturity. That kind of pricing says the market is treating the AI thesis as real, but not risk-free.
SoftBank’s OpenAI commitment is the other half of this story. Bloomberg, via The Star Malaysia, said the follow-on investment commitments are nearing $65 billion. That scale makes the bond sale more than a routine refinancing. It is a funding bridge for a strategic stake in one of the hottest names in AI. And it helps explain why the market is insisting on a bigger coupon. The bonds are not financing a utility-like asset with stable cash generation. They are supporting a private-company position whose long-term value is still being tested in public markets and private valuations alike.
The market reaction is also shaped by what the debt says about SoftBank’s own balance sheet strategy. As BNP Paribas strategist Mana Nakazora put it: “This is a financing decision that tells you how Son values the OpenAI stake relative to the cost of his balance sheet. The market will price the notes on the assumption that the proceeds go out the door as equity in a private company, not into anything that generates near-term cash flow.” That framing captures the central trade-off: the more SoftBank leans into AI, the more creditors will want to be paid for taking the ride.
For now, demand appears to be there. Bloomberg, via Seoul Economic Daily, said bookbuilding drew more than $20 billion in orders for the $11 billion offering. That is a strong result, even for a marquee borrower. It suggests investors are still willing to fund AI-linked stories when the compensation is rich enough. But robust orders do not mean cheap financing. If anything, they show that the market can be enthusiastic and demanding at the same time.
SoftBank’s debt sale is likely to be priced on Sept. 24, 2026, with settlement set for Sept. 29, 2026, according to Reuters/CNBC TV18 via NAI500. The third tranche of SoftBank’s OpenAI follow-on investment is expected to close on Oct. 1, 2026. Those dates leave little time for the market to cool off. If anything, they set up a near-term test of whether investors are comfortable underwriting the next phase of the AI boom at near-double-digit yields.
The bigger message is simple: AI is no longer just bid up in stocks and private valuations. It is moving into credit, where the cost of capital is more visible and the risk premium is harder to ignore. SoftBank’s deal makes that shift plain. The AI race is still open for financing. It is just becoming a much more expensive race to run.
