SoftBank billion turbine loss: Son's huge bet backfires

The SoftBank billion turbine loss is now clear: SoftBank Group took a $3.7 billion earnings hit after canceling excess gas-turbine reservations for U.S. AI data centers, swinging to a $7.1 billion quarterly net loss as Tokyo shares fell about 11%. Founder Masayoshi Son’s scramble for scarce power capacity overshot what the company could use.

Key Takeaways

• SoftBank reserved well over 13 gigawatts of gas turbines—more than six Hoover Dams’ worth—then canceled excess bookings as further payments loomed.
• The $3.7 billion power-equipment charge helped drive a ¥1.12 trillion ($7.1 billion) fiscal second-quarter net loss, SoftBank’s first in nearly two years.
• Shares fell about 11% after the results, extending a slide of more than 55% since June that erased over $160 billion in market value.
• CFO Yoshimitsu Goto denied Elon Musk’s claim of a “default,” saying SoftBank is balancing “gas and brake” after an all-offense AI push.
• For investors tracking AI infrastructure plays, the episode shows how prepaid capacity bets can crush returns when demand plans shift.

That writedown sits at the center of SoftBank’s unexpected quarterly wipeout and a sharp reaction in Tokyo trading. It also reframes Son’s AI power strategy from cornering scarce turbines to absorbing nonrefundable reservation fees.

Readers following capital allocation and income-style bets in the AI build-out can find related coverage in our Wealth Hacks & Passive Income hub.

What exactly went wrong with SoftBank’s turbine bet?

According to The Wall Street Journal, Son was so caught up in the 2025 AI frenzy that SoftBank reserved a large slice of the U.S. market for gas power-plant turbines—the machines that turn natural gas into electricity for data centers.

Turbines must be ordered years ahead, and manufacturers have struggled to expand output as power-plant demand surged. SoftBank began placing reservations largely with GE Vernova between late 2025 and early 2026, people familiar with the orders told the Journal.

Those bookings carried substantial, typically nonrefundable fees—often tens of millions of dollars per gigawatt. By early 2026 SoftBank had reserved capacity exceeding Connecticut’s power generation and a sizable share of global turbine output. Market leader GE Vernova ships about 20 gigawatts a year.

Even after covering the world’s largest gas-fired plant SoftBank is tied to—and other smaller plants it still plans—the company held far more reservations than it could use. Its earlier plan was to deploy the surplus in future plants. As additional payments approached in recent months, SoftBank canceled the excess and unwound some contracts that were further along.

In Tuesday’s earnings, SoftBank linked the loss to reservation fees and contract modifications for power equipment secured “in anticipation of shortages.” A GE Vernova spokesperson said SoftBank’s moves had “no negative or positive impact” on the supplier.

How big was the SoftBank billion turbine loss for earnings?

SoftBank reported a net loss of ¥1.12 trillion, or about $7.1 billion, for the July–September fiscal second quarter. That compared with a ¥2.5 trillion profit a year earlier and was SoftBank’s first quarterly net loss in nearly two years.

Analysts on average had expected a profit of about ¥129 billion. Roughly $3.7 billion of the hit came from revised procurement plans and nonrefundable advance payments for power-generation equipment tied to U.S. AI data-center projects, including SoftBank’s Ohio campus plans.

Bloomberg reported that SoftBank Group shares dropped about 11% after the company booked the roughly $7 billion loss. Declines in Arm Holdings, Intel, and Coupang also weighed on results, while the Vision Fund segment logged an ¥81.6 billion loss against expectations of a profit.

The gap versus consensus underscores how quickly SoftBank’s AI infrastructure accounting can swing when prepaid power capacity is written down.

Why does Masayoshi Son’s AI power bet still matter to investors?

SoftBank remains a proxy for OpenAI and broader AI sentiment. Son has poured about $65 billion into OpenAI while building a large data-center push and a robotics unit. SoftBank held a roughly 13% OpenAI stake as of early October.

The Ohio project illustrates the scale. A March deal calls for Japan to fund a $33 billion, 9.2-gigawatt gas plant on a former uranium-enrichment site in southern Ohio, with SoftBank as builder and operator. Power is meant for a neighboring 8-gigawatt data center leased to OpenAI—pitched as the world’s largest. Construction is slated to start early next year.

Reservations strengthened SoftBank’s hand in those talks, the Journal reported. Yet prepaid turbine capacity became a liability once plans no longer matched the order book.

Since June, when SoftBank briefly passed Toyota as Japan’s most valuable company, SoftBank shares have fallen more than 55%, wiping out over $160 billion in value. SoftBank recently agreed to pay 9.75% on a bond issue, and the cost of insuring its debt against default has surged. Cash-hungry SB Energy also delayed its planned IPO amid choppy markets.

The Financial Times had already flagged that a tech sell-off was raising risks for SoftBank’s massive AI bet, with Son facing falling values of holdings ahead of earnings. SoftBank later pointed to “evolving requirements” in U.S. projects when explaining the data-center writedown.

Did SoftBank default on the turbine orders?

Elon Musk alleged on social media that SoftBank had “defaulted” on turbine orders. Goto rejected that claim: “This is simply not correct at all. There is no such default.”

Goto also signaled a strategic pullback, saying SoftBank cannot be “all offense” in the current environment and must “balance between gas and brake.”

That message matters for anyone treating SoftBank—or AI infrastructure equities—as a leveraged way to ride the compute boom. Reservation fees that looked like a moat became a multibillion-dollar charge when SoftBank stepped back from excess capacity.

Son was late into generative AI after earlier WeWork wounds and after selling an early Nvidia stake before its biggest run. In early 2025 he invested $30 billion in OpenAI at a $300 billion valuation and pushed Arm deeper into data-center chip design. Rising OpenAI and Arm values emboldened further bets—until market skepticism about AI debt and payback times intensified.

The SoftBank billion turbine loss does not erase SoftBank’s Ohio power and OpenAI data-center ambitions. It does show how over-reserving scarce hardware can turn a scarcity play into a balance-sheet shock when the build-out calendar slips or shrinks.

For wealth-focused readers, the lesson is blunt: prepaid “picks and shovels” exposure can look like passive upside until contract terms make cancellation expensive. SoftBank still aims to supply the Ohio plant and other sites it needs. The cost of the excess is already on the books.