tech-ai

SoftBank’s $25B Arm Loan Shows AI’s Debt Bill Is Getting Dangerous

SoftBank is borrowing $25 billion against Arm to buy more OpenAI exposure. That is not conviction; it is leverage wearing a robot costume.

SoftBank’s $25B Arm Loan Shows AI’s Debt Bill Is Getting Dangerous

SoftBank is borrowing $25 billion against Arm to buy more OpenAI exposure. That is not conviction; it is leverage wearing a robot costume.

Masayoshi Son has built fortunes by making huge calls before the crowd was comfortable. He also knows better than most that a brilliant asset can become a dreadful investment if you pay too much, borrow too hard, or need the market to stay euphoric long enough to refinance the bet.

This week, SoftBank increased a margin loan backed by Arm Holdings shares from $20 billion to $25 billion. Bloomberg also reported that Apollo Global Management is discussing increasing a separate SoftBank loan to $9 billion from $5.4 billion to support its OpenAI investment. Add an expanded $6.5 billion credit line and the possibility of a further $10 billion to $20 billion bond sale, and the message is pretty hard to miss: SoftBank is not merely investing in AI. It is constructing a financing machine around it.

That is the real tech story of September 2026. Not another chatbot trick. Not a bloke in a black turtleneck saying “agentic” 14 times. The AI race is becoming a balance-sheet contest.

SoftBank Has Turned Arm Into Its AI ATM

The immediate headline is simple enough. SoftBank raised the size of a margin loan secured by Arm shares to $25 billion. It is reportedly the third time the company has enlarged this facility.

A margin loan is not free money. It is borrowing against the value of an asset. In this case, the asset is Arm, the chip designer whose architecture sits underneath an enormous amount of the world’s smartphones and increasingly matters to data-centre computing. If the collateral falls far enough, lenders can demand more security, repayment or, depending on the agreement, sell pledged shares.

That is the bit people skip while applauding the “vision.”

SoftBank’s public filings have previously warned that margin loans backed by Arm shares can be subject to early settlement after a significant decline in the value of the pledged stock. That does not mean Arm will collapse. It means the financing has a pressure point, and anyone pretending otherwise is doing public relations, not analysis.

Son needs the cash because SoftBank’s OpenAI commitment is enormous. SoftBank announced on February 27 that it would make up to $30 billion in follow-on investments in OpenAI during 2026 through Vision Fund 2. It had already invested $34.6 billion in OpenAI since September 2024. The company has said the new deal could lift its cumulative OpenAI investment to roughly $64.6 billion and its ownership to about 13%.

That is not a venture punt. It is a portfolio-defining position.

SoftBank also signed a $40 billion bridge facility in March, primarily to finance that OpenAI follow-on investment. The company drew $10 billion in April for the first tranche, then completed a second $10 billion tranche on July 1. A bridge loan is called a bridge for a reason: eventually, you need to get to the other side. That generally means selling assets, raising equity, issuing bonds, refinancing, or having your investment become sufficiently liquid and valuable that the problem disappears.

The fourth option is the one everyone is pricing in.

The AI Boom Is Moving From Equity Risk to Credit Risk

Here is the overlooked angle: the question is no longer only whether OpenAI can justify its valuation.

The question is who is carrying the financing risk while the industry waits for the revenue to catch up with the capital expenditure.

For years, AI enthusiasm has been expressed through equity markets. Buy Nvidia. Buy cloud companies. Buy anything with a data centre, a GPU allocation or a CEO capable of saying “inference” without blinking.

Debt changes the game.

Equity investors can be wrong and wait. They might suffer, but there is no maturity date attached to disappointment. Lenders are less romantic. Debt has interest, covenants, collateral values and refinancing dates. It turns a long-term technology thesis into a series of nearer-term financial tests.

SoftBank’s moves are a clean example. Its assets may be excellent. Arm is a real strategic asset. OpenAI may well be one of the most important companies of this decade. But excellent assets do not eliminate financing risk; they can tempt founders and investors into taking more of it.

That is how leverage works. When things go well, it makes you look cleverer than you are. When things go badly, it reveals the difference between owning a valuable asset and having control of your financial destiny.

I am not calling this a blow-up. Anyone who claims certainty around a business moving this quickly is kidding themselves. I am saying the AI trade is becoming more fragile than the headlines suggest because more of it is now being funded with obligations that cannot be paid with PowerPoint slides.

Why Son May Still Be Right

Before the usual brigade starts throwing empty bottles, let’s be fair: Son’s basic thesis is not stupid.

If OpenAI becomes foundational infrastructure for knowledge work, software, search, customer service, design, research and autonomous systems, ownership at scale could be immensely valuable. If Arm becomes a larger beneficiary of AI computing, then borrowing against it may look, in hindsight, like a rational use of a high-quality asset.

Son has made contrarian bets that looked unhinged before they looked obvious. He also built SoftBank around the idea that technology’s biggest winners can compound at a rate traditional finance struggles to comprehend.

The case for this strategy rests on three things going right at once.

First, OpenAI needs to turn extraordinary usage and strategic importance into durable, expanding cash flow.

Second, Arm needs to retain enough market value that collateral-backed financing remains comfortable rather than stressful.

Third, capital markets need to stay open and receptive while SoftBank refinances bridge debt and considers further bond issuance.

None of those things is impossible. All three are plausible. But stacking plausible outcomes is not the same as having a margin of safety.

That distinction matters.

The Contrarian Take: This Is Good News for Smaller Operators

The big AI players are spending like they have discovered a new continent. That can be intimidating if you are building a company, running an established business or investing your own money.

Ignore the theatre. You do not need a $25 billion margin loan to benefit from AI.

In fact, the debt-fuelled arms race creates a useful opening for smaller operators. The giants will spend billions on chips, data centres, model training, talent wars and grand strategic partnerships. Your job is not to beat them at that game. Your job is to use the tools they subsidise to build a sharper, cheaper and more useful business.

The winners beneath the model layer will often be businesses that solve ugly, expensive, specific problems: reducing quote turnaround time, improving inventory accuracy, finding fraud, triaging customer enquiries, automating compliance work, helping salespeople prepare better, or making a niche marketplace less painful.

That is where sensible people make money.

The infrastructure owners are fighting a war of capital intensity. Most founders should be fighting a war of customer value.

And investors should be careful not to confuse the two. Buying a business because it says AI is not investing. It is outsourcing your thinking to a marketing department.

What this means for you

If you run a business, do three things this week.

1. Treat AI spend like a hiring decision. Do not buy a dozen tools because your competitors posted about them on LinkedIn. Pick one costly, repetitive workflow. Set a baseline: hours, error rate, conversion rate or margin. Run a 30-day test. Keep the tool only if it produces a measurable gain.

2. Do not finance experimentation with permanent overhead. The lesson from SoftBank is not “never take risk.” It is that financing structure matters. Use monthly contracts, pilots and reversible decisions until the return is proven. Fixed costs are lovely right up until revenue has a bad quarter.

3. Build around the customer problem, not the model. Models will improve and prices will fall. Your advantage needs to be workflow, distribution, trust, proprietary data, brand or a customer relationship that took years to earn.

For investors, the rule is even simpler: separate the AI opportunity from the AI capital structure. A fantastic technology can sit inside an overextended vehicle. A mediocre company can ride a hot theme for longer than it deserves. Ask what must go right, what debt is involved, when it matures, and who gets hurt if the collateral drops.

AI will create enormous wealth. I am convinced of that.

But the money will not all go to the loudest company, the biggest debt package or the founder making the grandest prediction. It will go to the people who can turn powerful technology into cash flow without needing the market to stay drunk forever.

Sources