- SoftBank posted a ¥1.12 trillion ($7 billion) net loss for July–September, missing every analyst forecast.
- The loss includes a $3.7 billion charge from revised US AI data-center equipment plans, while financing costs more than doubled.
- SoftBank raised another $5 billion against Arm shares, bringing its Arm-backed loan to $25 billion.
A Quarter to Forget
SoftBank Group (TSE:9984) on Tuesday reported a net loss of approximately ¥1.12 trillion ($7 billion) for the three months through September, a sharp reversal from a ¥2.5 trillion profit a year earlier. The result blew past the most pessimistic analyst estimate—a ¥113 billion loss—according to LSEG (LSE:LSEG)-polled forecasts, underscoring the mounting financial strain of the conglomerate’s aggressive AI infrastructure push.
The loss was driven by a combination of factors: a $3.7 billion equipment-related charge tied to changes in procurement plans for power-generation equipment at US AI data centers, financing costs that more than doubled to ¥462 billion, and weaker investment performance at the Vision Fund. The fund segment posted a quarterly loss of ¥81.58 billion, compared with a ¥2.38 trillion profit a year earlier, as publicly traded holdings like Coupang (NYSE:CPNG) and Grab (NASDAQ:GRAB) slumped.
Revenue, however, continued to grow. Consolidated sales rose 16% to about ¥2.2 trillion, supported by telecommunications and AI computing. But that top-line strength couldn’t offset the investment losses and higher borrowing costs.
Procurement U-Turn Costs Billions
The $3.7 billion charge stems from advance commitments SoftBank made to secure power-generation equipment for its US data-center projects. Facing potential shortages and long delivery times, the company locked in orders early. But when project requirements changed, those commitments became costly. Cumulative payments for equipment supporting the US AI data centers reached $9.48 billion by September’s end.
CFO Yoshimitsu Goto said the data-center projects are expected to expand over the next five to ten years. SoftBank could recover up to $2.1 billion through future equipment rebates, but those are conditional and not an immediate reversal of the loss.
Financing Burden Grows
Financing costs more than doubled to ¥462 billion in the quarter, reflecting the scale of capital required for SoftBank’s AI ambitions. The company raised a record $56.5 billion during the quarter through financing and asset-related transactions, including an $11.87 billion term loan, $10 billion in dollar bonds, and ¥1 trillion in yen bonds.
Additionally, SoftBank borrowed another $5 billion against its shares in Arm Holdings (NASDAQ:ARM), bringing the total Arm-backed margin loan to approximately $25 billion. That increases reliance on a major portfolio holding as collateral—a structure that links the company’s financing capacity directly to Arm’s market value.
Broader AI Pressures
The results come as AI developers face growing scrutiny over the massive power and infrastructure needs of data centers. SoftBank’s Ohio project, linked to a broader US–Japan trade arrangement, includes a planned $4.2 billion investment with AEP Ohio (NASDAQ:AEP) in transmission infrastructure. But federal support doesn’t eliminate local opposition or regulatory hurdles. In October, PJM proposed studying the project’s generation and transmission requirements, a step that doesn’t guarantee interconnection rights.
SoftBank’s strategy is not retreating. On November 9, it announced plans with Grab and Malaysia’s PETROS for a combined $1.85 billion initial investment in Malaysian AI infrastructure, targeting at least five gigawatts over the longer term. The company also highlighted a 20-year OpenAI lease for eight gigawatts of IT capacity at its Ohio project.
What to Watch
The key question is whether SoftBank can contain financing costs and avoid further procurement charges while its AI projects mature. Management’s counterargument is that long-lived infrastructure and contracted demand will justify today’s capital commitments. As a point of balance-sheet context, Bloomberg reported an end-September loan-to-value ratio of 15.9%, below SoftBank’s stated 25% threshold.
But the distinction between valuation gains and cash returns is becoming more critical. A higher valuation for an AI holding can flatter earnings, but operating data centers must eventually generate enough cash to cover construction and financing costs. For now, the market is left weighing SoftBank’s vision against the mounting bill.