Everything About AI
NextFin News - The cost of insuring SoftBank Group's debt is hovering near its highest level in roughly three years, as credit investors reprice the conglomerate's $64.6 billion wager on OpenAI after the chatbot maker said it will not go public in 2026. The message from the default-protection market is blunt: the liquidity event that SoftBank's financing stack was built around has been pushed out, and the bill for that delay is now coming due in wider spreads.
Sam Altman, OpenAI's chief executive, said in a mid-September interview that taking the company public this year would be "an ill-advised moment" given mounting artificial-intelligence safety concerns, adding "not 2026" when pressed on timing. OpenAI had been exploring a listing that could have valued it at around $1 trillion. For SoftBank, which has concentrated more of its balance sheet in a single pre-IPO company than at any point in its history, the delay is not a calendar slip. It is a stress test of the funding architecture Masayoshi Son constructed around the bet.
The Setup: A Bridge That Assumed an Exit Inside Twelve Months
The mechanics of SoftBank's position are now a matter of public record. On Feb. 27, the company announced a $30 billion follow-on investment in OpenAI through SoftBank Vision Fund 2, payable in three $10 billion tranches on April 1, July 1 and Oct. 1. Once the third tranche closes, cumulative investment reaches $64.6 billion for an ownership interest of approximately 13 percent, with preferred shares that convert to common upon an initial public offering. That followed a $22.5 billion investment completed in late December 2025, which lifted the stake to about 11 percent, and an initial commitment of up to $30 billion made in April 2025.
To fund that buildout, SoftBank secured a $40 billion unsecured bridge financing facility in late March, syndicated by a consortium of American and Japanese banks and carrying a twelve-month maturity. The structure is the tell: a one-year, unsecured loan of that size only makes sense if the borrower expects a liquidity event inside the window. The market read it that way. Shares rose on the announcement, and the company proceeded to broaden the lender base, adding 21 new institutions by late July for roughly $7 billion of allocations.
Then the exit moved. With the IPO now targeted at 2027, the bridge no longer looks like a bridge. It looks like a refinancing that will have to be executed in a credit market that has turned more hostile to AI-linked balance sheets. That repricing is what shows up in the default-swaps market. SoftBank's five-year credit default swaps last drew widespread attention at about 355 basis points in early March, the widest level since April 2025 and roughly 100 basis points wider than Nissan Motor, leaving them the costliest protection among Japanese corporates. The spread has since widened again toward a multi-year high on the IPO-delay news.
The equity market delivered the same verdict faster. SoftBank Group shares fell about 11 percent on Monday, Sept. 14, to their lowest level since August 2025, as the AI-safety debate that Altman himself joined spilled over into the stocks of companies financing the buildout.
"I actually think that, given everything happening with safety, right now would be an ill-advised moment to go public, and we don't feel pressure on that," Altman said in the interview. "I would say not 2026. Yeah, we got a lot of stuff to do, like meeting this moment of what is going to be required for safety and alignment, and how the industry and governments can work together."
When Son completed the December tranche, he framed the relationship differently. "We are deeply aligned with OpenAI's vision of ensuring AGI benefits all of humanity," he said in a company statement. A year later, alignment and exit timing have become two different conversations.
How an IPO Delay Becomes a Credit Spread
The first-order channel is mechanical. An IPO converts a large, privately valued, illiquid stake into a tradable, mark-to-market holding that can be sold, pledged, or used to repair balance-sheet ratios. Without an IPO, the stake stays locked on the books at a value that lenders cannot independently verify in a liquid market.
The second-order channels are where the damage compounds, and there are three of them.
The collateral channel. By late summer, reports indicated that SoftBank's attempt to use its OpenAI holding as collateral for additional borrowing had been held up as lenders grew more cautious about extending credit backed by private companies. A delayed IPO makes that problem structural rather than temporary: the single asset that could unlock fresh funding is now harder to underwrite precisely because its exit path has lengthened.
The refinancing channel. The $40 billion bridge matures in March 2027. If OpenAI is not public by then, SoftBank must refinance at spreads that reflect the added uncertainty, or draw down the cash buffer it has been carefully maintaining. That is not a hypothetical: the company is already in the market arranging additional borrowing, having sought a $10 billion loan to help refinance debt used for the OpenAI investment.
The rating channel. S&P Global Ratings cut its outlook on SoftBank to negative from stable in March, citing the time needed to restore asset liquidity and portfolio quality, and flagged OpenAI as among the company's investments with the weakest credit quality. The agency said it may consider a downgrade from BB+ if it sees a heightened likelihood of adverse scenarios over the next six to twelve months. A delayed IPO pushes the restoration of liquidity further out and raises the probability that the rating action arrives before the refinancing does.
Here is the distinction that matters. A credit default swap spread is not primarily a view on whether a company will default tomorrow. It is the market's price for the option that things get worse before they get better. What the CDS market is pricing now is not SoftBank's insolvency. It is the loss of optionality.
Cyclical Shock, Structural Shift: Two Forces, One Spread
Getting the cyclical-versus-structural call right determines whether this spread is a buying opportunity or a new normal. Both forces are present, and they point in opposite directions.
The cyclical leg is a timing problem, and timing problems revert. OpenAI has not said it will never go public; Altman pointed to 2027 as a more plausible window. The bridge has a defined maturity. SoftBank reported a loan-to-value ratio of 13 percent at the end of June, well inside its self-imposed 25 percent limit, and cash of 2.3 trillion yen even after the position buildout. The company maintains a financial policy of keeping LTV below 25 percent in normal times and holding at least two years' worth of bond redemptions in cash. If OpenAI lists in 2027 at a valuation that validates today's private marks, the spread can give back a meaningful portion of the widening. On that reading, this is a liquidity-timing shock, and liquidity shocks mean-revert.
The structural leg is a concentration problem, and concentration discounts do not revert on their own. A $64.6 billion position in a single private company is not the old SoftBank model of a diversified Vision Fund portfolio spread across hundreds of holdings. It is a concentrated, venture-style bet executed at the holding-company level and funded with corporate debt. That is a structural change in the risk profile of the credit. Even if the IPO happens on schedule in 2027, the market is unlikely to underwrite SoftBank's credit again on the assumption that its largest position is readily exitable. The concentration discount, once applied, tends to stick.
The judgment: the widening is roughly two-thirds cyclical and one-third structural. That split matters for how to read the tape. It means the CDS can surrender a large share of the move without the underlying credit story being fixed, and it means a successful 2027 IPO would not fully erase the premium.
The Question the Market Is Not Asking
The consensus read is simple and it is already priced: no IPO this year is bad for SoftBank. The question the market is not asking is whether the delay is actually the second-best outcome for the very creditors now demanding more compensation.
Consider the alternative. An IPO launched into a deteriorating risk environment could have broken issue. The SpaceX offering in June raised $85 billion and briefly pushed that company's valuation to $1.8 trillion before shares tumbled, a sequence Altman himself cited as a caution against rushing. Had OpenAI listed into that backdrop and traded below its private marks, SoftBank's 13 percent stake would be marked down in real time, and the collateral value that lenders care about would be falling rather than merely illiquid.
The uncomfortable second-order point is this: illiquidity is not impairment. A delayed IPO at higher capability and safety maturity may produce a higher exit valuation than a rushed 2026 listing. The CDS market may be over-penalizing a delay that protects the very asset backing the debt. That does not make the wider spread wrong. It makes the magnitude debatable.
The broader AI-credit complex supports the caution without confirming the panic. Oracle's five-year CDS blew past 200 basis points in July, roughly four times the investment-grade corporate index and a level not seen since the global financial crisis, as investors questioned how much debt the company had taken on to fund its own AI commitments. Nvidia's five-year CDS climbed to about 80.77 basis points in mid-August, up roughly 90 percent year-to-date and more than doubled since late May. Long-term government yields have reached their highest levels in decades, raising the cost of every leveraged bet in the complex. SoftBank is not being singled out. It is the most concentrated balance sheet in a sector where leverage is suddenly being re-underwritten.
The Bear Case, Stated at Full Strength
The strongest argument against the "over-penalized" reading is that SoftBank's financing stack is more fragile than the headline LTV figure suggests, and a delayed IPO exposes the seams. The bear case runs like this.
The $40 billion bridge is unsecured and short-dated. Rolling it in a credit market where AI-linked names are already under pressure means paying up, and the company has limited room to maneuver. Cash fell from 3.5 trillion yen at the end of March to 2.3 trillion yen, a draw of 1.2 trillion yen that shrinks the cushion available for bond redemptions and for the remaining commitments on the table: the final $10 billion OpenAI tranche due Oct. 1, a $5.4 billion acquisition of ABB's robotics business, and a $3.1 billion purchase of data-center investment firm DigitalBridge, totaling about $18.5 billion of committed investment through year-end.
In a March research note, analyst Sharon Chen projected that SoftBank's loan-to-value ratio could rise to 27 percent to 28 percent, above the 25 percent self-imposed ceiling, if asset values soften. S&P has already put the rating on a negative path. If a downgrade arrives before the refinancing, funding costs rise across the capital structure at precisely the wrong moment.
This is not a strawman. It is backed by the rating agency's own stated concerns about asset liquidity and portfolio quality, and by the reported difficulty in pledging the OpenAI stake as collateral. A company whose largest asset cannot be borrowed against is a company with less financial flexibility than its balance sheet implies.
The falsifying signal is quantifiable and it is near. Watch the next quarterly LTV disclosure and the cash balance. If LTV holds at or below 25 percent and cash remains above 2 trillion yen through the October tranche close, the bear thesis that the balance sheet is slipping loses its footing, and the CDS spread should tighten on that confirmation alone. If LTV prints above 25 percent or cash falls below 2 trillion yen, the structural-concentration thesis is confirmed and further widening is likely. There is no ambiguity in that test.
Who Is Exposed, Who Benefits, and What to Watch
The direct exposure sits with SoftBank's bondholders and the banks on the bridge facility, who now carry credit risk priced for a longer, less certain exit. Japanese retail investors being courted for the company's roughly 1 trillion yen retail bond issue, priced at a 4.75 percent coupon in what would be the largest such sale by a Japanese company, are being asked to absorb a credit whose risk profile has just been repriced upward.
On the other side, the relative beneficiaries are the counterparties who sold protection at wider premiums, rival AI investors whose balance sheets are more diversified, and companies that can fund AI capital expenditure from operating cash flow rather than balance-sheet leverage. The asymmetry is clear: SoftBank's downside is bounded by its asset base, which includes Arm Holdings, the OpenAI stake, and remaining Vision Fund holdings, but the cost of carrying that credit through a delayed-exit period is not bounded. It compounds in spread terms every quarter the IPO stays away.
The outlook splits cleanly by horizon.
Short term, zero to six months, is volatile with a widening bias. The October tranche close, the final pricing and absorption of the retail bond, and any further AI-safety headlines will drive both spreads and shares. Two-way volatility is the base case.
Medium term, six to eighteen months, turns on the March 2027 bridge maturity. The base case is a successful refinance at wider spreads, not a default. The spread should stabilize once the refinancing path is clear, even if it does not fully revert to pre-March levels.
Long term, eighteen months and beyond, is structural. SoftBank's credit will trade with a persistent concentration premium for as long as a single pre-IPO company dominates the asset base. That is a regime change in how the market underwrites the name, not a cycle.
Three scenarios frame the path. In the base case, OpenAI lists in 2027 at a valuation above today's private marks, SoftBank refinances the bridge at wider but manageable spreads, and the CDS gives back roughly half the widening. In the upside case, AI safety concerns ease, a formal industry pact is signed, and the IPO clears at a premium valuation, the concentration is rewarded rather than discounted and the spread returns toward pre-March levels. In the downside case, the IPO slips past 2027 or AI capital-expenditure economics disappoint, S&P follows through with a downgrade, LTV breaches 25 percent, and the bridge is refinanced only at punitive spreads or with asset sales.
The CDS market is not saying SoftBank cannot pay its debts. It is saying the company has bet its balance sheet on a single door, an OpenAI IPO, and that door just closed for the year. The question for 2027 is not whether the door reopens, but whether the market charges SoftBank rent for waiting in the hallway.
Explore more exclusive insights at nextfin.ai.
