SoftBank Is Borrowing Its Way Into OpenAI — Again

The Borrowing Never Stops

SoftBank Group is issuing approximately $11 billion in bonds to fund its latest OpenAI investment. The deal — split between a $10 billion dollar tranche and a €1 billion euro tranche — could become the largest bond issuance by a non-financial company in Asia-Pacific or Japan. But the headline number obscures a more pressing story: SoftBank is paying for its AI ambitions with borrowed money it doesn’t have, on terms it can barely afford.

This isn’t new behavior. It’s a pattern, repeated with increasing frequency.

Last year alone, SoftBank borrowed $41.5 billion across three rounds of financing. In February, it agreed to invest an additional $30 billion in OpenAI. Within a month, it was borrowing another $40 billion as a bridge facility, with JPMorgan Chase, Goldman Sachs, Mizuho, SMBC, and MUFG lined up as lenders. The bridge matures in March 2027. Just last month, SoftBank requested yet another $10 billion in two-year financing, partly to repay some of that bridge debt.

The cumulative commitment to OpenAI is now nearing $65 billion by October 2026. That figure is not merely a record for SoftBank — it is among the largest single corporate investments in a private company in history, surpassing previous benchmarks set during the 2021 tech investment boom. Yet unlike those earlier commitments, which were largely funded through asset sales and existing cash reserves, this round is being financed almost entirely through new debt.

Junk-Grade Faith in AI

What makes this sequence notable isn’t the scale — SoftBank has been big and leveraged for years. It’s the composition. SoftBank is increasingly funding equity stakes in AI companies with speculative-grade debt. The bonds being issued for OpenAI carry maturities of 3.5, 5.5, 7.5, 4, and 6 years. That means SoftBank is locking in long-term, high-cost borrowing to finance an investment in a company that is still not profitable.

S&P Global Ratings downgraded SoftBank to “BB” in 2023, citing concentrated asset risk. The rating was later raised to “BB+”, but analysts have flagged the single-name exposure to OpenAI as a concern. The rating agencies see the problem clearly: SoftBank is leveraging itself further precisely when its biggest bet has yet to generate any financial return.

The implications extend beyond credit metrics. A BB+ rating sits just one notch above “junk” territory, meaning a single adverse event — a regulatory setback for OpenAI, a prolonged AI Winter, or a broader credit market contraction — could trigger automatic selling by institutional holders bound by rating thresholds. That creates a cascading risk: forced selling depresses the value of SoftBank’s holdings, which weakens its balance sheet further, which raises borrowing costs, which forces more liquidation. It is a feedback loop that Magnussen Economics has flagged in internal modeling, though the firm declined to comment publicly on its findings.

This matters beyond SoftBank’s balance sheet. The conglomerate is one of the most visible symbols of corporate AI investment in Asia, and its financing strategy sends a signal to markets about the cost of AI capital. If the world’s largest tech investor is borrowing at junk rates to fund AI equity positions, the question isn’t whether the AI buildout is expensive — it’s whether the current model of financing it is sustainable.

The Bridge That Keeps Getting Longer

Some of what SoftBank is doing may be accounting theater. Rolling over bridge loans, reshaping debt tranches, and layering new borrowings against old ones are standard moves for a company with a complex capital structure. A portion of the recent borrowings may simply be refinancing rather than representing genuinely new debt.

But even if you strip out the recycling, the direction is unmistakable. SoftBank sold between $5.8 billion and $6 billion worth of NVIDIA stock to help fund its OpenAI commitments. The fact that it is liquidating one of the most valuable holdings in the AI supply chain to finance a downstream AI bet suggests the company is running out of easy options for raising capital on its own terms.

There is a second-order effect here that warrants attention. When a company this size sells a core holding like NVIDIA shares, it signals to the market that even the most confident AI investors are feeling liquidity pressure. That signal ripples outward. Other institutional holders of AI-equivalent assets may reassess their own positions, not because their thesis has changed, but because the market interprets SoftBank’s liquidation as insider knowledge of deeper strain. The result is a subtle but real drag on the valuations of the very companies SoftBank is trying to support.

The market will price this deal through its coupon. Investors will demand a risk premium for lending to a junk-rated company that is doubling down on a single, unprofitable portfolio position. If the rates come in steep, it confirms what the ratings already say: the market sees risk. If they come in relatively moderate, it may reflect a broader conviction among investors that AI equities will eventually pay off — and that SoftBank’s timing is sound.

The Cost of Conviction

What distinguishes this cycle from SoftBank’s earlier borrowing sprees is the absence of a diversification cushion. During the 2020–2021 period, the Vision Fund portfolio contained dozens of companies across autonomous vehicles, logistics, fintech, and biotech. Many of those bets failed or underperformed, but the breadth meant that no single loss was catastrophic in isolation. Today, SoftBank’s AI exposure is narrowly concentrated. OpenAI alone accounts for a growing share of the conglomerate’s total market value. ARM, its other crown jewel, has delivered modest returns since its 2023 IPO but does not come close to offsetting the capital deployed at OpenAI.

This concentration creates a dependency on OpenAI’s trajectory that SoftBank has never before shouldered. The company is not merely betting on AI — it is betting that OpenAI wins decisively, quickly, and profitably. If OpenAI remains private and continues to burn cash, SoftBank’s equity stake may never realize liquidity. If OpenAI goes public at a valuation lower than expected, the paper gains that currently justify the borrowing assumption could evaporate. And if the company faces regulatory constraints that slow its product roadmap, the entire leverage thesis unravels on a timeline that could coincide with SoftBank’s debt maturities.

What Happens Next

The immediate test is pricing. Bond markets will reveal how much extra yield investors require for the 3.5- to 7.5-year tranches. That number will matter more than the headline figure. A tight book would suggest that lenders remain confident in SoftBank’s AI narrative despite the credit rating. A wide spread would confirm that the market is pricing in tail risk — the possibility that this cycle ends differently than the last.

The longer-term question is whether OpenAI ever produces enough value to justify the leverage. SoftBank’s entire AI thesis rests on a handful of bets — OpenAI leading, presumably, with ARM and others contributing. None of these have delivered profits commensurate with the capital deployed. The company is betting that the AI infrastructure buildout will eventually generate returns that justify the cost of debt. That bet is not wrong in principle. It is wrong only if the timeline is longer than the maturity of SoftBank’s bonds.

There is also a structural question that will define SoftBank’s options going forward. The company has demonstrated an ability to raise capital in deep markets, but that ability is not infinite. Each new issuance increases the debt overhang, which makes subsequent borrowings more expensive and less attractive. At some point, the marginal cost of raising another dollar exceeds the marginal return that SoftBank expects from deploying it. The company has not reached that inflection point yet — but it is moving toward it with every tranche it places.

For now, SoftBank is throwing borrowed money into a well with no visible bottom. The question isn’t whether the company will stop — it’s whether it will run out of lenders before the bets pay off. And if they don’t pay off soon, the next bond offering won’t be priced by conviction. It will be priced by desperation.