On Thursday, SoftBank Group issued $11.1 billion in dollar- and euro-denominated bonds, the largest high-yield corporate bond sale on record, to fund the final $10 billion tranche of its $30 billion commitment to OpenAI. The bonds pay interest rates of 8.625, 9.25, and 9.75 percent across three dollar tranches, plus two euro tranches yielding 7.125 and 8 percent. When the deal closes, SoftBank will have committed $64.6 billion to OpenAI and will own roughly 13 percent of the ChatGPT maker. This is what financing the AI boom looks like when the money runs through the junk bond market.
The numbers deserve a pause. In June 2021, SoftBank issued a comparable stack of dollar and euro bonds paying between 2.125 and 5.25 percent. This week it paid up to 9.75 percent, the highest coupon it has ever paid on dollar debt, according to Reuters. The previous global record for a high-yield corporate sale was a $10.9 billion telecom deal in 2014. SoftBank beat it while paying nearly double the rates of five years ago. That is the arithmetic of the AI buildout: ambition expanding faster than cash flow, bridged by debt.
The market’s verdict was mixed in the way that tells you everything. Demand was enormous: preliminary indications of interest topped $20 billion before the official launch, and final demand for the dollar portion passed $30 billion, roughly three times the amount on offer, allowing SoftBank to price at the bottom of its guidance ranges. Investors want this paper. But the price of insuring SoftBank’s debt against default tells the other side: its five-year credit default swap spread moved above 400 basis points this week, up from about 280 in June, Reuters reports. Eager buyers, nervous insurers. That is a market that believes in the story and fears the balance sheet at the same time.
SoftBank is not alone in borrowing to fund AI; it is just the most dramatic example of the debt layer of the boom. The largest U.S. hyperscalers have been issuing record amounts of zero-coupon convertible bonds, and S&P has warned that Big Tech credit ratings could be at risk. Investor Michael Burry wrote in a widely shared Substack post this week that the S&P 500’s net capital expenditure has reached 2.07 percent of GDP, the highest since March 2000, and put total AI-related obligations near $3 trillion. Whether or not his numbers prove exact, the direction is undisputed: the AI buildout is being financed with promises, and promises have coupons.
What makes the SoftBank case study instructive is the structure of the bet. Under Masayoshi Son, SoftBank has committed to acquiring ABB’s robotics business for $5.4 billion and DigitalBridge for $3.1 billion, alongside the OpenAI stake. It has also used loans backed by holdings like Arm and OpenAI itself, which, as Finimize notes, can mechanically tighten financing conditions if those assets fall in value. This is leverage stacked on leverage: borrowed money secured by volatile equity, funding a stake in a private company whose valuation depends on the continued enthusiasm of the very market doing the lending.
Satoru Aoyama, a senior director at Fitch Ratings, captured the moment’s strangeness: “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.” Translation: the safest borrowers started this trend, and now the riskiest are joining it. When junk-bond investors are funding AI data centers, the boom has moved from the core of the credit market to its edges.
There is a reasonable bull case, and it deserves its turn. SoftBank’s bet is that OpenAI becomes one of the defining companies of the century, and that a 13 percent stake bought with 9.75 percent debt looks like genius in retrospect if the equity compounds at multiples of that. Son has made this kind of concentrated, leveraged bet before, on Alibaba, on Arm, and been vindicated. The $30 billion in orders suggests professional credit investors see enough asset coverage to sleep at night. And the 1 trillion yen retail bond SoftBank sold in Japan this month shows it can tap multiple pools of capital.
But the case study’s warning is about what happens if the music slows. Burry’s post suggested write-offs could arrive “perhaps in 2028 or 2029.” If AI revenues disappoint, the debt does not renegotiate itself. The 9.75 percent coupons are fixed; the OpenAI valuation is not. A downturn that cut the value of SoftBank’s AI holdings would simultaneously raise its borrowing costs, tighten its loan covenants, and widen those credit default swap spreads further. That is the doom loop of leveraged investing, and it is why the CDS market is charging 400 basis points while the bond buyers cheer.
For the individual investor, the lesson is closer to home than it looks. The same dynamic, borrowing against optimism, plays out in miniature whenever someone buys stocks on margin or stretches into an adjustable-rate mortgage because asset prices “always go up.” SoftBank is doing it with eleven billion dollars and a team of bankers; the mechanics are identical. Leverage magnifies conviction, and conviction is not a repayment plan.
My take is that this bond sale will be remembered as a landmark either way. If the AI boom delivers the revenues to cover these coupons, September 2026 will be the week the market funded the future at scale. If it does not, this will be the exhibit in every documentary about the bubble: the largest junk bond in history, priced at nearly 10 percent, to buy a piece of a private AI lab. Both outcomes are still on the table. The coupons, unfortunately for SoftBank, are due either way.
What to watch next
OpenAI’s revenue trajectory, SoftBank’s credit default swap spreads, and whether other AI borrowers follow into the high-yield market. If junk-bond AI issuance becomes a trend rather than an exception, the credit cycle has a new chapter.
