SoftBank’s $7 Billion Lesson in the AI Power Rush

Human Verified 0

Share to: , and

Data centre construction site at dusk with power turbines and transformers in the foreground

SoftBank Group has posted a net loss of about $7 billion for the July-to-September quarter, its first quarterly loss in almost two years, after writing off $3.7 billion tied to power equipment it bought early for artificial intelligence data centres in the United States. The Japanese investment group said a subsidiary had paid in advance for turbines and other generation gear, only to revise its plans when the specifications changed. The deposits could not be refunded. SoftBank also said its financing costs more than doubled in the quarter as it borrowed heavily to fund its AI ambitions.

The charge in detail

The loss comes from a unit called Silver Bands 7, which paid $9.48 billion over 18 months to reserve factory capacity and buy power generation equipment for US data centres. Of the $3.7 billion charge, $2.85 billion covers advance payments that cannot be recovered, and $0.88 billion covers contract changes and settlements with suppliers.

SoftBank said the remaining equipment will be used, including at the PORTS-Pike campus in Ohio, where OpenAI has signed a 20-year lease for eight gigawatts of computing capacity. The group added that it could earn rebates of up to $2.1 billion on future equipment purchases over the next ten years, though it has not booked that money as an asset.

Why SoftBank bought early

Chief financial officer Yoshimitsu Goto told reporters in Tokyo that intense competition to secure energy supply and power generation equipment had pushed the group to order before long-term specifications were finalised. Turbines, transformers and other gear can take years to arrive, and queues for the largest machines stretch into the next decade.

Buying early was meant to guarantee supply; instead it locked SoftBank into kit that no longer fits its plans. Goto said the group was still working by trial and error on assets it no longer needs, and he did not rule out further charges of the same kind. The admission is striking from the industry’s most aggressive AI investor.

Glass office towers at dusk under dramatic clouds

A quarter of heavy borrowing

Financing costs more than doubled to 462 billion yen in the quarter, as SoftBank raised $56.5 billion — its highest ever quarterly total — to keep funding its AI bets. Borrowing included an $11.87 billion term loan, a $10 billion dollar bond and a one trillion yen bond.

The group also increased the margin loan on its stake in chipmaker Arm by $5 billion, to roughly $25 billion. SoftBank’s shares had surged 14% on Tuesday before the results, but they have almost halved in value since the start of June.

The borrowing sits alongside Apollo’s backing of SoftBank’s OpenAI bet, with more than $15 billion of debt, and Nvidia’s $3 billion commitment to SB Energy, SoftBank’s data centre development arm. The buildout is being funded on leverage at every level, which is why a $3.7 billion writedown lands as more than an accounting entry.

The OpenAI connection

SoftBank is the largest backer of the AI buildout, with $64.6 billion committed to OpenAI for a 13% stake, and the Ohio campus sits at the heart of the Stargate venture it launched with OpenAI and Oracle. Goto declined to discuss OpenAI’s funding round, reported to value it at $1.25 trillion, but said an announcement was highly likely this quarter.

Our Read

This is the first hard evidence that the AI infrastructure rush can burn money at scale, and it comes from the investor most committed to it. SoftBank did not lose $3.7 billion because demand for AI disappeared; it lost it because the physical buildout is harder than the financial models assumed.

Power, not chips, is the constraint — and power has lead times no amount of capital can shorten. The same shortage explains why Cerebras blamed land and power for delays under its OpenAI deal, and why SpaceX is planning $74 billion of AI campuses. The race is now about electricity as much as silicon.

For SoftBank, the question is whether the loss is a one-off or the first of several. Goto’s refusal to rule out further charges, and his talk of trial and error, suggests the group is still learning what its AI bet actually costs. Investors who cheered the 14% rally before the results may find the next quarters less forgiving.

Why It Matters

It matters because SoftBank is the bellwether for AI investment. When the biggest backer of the buildout takes a $7 billion quarterly loss and warns more charges may follow, every other funder of data centres must ask whether their early orders will age as badly. The financing questions Moody’s raised now have a concrete example.

It also matters for the companies building the campuses. If the constraint is power and sites rather than demand, the winners are the firms that can secure both — and the losers are those that prepaid for gear before the designs were settled. Nvidia’s leadership change this week has unsettled the sector; SoftBank’s loss adds a financial warning to it.

Background

SoftBank reported the results on Tuesday 10 November, swinging from a 2.5 trillion yen profit a year earlier, when rising OpenAI valuations lifted its portfolio. The Vision Fund segment posted a loss of 81.6 billion yen, and the group’s total gain on investments swung to a loss of 133.6 billion yen.

SoftBank’s earnings are volatile by design — but this quarter’s loss came from the buildout itself, not the portfolio. That distinction is why the result is being read as a warning about the AI infrastructure rush, rather than as another swing in the value of its holdings.

Source: Reuters report.

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.