What happens when you order billions of dollars worth of power equipment before the blueprints are even finished? SoftBank just found out the hard way. The Japanese investment giant reported a net loss of about $7.1 billion for the July-September quarter, its first red ink in nearly two years, and the biggest single hit came from a $3.7 billion write-off on gear that was supposed to keep the lights on at its ambitious AI data centers.
SoftBank’s Costly Pivot on AI Power Plans
I’ve been watching SoftBank’s moves in the AI space for a while, and this quarter feels like a moment of truth. The company, led by Masayoshi Son, has poured enormous resources into OpenAI and the surrounding infrastructure. Yet the latest numbers show how quickly plans can unravel when competition for energy supply turns fierce and project requirements shift mid-stream.
Through a subsidiary known as SB7, jointly owned with one of the Vision Funds, SoftBank had already paid out roughly $9.5 billion in advance for manufacturing slots and power generation equipment. Much of it was earmarked for a huge campus in Ohio and other sites. Then the strategy changed. Project needs evolved, and suddenly a large chunk of those non-refundable deposits had to be written off.
The breakdown is stark. About $2.85 billion in advance payments simply disappeared because the contracts offered no refunds. Another $880 million covers contract modifications and settlements that will start hitting the cash flow this quarter. In total, nearly 30 percent of every dollar prepaid for this equipment has effectively vanished. What remains, around $6.6 billion, has been reassigned to the Ohio site and a few other projects that are now on hold pending further review and a possible IPO that may never materialize.
There is talk of potential rebates totaling up to $2.1 billion on future purchases over the next decade, but SoftBank’s own accountants declined to book any of that as an asset. The message is clear: today’s loss is certain, while tomorrow’s coupon remains highly conditional.
Why SoftBank Rushed the Orders
CFO Yoshimitsu Goto explained that intense competition for energy equipment forced the company to place orders before long-term specifications were locked in. The Ohio project itself is not being scaled back, he insisted, but a thorough reassessment is now underway. He would not name the vendors involved or detail exactly what changed in the requirements. More importantly, he could not guarantee that similar write-downs would not appear in future quarters.
This approach of securing capacity first and refining the plan later is not unique to SoftBank. Across the industry, the race to build AI infrastructure has created bottlenecks in turbines, transformers, and specialized power systems. Delivery timelines already stretch into 2030 in many cases. When a major player walks away from reserved slots, those openings can quickly be claimed by competitors with stronger balance sheets or clearer project pipelines.
In my view, the real lesson here is about the cost of speed. SoftBank was trying to move faster than the market, funding turbine reservations with short-term bridge financing. Any change of course then becomes extremely expensive. The irony is that the underlying concept—building behind-the-meter gas generation on site at a former industrial facility—makes sense for large data centers that cannot rely solely on the existing grid. The balance sheet simply could not absorb the flexibility required.
The Interest Bill Doubles and Debt Climbs
The equipment write-off grabbed the headlines, but the surge in financing costs may prove more persistent. Consolidated finance costs for the six months to September reached ¥791 billion, more than double the ¥340 billion recorded a year earlier. At the parent level alone, interest expense jumped to ¥676 billion from ¥269 billion.
Part of that spike came from accelerating the amortization of costs tied to large bridge loans that were repaid early. SoftBank had lined up a $40 billion facility earlier in the year to support its latest OpenAI commitment. It drew $30 billion and then repaid the entire amount by mid-September, six months ahead of schedule. The early repayment looked positive to lenders, yet the company still had to refinance the exposure.
The replacement package included a new two-year term loan of nearly $12 billion, a $7.35 billion prepaid forward structure on its Intel shares, an expansion of the margin loan backed by Arm holdings (now standing around $25 billion), and additional bond issuance. Over six months the group raised roughly ¥17.2 trillion and repaid ¥9.7 trillion. Net interest-bearing debt rose by ¥7.9 trillion since March to ¥32.6 trillion. The equity ratio slipped from 29 percent to 23.6 percent.
Operating cash flow stayed negative at ¥527 billion for the half, while investing cash flow was a deep negative ¥5.4 trillion. Financing inflows covered the gap with room to spare, but the pattern is clear: SoftBank is still relying heavily on new capital to fund both its investments and its interest payments.
The Intel Hedge That Worked Against Them
One of the more surprising elements in the numbers involves SoftBank’s Intel position. The company bought 87 million shares for $2 billion last year. By the end of March the stake had risen to $3.8 billion, and by September it reached $10.5 billion, generating a large paper gain in the first half.
Instead of realizing that gain, SoftBank entered into prepaid forward contracts structured as collars. It raised $7.35 billion against the entire holding, locking in a floor while capping the upside. When Intel shares continued to climb after the contracts were signed, the derivatives moved against SoftBank, producing a ¥322 billion mark-to-market loss in a short period. At the same time, because the shares finished the quarter lower than they started, SoftBank also booked a separate valuation loss on the equity itself.
The result? Losses on both the stock and the hedge in the same quarter. Of the pledged shares valued at ¥1.65 trillion, banks hold rights of use over ¥1.1 trillion, meaning they can lend or otherwise deploy the collateral. The Intel proceeds were already allocated to other financing needs, so the economic exposure remains complex.
OpenAI Exposure and Circular Structures
SoftBank’s cumulative investment in OpenAI now stands at roughly $64.6 billion. The stake is carried at $99.6 billion against a cost of $54.6 billion as of the end of September, a mark that has not moved since March. Management expects the latest fundraising round to lift the net asset value in the coming quarter, though OpenAI itself has adjusted its valuation target downward and raised less capital than initially sought.
A subsidiary of Vision Fund 2 borrowed the full $10 billion against OpenAI preferred shares. SoftBank Group guarantees the loan, and the facility includes cash collateral and mandatory prepayment triggers if the fair value of those shares declines significantly. That clause sits on the balance sheet at a time when private market valuations in the AI sector face greater scrutiny.
Perhaps the most unusual structure involves warrants granted by SoftBank’s energy subsidiary to OpenAI as a lease incentive. The landlord effectively paid the tenant in equity to sign a data center lease. Some warrants were later forfeited when a longer Ohio lease was signed, yet ¥542 billion still sits as a derivative liability. Because the warrants increase in value as the energy business is marked higher, SoftBank books losses when the underlying project succeeds. Additional share-based compensation for the unit’s employees compounds the effect.
Other circular arrangements appear as well. One SoftBank entity sold its stake in the energy business to a Vision Fund vehicle and booked a gain that disappears on consolidation. The Vision Funds themselves marked down holdings in SoftBank subsidiaries, producing paper losses that also net out at the group level. These moves improve the optics at the individual company level while leaving the consolidated picture largely unchanged.
Liquidity Cushion and Future Funding Needs
On paper, SoftBank ended September with about $32 billion in liquidity. After subtracting near-term bond maturities and the final $10 billion OpenAI tranche paid in early October, the excess cushion shrinks to roughly $10 billion. Annual interest expense is now running above ¥1 trillion. Senior loan repayments totaling ¥1.4 trillion come due in the fiscal year ending March 2028, on top of ongoing data center spending that already reached ¥1.7 trillion in the first half.
The famous loan-to-value ratio stood at 15.9 percent at the end of September, well below the internal 25 percent ceiling. Management has since indicated that net asset value has fallen from ¥59.8 trillion to around ¥44.7 trillion. Holding net debt constant, that implies an LTV closer to 20 percent today—and the OpenAI mark has not yet been updated for any new fundraising.
Credit markets have taken notice. SoftBank credit default swaps have pushed to multi-year highs, reflecting investor concern about the combination of rising interest costs, large refinancing needs, and continued capital commitments to high-risk, high-reward projects.
Balancing Acceleration and Caution
Goto’s message on the earnings call was that SoftBank can no longer run purely on offense. The company must now press the accelerator and the brake at the same time to keep leverage within targets. For an organization long associated with bold, concentrated bets, that shift in tone is significant.
The timing of the procurement review coincides with broader headwinds. Political changes in certain U.S. states could slow data center approvals. International deal complications have already appeared. Other large AI-related projects across the industry have faced delays or scaled-back ambitions. Against that backdrop, questions remain about how much of the remaining prepaid power equipment will ultimately be put to use.
Bulls will argue that this was a kitchen-sink quarter. The bridge facility is gone, the equipment deposits are written off, and a higher OpenAI valuation could boost net asset value next period. That optimistic reading is possible. The more cautious view focuses on the structural realities: ¥32.6 trillion of interest-bearing debt, negative operating cash flow, a guaranteed loan against private shares, and a CFO who cannot rule out further impairments.
SoftBank is likely to return to the capital markets in 2027. The price it pays will depend on how successfully it manages the current portfolio of investments and whether the AI infrastructure buildout begins to generate more predictable cash flows. The $3.7 billion equipment charge was the cost of changing the plan. The elevated interest expense is the ongoing cost of having pursued that plan at such scale and speed.
Key Takeaways from the Quarter
- SoftBank recorded a ¥1.12 trillion net loss, missing expectations of a modest profit
- A ¥597 billion impairment on power generation equipment for AI data centers was the largest single item
- Financing costs more than doubled year over year as the company refinanced large bridge facilities
- The Intel collar structure produced derivative losses even as the underlying shares had previously generated substantial gains
- Liquidity looks solid on the surface but is tighter after accounting for near-term maturities and remaining OpenAI commitments
- Management now emphasizes the need to balance growth investments with tighter control of leverage
The broader AI infrastructure boom continues, yet SoftBank’s experience illustrates the financial risks that accompany aggressive capacity reservation in a supply-constrained market. Turbine slots and specialized power equipment remain scarce. Companies that can fund long-term projects with more stable capital structures may ultimately capture the benefits of the buildout that SoftBank helped accelerate.
For investors following the intersection of private technology investment and public market leverage, the next several quarters will be telling. SoftBank has demonstrated an ability to raise capital repeatedly and at scale. Whether that capital continues to generate the returns needed to service a growing interest burden remains the central open question.
The numbers released this week do not end the SoftBank AI story. They do, however, mark a clear shift from pure expansion toward a more careful balancing act between ambition and financial reality. That adjustment was overdue. How effectively it is executed will determine whether the current losses become a temporary setback or a longer-term constraint on the group’s strategy.
