SoftBank Group reported a net loss of ¥1.12 trillion, about $7.1 billion, for its fiscal second quarter on Nov. 10, its first quarterly loss in almost two years, after taking a $3.7 billion writedown on U.S. data-center projects. The result reversed a ¥2.5 trillion profit a year earlier and came in far below the ¥129 billion profit analysts had expected. The Tokyo company's shares nonetheless closed up 13.9% at ¥3,872, a sign of how far expectations had already fallen: the stock is still more than 50% below its record high in June.
The loss lands at a delicate moment for the AI trade. SoftBank is one of OpenAI's largest backers and has spent the past two years borrowing heavily to buy stakes in AI companies, chip designers, and data-center operators. Its results are now a test of whether the returns from those bets can arrive before the financing costs do. This quarter, they did not.
The numbers
The headline loss was driven by the data-center writedown, but the damage was broad. SoftBank's Vision Fund segment, which holds stakes in hundreds of technology startups, lost ¥81.6 billion against expectations for a profit. Declines in the share prices of Arm Holdings, Intel, and Coupang also weighed on the portfolio. Financing costs more than doubled to ¥461.9 billion, about $2.9 billion, as the company's debt load grew.
The Vision Fund result is the part of the report that matters most for the AI trade. The fund marks its holdings to market, and the marks move with the same public-market sentiment that has hit technology stocks this fall. SoftBank has been counting on exits and initial public offerings to convert those marks into cash. A quarter in which the fund loses money is a quarter in which the exit window is narrower.
The balance sheet still has room. SoftBank's loan-to-value ratio stood at 15.9% at the end of September, well below the 25% threshold it has set for financial stability, according to Chief Financial Officer Yoshimitsu Goto. But the company's credit default swaps, which measure the cost of insuring its debt, have risen to their highest level in years, a signal that bond investors are watching the AI spending more closely than the equity market has been.
The share move and the bond market tell different stories. The stock rose, while the credit signals worsened. That gap is a reminder that equity and debt investors are pricing different risks: equity holders own the upside if the AI bets pay off, while bondholders care about the $3.7 billion of advance payments and the debt that financed them. Both can be right at the same time.
The Ohio writedown
The writedown centers on SoftBank's plans for a large data-center campus in Pike County, Ohio. The company had reserved manufacturing capacity and paid in advance for power-generation equipment to serve the site, part of an 8-gigawatt buildout. When the global AI outlook shifted, SoftBank revised its procurement strategy, and the non-refundable payments became a loss.
Goto said the company had decided to conduct "a thorough reassessment" of the procurement plan, according to Business Standard's account. He also said SoftBank would watch closely how the midterm elections affect AI sentiment, a reference to the Democratic candidates who ran on curbing data-center growth and won. The company did not say it was abandoning the Ohio project; it said it had revised how it buys the equipment. But the writedown is an admission that the buildout will be slower or more expensive than the company assumed when it placed the orders.
That is the more important signal in the quarter. SoftBank has argued that AI demand will justify almost any amount of capital spending, and it has put its balance sheet behind that view. The Ohio writedown shows that the spending is not frictionless. Power equipment has to be ordered years in advance, construction schedules can slip, and the political environment can change the economics of a site after the money is committed.
The AI portfolio
SoftBank's exposure to AI runs through several channels. It held a 13% stake in OpenAI at the beginning of October and has invested almost $65 billion in the company cumulatively. It also acquired DigitalBridge, a data-center operator and infrastructure investor, extending its reach from model developers to the physical layer of the AI economy. Arm, the chip designer it controls, is another large holding, and its valuation moves with the same AI sentiment.
Arm is the other major variable. SoftBank controls the chip designer, whose architecture underpins most mobile processors and a growing share of data-center chips. Arm's share price fell during the quarter, and its valuation is tied to the same AI capital-spending cycle that SoftBank is financing. When Arm's stock falls, SoftBank's net asset value falls with it, which is why the company's shares are so sensitive to the AI trade.
Those stakes make SoftBank a proxy for the entire trade. When AI valuations rise, its shares rise more. When they fall, the leverage cuts the other way. The past six months have illustrated both directions. Investors have been reassessing the private valuations that underpinned the rally: Anthropic has been discussed at a valuation as low as $1.1 trillion, down from the $2 trillion figure that circulated earlier. OpenAI raised about $11 billion at a $1.25 trillion valuation after initially seeking $30 billion at $1.4 trillion. Earlier reporting traced how quickly those private marks had moved.
Those numbers are still enormous by any historical standard. But they are lower than the marks that were being used a few months ago, and the direction matters for a company that has borrowed against its stakes. SoftBank's thesis depends on the value of its portfolio rising faster than the cost of the debt it used to buy it. A quarter like this one tests that thesis directly.
The political turn
The timing of the writedown is not incidental. The midterm elections gave Democrats control of Congress, and several of the party's candidates campaigned against the data centers that have driven electricity demand and local opposition across the country. SoftBank's Goto acknowledged that the political shift could affect AI sentiment, and the company's revision of its procurement plan came after the election.
The midterms did not create the data-center backlash; they gave it a voice. Local opposition to power demand, water use, and land use has been building for two years, and several states have moved to tighten rules or tax breaks for new campuses. What changed after the election is the national political signal, which matters for companies that plan projects over decades and borrow against them.
The politics could affect the buildout in several ways. New data-center projects may face more state and local opposition, longer permitting processes, and higher power costs if the cost-allocation rules that advanced in the Senate this week become law. A federal permitting bill that speeds approvals would help developers, but it also requires large data centers to pay more of the grid costs they trigger. The net effect is a more expensive and more contested path to the same capacity.
SoftBank is not the only company exposed to that shift. The entire AI supply chain, from chipmakers to utilities to the private-credit funds that have financed data-center construction, is priced on the assumption that demand grows quickly and that the infrastructure gets built. The Ohio writedown is a small data point in that system, but it is the kind of data point that can change how lenders and investors price the next project.
What to watch
The next quarter will show whether the data-center revision is a one-time charge or the start of a slower spending plan. Watch SoftBank's disclosures on the Ohio project and on DigitalBridge's pipeline. Watch the company's loan-to-value ratio, which remains the clearest measure of how much room it has to keep borrowing. And watch the credit markets: if the cost of insuring SoftBank's debt keeps rising, the company may have to sell assets or slow its AI investments, which would ripple through the companies it backs.
The balance-sheet question is not whether SoftBank can survive a bad quarter. It is whether the company can keep funding its commitments without selling assets at a discount. The loan-to-value ratio gives it room, but the ratio uses SoftBank's own marks for private holdings. In a market where private valuations are falling, that cushion can shrink faster than the reported number suggests.
The bigger question is whether the AI trade can absorb a period of higher financing costs and lower valuations without a broader unwind. SoftBank's shares rose after this report, which suggests that investors are not yet treating the loss as a crisis. But the company's bonds and the writedown point the same way: the era of cheap capital for AI infrastructure is being tested, and the companies that borrowed most heavily against future demand will feel the test first.
SoftBank has survived worse. Masayoshi Son has rebuilt the company after the dot-com crash and the WeWork collapse, and he has a long record of raising money when the cycle turns. This quarter does not break the company. It does show that the AI buildout has a cost, and that even its most committed backer is now managing the bill.
Primary sources
- Bloomberg's report, published via Yahoo Finance, for the quarterly loss, the data-center writedown, the Vision Fund result, the OpenAI stake, and the market reaction.
- Business Standard's syndicated Bloomberg report for the Ohio procurement details, the Goto comments, the loan-to-value ratio, and the credit default swap pressure.
- Mavengity's earlier analysis of OpenAI's valuation for the private-market context.