Everything About AI
NextFin News - SoftBank Group's shares climbed about 7% this week even as the cost of insuring its debt against default jumped to the highest level in more than a year, a striking split that captures the central tension of the artificial-intelligence investment boom: equity investors are pricing the upside of AI while bondholders are charging a steep premium for the risk of funding it.
The divergence came into focus on October 1, when SoftBank closed the third and final $10 billion tranche of its $30 billion follow-on investment in OpenAI, bringing its cumulative commitment to the ChatGPT maker to $64.6 billion and leaving it with an approximate 13% ownership stake. To fund the payment, the Japanese conglomerate raised $11.1 billion through dollar- and euro-denominated bonds that settled on September 29 — the largest high-yield corporate bond sale on record, surpassing French telecom Numericable Group's $10.9 billion issuance in 2014 — at yields reaching 9.75%, levels more typical of substantially weaker borrowers.
Shares of SoftBank ended Thursday up 0.6% in Tokyo, after markets were closed from Monday to Wednesday for national holidays. The market is sending two messages at once. One says Masayoshi Son's concentrated bet on AI could reshape the global economy. The other says the bill for that bet is coming due, and it is expensive.
The Deal: Record Debt to Finance a Concentrated Bet
SoftBank's financing need was concrete, not theoretical. The company had arranged a $40 billion bridge loan facility in March, drawing $30 billion of it for the OpenAI commitment, with the facility maturing in March 2027. The bond sale converts part of that short-term obligation into debt that can remain outstanding for up to 7.5 years, and it also replaces a remaining $10 billion bridge facility tied to the October tranche.
The final terms put a hard number on the market's skepticism. According to the company's filing, SoftBank issued $1 billion of 3.5-year dollar notes at 8.625%, $4.5 billion of 5.5-year notes at 9.25%, and $4.5 billion of 7.5-year notes at 9.75%, alongside two €500 million euro tranches yielding 7.125% and 8%. S&P Global Ratings and Fitch rate the notes BB+, the highest rung of speculative grade but still below investment grade. By comparison, the average yield on BB-rated dollar bonds stood at 6.55%, according to index data cited in reporting on the transaction. SoftBank's shortest tranche started more than two percentage points above that benchmark; its longest approached the average associated with far weaker credits. For context, a $7.3 billion SoftBank issuance in June 2021 of senior dollar- and euro-denominated bonds yielded between 2.125% and 5.25%. The 2026 pricing is roughly four to five percentage points higher on comparable tenors.
And this was only the latest tranche of borrowing. Earlier in September, SoftBank sold 1 trillion yen ($6.3 billion) of seven-year retail bonds at a 4.75% coupon — the highest level for its straight bonds in 17 years, and more than double the 2.3% average coupon on yen-denominated retail corporate bonds issued in Japan this year. Taken together, SoftBank has issued $14.6 billion in high-yield bonds so far in 2026, accounting for 63.4% of the entire Asia Pacific and Japan high-yield corporate bond market.
The credit-risk gauge tells the same story from a different angle. The cost to insure SoftBank's debt against default over five years rose above 400 basis points in September, compared with around 280 basis points in June — its highest level since April 2025, when the OpenAI investment was first disclosed. Ratings as of September 11 show Fitch and S&P at BB+, while Japan Credit Rating Agency assigns A.
Why Investors Are Willing to Lend
Despite the pricing, demand was strong. Orders during bookbuilding reportedly exceeded $20 billion for the $11.1 billion offering. "I was positively surprised by the market appetite," said Satoru Aoyama, senior director at Fitch Ratings. That appetite reflects a simple calculation: at nearly 10%, SoftBank's bonds pay income that few other large corporate credits can match, and the collateral story has improved.
SoftBank's balance sheet is no longer just a collection of speculative venture stakes. It holds a controlling position in Arm, the chip designer whose shares have rallied on AI demand, and its OpenAI stake is now one of the most valuable private equity positions in technology. The company's 2026 annual report emphasizes that for AI infrastructure investments such as power and data centers, it will pursue debt through project finance as non-recourse financing that does not impact its loan-to-value ratio. Chief Financial Officer Yoshimitsu Goto has said asset-backed financing should be prioritized, noting that holdings in OpenAI "could provide financing capacity in the future."
The company's financial policy is to keep its LTV — adjusted net interest-bearing debt divided by the equity value of holdings — below 25% in normal times, with an upper threshold of 35% in emergencies. As long as Arm and OpenAI hold their value, SoftBank retains borrowing capacity even at speculative-grade pricing.
There is also the view that the bond market is being paid adequately for a risk that equity investors may be underpricing. A coupon near 10% compensates creditors even if OpenAI's path to profitability takes years. Bondholders do not need OpenAI to become the most valuable company in history; they need SoftBank to service debt and refinance maturities. On that narrower test, the current liquidity position looks manageable: S&P Global Ratings assesses SoftBank's liquidity as adequate and expects liquidity sources to exceed uses by more than 1.2 times over the next 12 months.
The Counter-Thesis: Leverage Amplifies a Single Outcome
The bear case is not that SoftBank cannot pay its bills today. It is that the company has tied its fate to a single asset class at valuations that assume AI delivers on its most optimistic forecasts — and it has done so with borrowed money.
History offers a cautionary precedent. In 2022, the SoftBank Vision Fund posted a record 3.5 trillion yen loss — about $27.4 billion — as the valuation of its stock portfolio plummeted, and Son later said he was "embarrassed" and "ashamed" at the fund's dismal performance. The company's cumulative loss on WeWork alone exceeded $14 billion. After a $32 billion Vision Fund loss in fiscal 2022, Son said SoftBank would shift into "defense" mode, selling down stakes and taking Arm public. Little more than a year later, as AI hype mounted, he switched back to "offense."
The concern gained fresh visibility in mid-September, when SoftBank's shares fell about 11% in a single session after OpenAI said it would not seek a public listing in 2026. The selloff underscored a structural problem: much of SoftBank's financing strategy implicitly assumes an OpenAI IPO as a liquidity event. A margin loan reportedly expanded to $25 billion against Arm shares creates margin-call dynamics if chip-stock valuations roll over. The 13% OpenAI stake, while valuable on paper, may not carry proportional governance rights because OpenAI's nonprofit parent retains control and the company has scaled back plans to fully convert to a for-profit entity.
Broader market warnings are accumulating. The flood of AI funding has made some bond investors uneasy, with global AI-related debt issuance already topping $575 billion in 2026, according to a report by Goldman Sachs Group credit strategists. The concern is that the growing pile of debt could cause a hangover for markets if the technology fails to pay off for the firms making the biggest investments. SoftBank is not alone — Amazon, Alphabet, and Microsoft are all borrowing heavily — but SoftBank's concentration is unmatched. "Risks for SoftBank credit are material and have increased as concentration has increased and cash flow has come under material strain," Mark Chapman, head of telecom and media at CreditSights, wrote in a note. He calculated that SoftBank's holdings in Arm and OpenAI together make up three-quarters of the group's asset value at current levels. Not every credit analyst is as concerned. Makiko Yoshimura of S&P Global Ratings wrote that Arm's strong share-price performance has supported SoftBank's credit quality, such that the delay to OpenAI's IPO does not have an immediate negative impact on SoftBank's creditworthiness. Chapman, for his part, sees the potential for a ratings upgrade in the intermediate term if OpenAI's IPO is "timely and successful."
The strongest version of this argument, however, comes from the pricing itself. If the AI thesis were as certain as Son argues, creditors would not demand nearly 10%. The bond market is effectively telling equity investors that they are receiving insufficient compensation for the same risk — or, put differently, that the recent rally in SoftBank shares is being underwritten by yield-hungry investors who are being paid to absorb the downside first.
SoftBank in the Wider AI Debt Boom
Placing SoftBank alongside its peers sharpens the picture. The total capital raised by hyperscalers, data-center vehicles, and neo-clouds in the form of investment-grade debt, high-yield debt, and equity rose from $172 billion in 2025 to $346 billion through mid-2026, more than a 100% increase, according to research from Barclays strategist Venu Krishna. Alphabet, Amazon, Meta, Microsoft, and Oracle collectively raised about $159 billion in bonds through early June 2026, a 47% increase over their full-year 2025 issuance. The value of bond sales from hyperscalers has more than doubled this year to over $200 billion, according to LSEG data.
The difference is balance-sheet quality. Those peers carry investment-grade ratings and, in several cases, large cash balances. Oracle, for example, has been preparing multibillion-dollar debt offerings to finance AI-related data centers while reporting remaining performance obligations above $450 billion — but its borrowing costs sit well below SoftBank's because creditors underwrite cash-generating software revenue, not a private stake whose liquidity depends on an eventual listing. SoftBank is paying a speculative-grade premium precisely because its AI payoff is optionality rather than contracted cash flow.
That premium is the mechanism by which the market separates conviction from cheap capital. In earlier cycles, Son could fund big swings with telecom cash flow, Alibaba proceeds, and patient Vision Fund capital. Today's AI buildout is too large for that model: commitments approaching $65 billion to a single company require the bond market, and the bond market charges for what it cannot see.
What Would Prove the Bulls Wrong
The bull case rests on a chain of assumptions: AI adoption continues to accelerate; OpenAI's valuation keeps rising toward and beyond $850 billion; Arm remains a prized AI asset; and refinancing remains available. Break any link and the leverage works in reverse.
The single most important falsifying signal is the OpenAI IPO timeline. SoftBank's own July 1 filing noted the third tranche's closing date "may be subject to acceleration in the event of a public listing of OpenAI Group PBC shares." If OpenAI does not list by the end of 2027 and its private-market valuation stalls or declines, SoftBank's most liquid source of takeout financing disappears precisely when its debt maturities begin to cluster. A second warning sign would be Arm's share price: a sustained 30% decline from current levels would pressure the margin-loan cushion and force either asset sales or additional expensive borrowing.
Son, for his part, has dismissed such caution. Speaking at SoftBank World 2026 in Tokyo on July 14, he called concerns about an AI bubble "absurd" and predicted the technology will require $5 trillion in annual investment by 2040.
"The business model will be viable because by 2040, if AI revenue makes up 20% of global GDP, spending 800 trillion yen a year is a rounding error," Son said.
That is the crux of the disagreement. Son is underwriting a structural shift in the global economy. The bond market is underwriting a loan.
Outlook: Three Scenarios for the Next 12 Months
Base case. SoftBank refinances its bridge facilities without distress, OpenAI continues private funding rounds at rising valuations, and the shares trade in a wide range as investors weigh AI optionality against financing costs. The credit spread stays elevated but stable, in the 350-to-450 basis point range.
Upside case. OpenAI lists in 2027 at a valuation above $1 trillion, Arm's AI-driven growth continues, and SoftBank's LTV falls as asset values rise. The shares re-rate higher and the company can refinance at lower spreads, potentially back toward investment-grade territory over time. Beneficiaries would extend to other AI-exposed borrowers who benefit from a functioning high-yield market.
Downside case. AI spending fails to generate expected revenue, OpenAI delays its IPO again or downsizes, and Arm's valuation compresses. SoftBank's refinancing costs rise further, potentially forcing asset sales at unfavorable prices. The most exposed are the high-yield bondholders, followed by equity.
Short term, the completed financing removes a near-term funding overhang, and the shares may find support from the removal of uncertainty. Medium term, the key variable is execution — whether the Stargate data-center project and pending acquisitions of ABB Robotics and DigitalBridge generate returns that exceed the roughly 9% cost of debt. Long term, the question is structural: whether AI becomes the 20% of global GDP that Son forecasts, or whether the current investment wave proves cyclical and mean-reverting.
The market's split verdict may be the most honest assessment available: the AI opportunity is real, but so is the price of betting on it with borrowed money. SoftBank's investors are not ignoring the credit risk — they are simply being paid enough to tolerate it, for now.
Explore more exclusive insights at nextfin.ai.
