Coding Assistants Edge0 Claims a 145 Billion Parameter Model That Runs in 29GB of Memory on a Laptop, But the Download Is 105GB Business xFusion, the Former Huawei Server Business, Clears ChiNext IPO Review in 174 Days to Raise 8 Billion Yuan News Modine Surges 29 Percent as Samsung Holds Takeover Talks Over the 5.6 Billion Dollar AI Cooling Supplier AI Agents Amazon Drafts a Plan to Index Nearly Every Product on the Internet So AI Shopping Agents Never Leave Its Site Meta Meta Is Testing a Prediction-Market App Called Takes, Where Employees Bet Cred Points Instead of Real Money Claude Anthropic Ships Claude for Manufacturing: 17 Skills, Four Connectors, and a Rule That the AI Never Touches the Machines Business Moonshot AI Dismantles Its Cayman Holding Company and Moves Investor Stakes Onshore Ahead of a Hong Kong IPO News AMD Adds a 288GB MI450X to the Instinct MI400 Line and Claims Up to 30 Percent Higher Inference Than Vera Rubin NVL72 Coding Assistants Edge0 Claims a 145 Billion Parameter Model That Runs in 29GB of Memory on a Laptop, But the Download Is 105GB Business xFusion, the Former Huawei Server Business, Clears ChiNext IPO Review in 174 Days to Raise 8 Billion Yuan News Modine Surges 29 Percent as Samsung Holds Takeover Talks Over the 5.6 Billion Dollar AI Cooling Supplier AI Agents Amazon Drafts a Plan to Index Nearly Every Product on the Internet So AI Shopping Agents Never Leave Its Site Meta Meta Is Testing a Prediction-Market App Called Takes, Where Employees Bet Cred Points Instead of Real Money Claude Anthropic Ships Claude for Manufacturing: 17 Skills, Four Connectors, and a Rule That the AI Never Touches the Machines Business Moonshot AI Dismantles Its Cayman Holding Company and Moves Investor Stakes Onshore Ahead of a Hong Kong IPO News AMD Adds a 288GB MI450X to the Instinct MI400 Line and Claims Up to 30 Percent Higher Inference Than Vera Rubin NVL72

SoftBank Swings to a $7.1 Billion Quarterly Loss on a $3.7 Billion Data Center Writedown, Its First in Nearly Two Years

SoftBank Swings to a $7.1 Billion Quarterly Loss on a $3.7 Billion Data Center Writedown, Its First in Nearly Two Years

SoftBank Group reported a 1.12 trillion yen ($7.1 billion) fiscal second-quarter net loss, its first in nearly two years, after booking a $3.7 billion charge on revised power-equipment procurement for US AI data center projects. The Vision Fund segment lost 81.6 billion yen. SoftBank also confirmed it has completed its $30 billion OpenAI top-up, lifting cumulative investment to roughly $64.6 billion for a stake of about 13 percent.

SoftBank Group on Tuesday reported a fiscal second-quarter net loss of 1.12 trillion yen, roughly $7.1 billion, its first quarterly net loss in nearly two years and a sharp reversal from a profit of about 2.5 trillion yen a year earlier. Analysts had expected a profit of around 129 billion yen. Revenue came in at 2.22 trillion yen, ahead of the 2.02 trillion yen that analysts polled by LSEG had modeled.

The single largest driver was a 597.2 billion yen charge, about $3.7 billion, taken across the six-month period after SoftBank revised its procurement plans for power generation equipment for AI data center projects in the United States, including an ambitious campus in Ohio. The company had rushed to lock in non-refundable advance payments for turbines and manufacturing capacity, and has now begun unwinding that posture. "We have decided to conduct a thorough reassessment" of the procurement strategy, chief financial officer Yoshimitsu Goto told reporters, adding that the company needs to watch closely how the US midterm elections affect AI sentiment.

Declines in the share prices of chip designer Arm, Intel and Korean e-commerce group Coupang dragged on the portfolio. The Vision Fund segment, which manages stakes in hundreds of startups, posted a segment loss of 81.58 billion yen against expectations for a profit, after investment gains collapsed to 85.45 billion yen from 2.81 trillion yen a year earlier. Vision Fund 2 recorded gains of 89.26 billion yen, while Vision Fund 1 slipped to a small loss despite a $1.3 billion gain on its ByteDance position.

The quarter also closed out the most consequential bet on SoftBank's books. The company confirmed it has completed the previously promised $30 billion additional investment in OpenAI through Vision Fund 2, delivered in three $10 billion tranches on April 1, July 1 and October 1, the last funded from senior foreign-currency bonds issued in September. That lifts SoftBank's cumulative investment in OpenAI to roughly $64.6 billion, about 13 percent of the company. As of the end of September the stake was carried at a fair value of $99.6 billion, a cumulative gain of $45 billion, a paper profit that now depends heavily on where AI valuations settle.

Those valuations are exactly what is moving. Bloomberg reported that investors are weighing an Anthropic valuation as low as $1.1 trillion in the wake of the midterms, down from a previously touted $2 trillion, while OpenAI's latest round closed at about $11 billion raised at a $1.25 trillion valuation after the company initially sought $30 billion at $1.4 trillion. SoftBank's own shares have fallen more than 50 percent from their June record, its credit default swaps sit at multi-year highs, and one analyst comparison put the company's loan-to-value ratio at 15.9 percent, still below its 25 percent stability threshold.

Markets took the loss with surprising calm. SoftBank shares closed 13.9 percent higher on Tuesday as Asian technology stocks rallied, and the company separately announced, the day before results, a joint commitment with Grab and Malaysia's PETROS of $1.85 billion in initial capital for AI infrastructure in Malaysia, with a long-term target of at least five gigawatts of compute capacity.

The writedown is a reminder that the AI buildout is no longer only a story about chips and models; it is a story about turbine lead times, land, and the cost of reserving capacity that may never be needed. SoftBank paid real money to guarantee it would not fall behind, then paid again, in accounting terms, for the privilege of slowing down. Few peers have reserved that honestly.

Comments (0)

Log in to join the discussion

Log In

No comments yet