NextFin News - SoftBank Group turned to the high-yield bond market to fund its artificial intelligence bets after the world's $23 trillion syndicated loan market began to push back. The Japanese conglomerate issued $11.1 billion of dollar- and euro-denominated notes on Sept. 24, 2026 — the largest high-yield corporate bond sale on record — at coupons as high as 9.75 percent, pricing a market willing to finance Masayoshi Son's AI ambitions, but only at a steep premium.
The deal capped a month in which SoftBank raised roughly $28 billion across three funding channels: a ¥1 trillion retail bond sale, an upsized $11.87 billion syndicated loan, and now this jumbo junk-bond offering. Together they cover the final $10 billion tranche of the $30 billion follow-on commitment to OpenAI. When that payment lands on Oct. 1, SoftBank's cumulative stake in the ChatGPT maker reaches $64.6 billion — one of the largest single-company wagers in technology history, struck at a $730 billion pre-money valuation for roughly a 13 percent ownership interest.
The question the market is now asking is not whether SoftBank can raise money. It can. The question is whether the cost of that money — and the structure of the balance sheet that carries it — is sustainable if the AI boom pauses before the payoff arrives.
The Funding Sprint: Three Channels, One Bet
SoftBank's September financing push reads like a company exhausting every available source of capital in sequence. It began with a record ¥1 trillion ($6.32 billion) of yen-denominated retail bonds. It then secured an $11.87 billion two-year syndicated loan from about 20 banks, upsized from an initial $10 billion target. And it closed the month with the $11.1 billion bond sale that became the headline event.
The bond structure itself tells the story of a borrower paying up for duration. The dollar notes split into $1 billion at 3.5 years priced at 8.625 percent, $4.5 billion at 5.5 years at 9.25 percent, and $4.5 billion at 7.5 years at 9.75 percent. Two euro tranches of €500 million each, maturing in four and six years, yielded 7.125 percent and 8 percent. The 9.75 percent coupon on the 7.5-year paper is a rate more typical of deeply stressed borrowers than of a company with SoftBank's profile — a direct read of the risk investors are underwriting.
The transaction is the largest high-yield corporate bond issuance globally on record, surpassing the $10.9 billion sold by Numericable Group in 2014, according to LSEG data. In 2026 alone, SoftBank has issued $14.6 billion of high-yield bonds — 63.4 percent of the entire Asia Pacific and Japan high-yield corporate bond market. One borrower now accounts for nearly two-thirds of a regional junk-bond market. That concentration is the clearest single indicator of how much of the AI buildout is being financed on credit.
The proceeds are earmarked for a specific, dated obligation: the third and final $10 billion tranche of the $30 billion follow-on investment in OpenAI announced in February, due Oct. 1. The first two $10 billion tranches were paid in April and July. Once complete, SoftBank's total OpenAI investment reaches $64.6 billion through SoftBank Vision Fund 2, giving it an ownership stake of approximately 13 percent at the $730 billion pre-money valuation set in the follow-on agreement.
Why the $23 Trillion Loan Market Said No
The headline figure — a $23 trillion market — refers to the outstanding stock of global syndicated loans, which reached $23.1 trillion at the end of 2025, according to the OECD's Global Debt Report 2026. That market, larger than the $36.4 trillion corporate bond market, has long been the preferred funding channel for large, relationship-driven borrowers. SoftBank itself leaned on it with the $11.87 billion facility.
But syndicated loans carry constraints that bonds do not. They are typically shorter-dated, bank-held, covenant-heavy, and sized by relationship capacity rather than investor appetite. A two-year loan must be refinanced or repaid quickly — a poor match for an equity investment whose exit depends on an IPO that OpenAI has now said will not happen in 2026. Bonds, by contrast, lock in long-duration funding and distribute the risk across a dispersed investor base that can absorb it without a single lender re-underwriting the exposure.
The shift from loans to junk bonds is the structural signal in this story. Satoru Aoyama, senior director at Fitch Ratings, put it plainly: "Hyperscalers in the US have been raising debt, but they have high credit ratings. Now AI-driven debt issuance has reached the high-yield market at scale." The distinction matters. Amazon, Google, Meta and Microsoft can borrow at investment-grade spreads because their operating cash flows cover their AI capital expenditure. SoftBank cannot. It is funding illiquid private equity stakes with public debt that must be serviced regardless of whether those stakes produce a return.
Fitch assigned SoftBank Group its first-time issuer rating of BB+ with a stable outlook on Sept. 11, 2026 — the highest speculative-grade tier, but speculative grade nonetheless. S&P rates the company BB+. For comparison, Alphabet and Amazon, the two largest corporate bond issuers of 2026, carry AA+ and AA ratings. That four-to-five-notch gap is the reason SoftBank pays 9.75 percent while the hyperscalers borrow at a fraction of that cost.
"I was positively surprised by the market appetite," said Satoru Aoyama of Fitch Ratings.
The appetite was real — the deal drew more than $20 billion of preliminary, non-binding demand before it priced — but the price was not cheap, and the cost of insuring SoftBank's debt has been climbing. Five-year credit default swaps hovered near 383 basis points in late September, close to the highest level since 2023, according to data provider CMA; one measure pushed above 400 basis points during the week, up from around 280 basis points in June. SoftBank's own 2031 dollar bond yield rose to about 8.2 percent this month from as low as 6.7 percent in January. The market is repricing the company even as it funds it.
The Broader Backdrop: A $109 Trillion Market Taking On AI Risk
SoftBank's move is the leading edge of a wider migration of AI financing into the credit markets. The OECD's Global Debt Report 2026 put combined sovereign and corporate bond markets at approximately $109 trillion, with outstanding corporate debt at $59.5 trillion split between $36.4 trillion of bonds and $23.1 trillion of syndicated loans. Into that vast pool, AI-related borrowers are now drawing down capital at a pace without precedent. Goldman Sachs credit strategists estimate global AI-related debt issuance has topped $575 billion in 2026. Hyperscalers issued $108 billion of debt globally in 2025; so far in 2026 they have issued $194 billion. Roughly one-third of AI capital expenditure this year is expected to be debt-financed.
The second-order implication is the migration down the capital structure itself: as AI capital expenditure outgrows free cash flow, companies move from investment-grade bonds to high-yield bonds to asset-backed loans to equity-linked structures. The transmission mechanism is straightforward, and each step down raises the cost of capital and the fragility of the buildout. SoftBank is simply the furthest point along that curve — the place where AI financing has reached the deepest, most expensive layer of the credit market.
The concentration risk is already visible in the data. In 2025, nine major technology firms — the so-called hyperscalers — raised a combined $122 billion from bond markets, representing nearly half of all global bond issuance by technology companies, according to the OECD. If half of their projected AI investments were financed through bonds, these firms alone would account for roughly 15 percent of historical annual global bond issuance by non-financial corporations. A single theme is becoming a large share of the market's supply, and SoftBank's 63.4 percent regional junk-bond share is the concentrated version of the same phenomenon.
The market has already begun to discriminate within that theme. Shares of debt-heavy neocloud CoreWeave rose almost 8 percent in a recent week, while Oracle — which has leaned heavily on the debt market for its AI expansion — fell 7 percent for the week and about 30 percent for the year. Investors are not rejecting AI financing wholesale; they are pricing the quality of the balance sheet behind it. SoftBank sits on the expensive side of that divide.
The Counter-Thesis: Son's $46 Trillion World
Bulls argue the entire concern rests on too narrow a view of the payoff. Masayoshi Son has laid out a vision in which AI-related industries account for 20 percent of global output by 2040 — equivalent to $46 trillion — and global AI infrastructure spending reaches $5 trillion a year. At the SoftBank World conference in July, he dismissed bubble talk as "absurd," argued that data centers could require 3 terawatts of power by 2040, about 1.8 times current global electricity consumption, and predicted 100 trillion autonomous agents by the end of the decade.
On that view, a $64.6 billion stake in the leading frontier model developer is not overpayment — it is early positioning in a market that will dwarf today's valuations. If AI revenue does reach 20 percent of global GDP, SoftBank's equity stake could be worth multiples of its cost, and the 9.75 percent coupon becomes a rounding error. The funding structure is a bridge to a payoff large enough to make the bridge's cost irrelevant.
There is also a credit backstop that bears on the bear case. SoftBank has not relied solely on unsecured bonds. It has taken out loans collateralized by its holdings in chip designer Arm and by the OpenAI stake itself, and it has sold assets along the way. S&P's Makiko Yoshimura noted in a research note that Arm's strong share-price performance has supported SoftBank's credit quality to the point where the OpenAI IPO delay does not have an immediate negative impact on creditworthiness. Fitch said in a note that it expects SoftBank's debt to rise as it funds committed investments, but the company should retain adequate liquidity and access to capital markets. Fitch's stable outlook reflects the same judgment: the collateral base buys time.
But the counter-thesis has a timing problem at its core. The debt is fixed in cost and near in maturity; the payoff is uncertain and distant. The 7.5-year notes at 9.75 percent must be serviced through 2034. OpenAI's IPO — the clearest liquidity event for SoftBank's stake — has been pushed out, with Sam Altman confirming it will not happen in 2026. Every quarter of delay adds interest expense without adding exit optionality. A structural bull case does not erase a cyclical cash-flow squeeze.
What to Watch
The base case is that SoftBank services the debt and waits for OpenAI's listing. The company has enough collateral and enough funding channels to avoid a near-term liquidity event, and the stable ratings outlook from Fitch supports that read. The Oct. 1 tranche payment will close, the notes will trade, and the story will shift to the next funding need.
The downside case turns on two variables: the cost of refinancing and the timing of the exit. If five-year credit default swaps push materially above 400 basis points and stay there, or if SoftBank's 2031 notes trade above 9 percent for an extended period, the market is signaling that the next raise will be even more expensive. If OpenAI's IPO slips into 2027 or beyond, the mismatch between debt service and liquidity widens.
The refinancing wall gives the downside case its teeth. The two-year syndicated loan comes due in 2028, the retail bonds mature in 2033, and the 7.5-year dollar notes run to 2034. Each of those dates is a potential repricing event if the AI narrative has cooled or if OpenAI has not yet listed. The bull case requires the IPO to arrive before the credit markets force a more expensive refinance — a race between a private-market listing calendar and a public-debt maturity schedule.
The falsifying signal for the bear thesis is specific: if OpenAI prices its IPO in 2026 at a valuation above $730 billion — the pre-money at which SoftBank's follow-on was struck — the entire funding critique collapses, because the equity gain would dwarf the interest cost. Conversely, the falsifying signal for the bull thesis is a five-year CDS level sustained above 450 basis points alongside an IPO delay past mid-2027; that combination would indicate the credit market no longer believes the payoff will arrive in time.
Short term, the relief rally in SoftBank shares — up more than 7 percent after the deal priced, following an 11 to 13 percent drop on Sept. 14 after the IPO delay — reflects funding secured, not risk resolved. Medium term, the watch items are the refinancing wall and the CDS trajectory. Long term, the thesis stands or falls on whether AI revenue actually reaches 20 percent of global GDP by 2040, as Son forecasts, or whether the buildout slows under the weight of its own financing cost.
The $23 trillion loan market did not stop SoftBank. It simply made the company pay a high-yield price to walk around it — and that price is the market's first honest assessment of what the AI boom costs when the cash-rich buyers step aside.
Explore more exclusive insights at nextfin.ai.
