Everything About AI
NextFin News - The cost of insuring SoftBank Group's debt has climbed to its highest level in four years, reaching 424 basis points on five-year credit default swaps, as investors price the strain of the conglomerate's all-in artificial intelligence funding spree. The widening in protection costs - up 24 basis points in a single week - places SoftBank alongside Oracle and CoreWeave in a rare, synchronized repricing of the balance sheets that have financed the AI buildout.
The Credit Signal Investors Cannot Ignore
Credit default swaps function as the bond market's truth serum. When the price of protecting against a borrower's default rises, it is investors voting with capital rather than headlines, and the vote on SoftBank has turned decisive. The five-year CDS spread of 424 basis points marks the widest level since 2022, according to data compiled by S&P Global Market Intelligence and LSEG. For context, a spread above 400 basis points sits at the threshold where credit markets begin to price meaningful default risk rather than routine funding volatility, and SoftBank's move through that line in October was not an isolated twitch.
SoftBank is not alone at that threshold. Oracle's five-year CDS spread hit a record 288 basis points on October 15, up more than 12 percent in a week and 26 basis points overnight. CoreWeave, the AI cloud-computing company whose fortunes are intertwined with both, saw its five-year swaps soar above 1,140 basis points before settling near 1,088 basis points - a 33 percent weekly jump and 177 percent higher than a year ago. A spread above 1,000 basis points is generally read as a signal of credit distress. Three companies, three different corners of the AI capital stack, one shared repricing.
The combination matters because it is no longer a single-name story. When the investor, the chip buyer, and the cloud renter all see their funding costs rise together, the market is no longer questioning a company - it is questioning a financing model. SoftBank sits at the top of that chain: it funds OpenAI, which anchors demand for CoreWeave's compute, which in turn justifies the chip orders that Oracle and others are financing. Stress at any link travels up and down the chain, and the CDS market is now pricing that linkage explicitly.
How SoftBank Got Here: Debt-Fueled Conviction
The path to 424 basis points was paved deliberately. On September 24, SoftBank announced a record $11.1 billion sale of dollar- and euro-denominated high-yield bonds, offering yields of up to 9.75 percent - among the richest coupons ever attached to a Japanese corporate issue. The proceeds were earmarked for the company's commitment to OpenAI, which has grown to nearly $65 billion across successive investment rounds. That issuance capped a year in which SoftBank had already sold $14.6 billion of high-yield bonds, accounting for 63.4 percent of all Asia Pacific and Japan high-yield corporate issuance in 2026. In other words, SoftBank alone made up nearly two-thirds of an entire region's junk-bond market in a single year. Concentration of that size does not go unnoticed by credit investors, and it does not come cheap.
The credit-rating agencies have been circling the position for months. In March, S&P Global Ratings cut its outlook on SoftBank to negative from stable while affirming its BB+ long-term issuer rating, citing the additional $30 billion OpenAI investment and warning that restoring asset liquidity and portfolio quality could take longer than expected. S&P noted it would consider a downgrade if certain scenarios materialized over the following six to 12 months - a window that now runs into early 2027. In September, Fitch Ratings assigned its first-ever rating on the group, BB+, one notch below investment grade, as it priced the proposed notes. Two agencies, one message: the AI bet has moved SoftBank's credit profile into territory where margin for error is thin.
Beneath the credit metrics sits a balance sheet of staggering scale. As of June 30, 2026, SoftBank reported net assets of ¥72.30 trillion and consolidated net interest-bearing debt of ¥22.29 trillion, with net debt of ¥10.81 trillion after adjustments for self-financing entities. The company's own net-asset-value disclosure showed a loan-to-value ratio of 13.0 percent - a figure management points to as evidence of conservative leverage. But LTV measures collateral coverage, not cash-flow coverage, and that distinction is where the credit argument turns. A portfolio of private technology stakes can be marked up in good markets and marked down, or rendered unsellable, in bad ones. Collateral that cannot be liquidated without moving the price is collateral in name only.
The Mechanism: Why AI Debt Is Different Debt
Traditional corporate debt is repaid from operating cash flow. AI infrastructure debt, in the current cycle, is repaid from a chain of promises: the cloud renter promises the hyperscaler will keep buying compute; the hyperscaler promises the AI model will keep generating revenue; the model promises the technology will keep improving fast enough to justify the spending. Every link in that chain is real, but every link is also forward-dated. That forward-dating is the entire story.
CoreWeave's own numbers illustrate the tension with unusual clarity. The company carries more than $51 billion of debt and obligations, yet it is running heavily negative free cash flow as it buys chips and builds data centers ahead of demand. Its 9.75 percent bond due 2031 was yielding 15.78 percent in mid-October, and its 2032 bonds traded near 77 cents on the dollar. The market is not saying demand is fake. It is saying the cash flows must arrive on schedule, and that any delay compounds because the debt service does not wait.
"It's obviously heavily negative free cash flow at the moment, and it needs that cash flow to arrive really to justify all the capex that it's spending," said Gavan Nolan, executive director of fixed income pricing at S&P Global Market Intelligence.
The Evercore ISI trading desk put the same point more bluntly: the market is "basically saying 'no mas' to AI debt without operating cash flow support." Jim Reid, global head of macro and thematic research, warned of the feedback loop: "The problem is this kind of concern can prove to be a self-fulfilling prophecy, and risks triggering broader issues in other AI-related names."
For SoftBank, the transmission channel runs through asset liquidity rather than operating earnings. The conglomerate's credit quality depends less on steady cash flow than on its ability to sell stakes - Arm, T-Mobile, and its listed holdings - at favorable prices when cash is needed. That worked in buoyant markets. In a credit-skeptical environment, forced selling into weaker demand is precisely the dynamic S&P flagged when it turned the outlook negative. The second-order effect is what makes this different from an ordinary spread widening: if SoftBank must sell assets to restore liquidity, it depresses the very portfolio values that back its debt, which widens spreads further, which raises the cost of the next bond issue. That loop is what a 424-basis-point spread is pricing.
There is also a cross-asset channel that most investors are not tracing. Higher CDS spreads on SoftBank feed directly into the pricing of its next bond deal. The September issuance already demanded yields up to 9.75 percent - a full percentage point or more above where SoftBank priced debt in calmer periods. Every incremental basis point of funding cost is a basis point less return on the AI portfolio, and the AI portfolio is a levered bet where small changes in the cost of capital compound into large changes in the terminal outcome. The credit market is effectively charging SoftBank a risk premium on its conviction, and that premium grows with every widening.
The Counter-Case: Conviction Backed by Cash
The bull case for SoftBank is not sentimental. It rests on a simple arithmetic claim: the AI trade is early, the returns will be enormous, and today's credit spreads will look like noise in hindsight. Chief Executive Masayoshi Son has been explicit about the scale he expects.
"In 2040, approximately 20% of the world's GDP will be replaced by AI-related industries, the world of superintelligence," he said in a recent interview
, adding that he estimates almost $5 trillion in annual global investment will be needed to build out data centers, chip production, and energy infrastructure. On that view, a 424-basis-point CDS spread is the price of admission to the defining investment cycle of the decade, not a warning sign.
SoftBank's earnings support the confidence, at least retrospectively. In the fiscal year through March 2026, the company reported net profit of approximately ¥5 trillion, nearly five times the prior year, as its AI investments began to pay off. The Vision Funds hold stakes in the companies building the AI layer, and the domestic telecom arm generates steady cash flow that services the group's debt even when investment gains are quiet. From this vantage point, the credit market is making a category error - it is underwriting SoftBank as a levered holding company when the bull case says it should be valued as the venture capitalist of the AI era.
There is also the matter of who is on the other side of the AI trade. CoreWeave's backlog exceeds $100 billion, with roughly two-thirds contracted to investment-grade counterparties including Microsoft, Meta, IBM, and Caterpillar, according to Deutsche Bank. If those customers honor their contracts, the cash flows do arrive.
"When you are building a new type of technology company there will be periods of time when the market misunderstands and acts irrationally," CoreWeave Chief Executive Mike Intrator wrote in a memo. "We are confident in our backlog which more than supports all of our debt obligations."
The counter-thesis is strongest on one point: timing. Even a correct long-term view can be wrong on financing. Debt has maturities. If credit conditions tighten before AI revenue matures, the companies with the most leverage - not the worst long-term prospects - face the most immediate pressure. That is the asymmetry the credit market is pricing, and it is why the bull case's confidence about 2040 does not settle the question of 2027.
Cyclical Strain or Structural Shift? The Call
The funding strain itself is cyclical. Credit spreads widen, sentiment sours, and capital becomes scarce - as it has in every technology boom from the railroad era to the dot-com bubble to the telecom buildout of the late 1990s, a cycle SoftBank knows intimately from its own broadband bet. Spreads that reach 424 basis points on fear can tighten back to 250 on a single strong earnings print from OpenAI or a successful asset sale. History says mean reversion is the default for credit sentiment.
But the underlying shift is structural, and that is what makes this cycle different from the ones before it. AI infrastructure is a capital-intensive, winner-take-most industry where the companies that secure financing early will own the compute layer for a generation. The capital requirements are not a passing phase - they are the business model. A data center built today must be serviced for a decade; a chip order placed today locks in depreciation schedules that do not care about the credit cycle. SoftBank's bet is that the structural prize outweighs the cyclical cost of capital. The credit market is asking for proof that the prize arrives before the bill comes due.
So the correct read is a cyclical wave riding on a structural shift. The spreads will mean-revert - but only if the structural revenue arrives first. If it does not, the cyclical rebound becomes a trap, and 424 basis points will look like the floor rather than the ceiling.
What This Means Across Time Horizons
Short term (sentiment and liquidity): The pressure is likely to persist. Credit spreads lead equity prices, and a CDS move of this magnitude typically forces a repricing of the equity as well. SoftBank shares have already reflected part of the strain, trading in a 52-week range of ¥3,365 to ¥9,074 and changing hands near ¥5,000 in mid-October - far below the year's highs. Any further widening toward 500 basis points would signal that credit investors are testing a downgrade scenario, not just expressing unease.
Medium term (fundamentals): The decisive variable is OpenAI's path to profitability and SoftBank's ability to monetize its stake without depressing the valuation. The $65 billion commitment is not a single outflow but a series of tranches, which gives management optionality - it can slow the pace if conditions deteriorate. If OpenAI's revenue trajectory justifies the valuation at each tranche, the credit story stabilizes. If the pace slows, the negative outlook from S&P becomes a downgrade, and funding costs rise again - a feedback loop that is difficult to exit. The six-to-12-month window S&P set in March is the clock to watch.
Long term (structural): If AI adoption follows Son's trajectory, today's credit spreads will be a footnote. The conglomerate that controls the largest stake in the leading model developer will be underwritten on cash flows that do not yet exist but may dwarf them when they do. The risk is not that the thesis is wrong; it is that the financing structure assumes a smooth path to a future that arrives in lumps.
What to Watch
- The 500-basis-point line: If SoftBank's five-year CDS pushes through 500 bps, the market is pricing more than strain - it is pricing a downgrade to sub-BB territory.
- S&P's next move: The negative outlook carries a six-to-12-month window from March 2026. Any downgrade to BB or below would trigger forced selling by investment-grade-only mandates and lift funding costs across the group.
- CoreWeave's cash flow: As the most levered link in the chain, CoreWeave's ability to convert its $100 billion backlog into operating cash flow is the canary. A 1,000-basis-point CDS level that holds or widens keeps pressure on the entire AI debt complex.
- SoftBank's asset sales: Management's stated option to restore liquidity is to sell holdings. The prices it achieves - and whether sales look strategic or forced - will tell investors how much cushion really exists.
- The next bond deal: After the $11.1 billion September issuance, the coupon SoftBank must pay on its next dollar or yen offering is the cleanest read of whether the credit market is calming or hardening.
The falsifying signal for the bearish read is precise: if CoreWeave reports positive operating cash flow within the next two quarters while SoftBank's CDS tightens back below 300 basis points, the stress is cyclical noise and the credit bears are wrong. If instead the 1,000-basis-point distress level in AI cloud debt holds and spreads to more counterparties - Oracle already at a record 288 basis points is the first domino - the strain is structural and the repricing has further to run.
The Bottom Line
SoftBank's 424-basis-point credit spread is not a verdict on artificial intelligence. It is a verdict on leverage, timing, and the price of conviction. Masayoshi Son has made the largest venture bet of his career with borrowed money, and the bond market is now charging him rent on that conviction. The AI thesis can still be right and the credit trade can still hurt - because in the window between a correct long-term view and a due near-term debt payment, the market asks only one question: can you make it to the future on the terms you borrowed to get there?
Explore more exclusive insights at nextfin.ai.
