Industory
SoftBank’s $3.7 Billion Turbine Loss: What the AI Power Bet Reveals
SoftBank Group recorded an approximately $3.7 billion loss from advance payments and contract adjustments for power-generation equipment intended for artificial-intelligence data-centre projects in the United States. The charge was a principal contributor to a fiscal second-quarter net loss of roughly $7 billion—the company’s first quarterly loss in nearly two years. Reuters
Through a related entity, SoftBank had reserved substantial gas-turbine manufacturing capacity at a time of acute scarcity. Reports indicated reservations exceeding 13 gigawatts—more than the output of six Hoover Dams. When project specifications later changed, SoftBank wrote down large non-refundable advances and associated modification costs. Days earlier Elon Musk had publicly claimed SoftBank was defaulting on turbine orders; SoftBank’s chief financial officer categorically rejected the assertion. Wall Street Journal
The episode supplies one of the clearest financial illustrations to date of both the scarcity value of power equipment in the AI build-out and the execution risk of over-committing capital before designs are finalised.
Table of Contents
- How Did the $3.7 Billion Loss Arise?
- What Volume of Turbines and Capacity Had SoftBank Reserved?
- Which Entity and Projects Were Involved?
- How Does This Relate to AI Data-Center Power Demand?
- What Did Management Say About the “Default” Allegations?
- Are There Any Offsetting Recoveries or Future Rebates?
- What Does the Loss Signal for the Broader AI Infrastructure Build-Out?
- How Have Markets and Analysts Reacted?
- Key Takeaways
- Frequently Asked Questions
How Did the $3.7 Billion Loss Arise?
A SoftBank-affiliated vehicle made cumulative advance payments of approximately $9.5 billion for US power-generation equipment by 30 September 2026. When technical requirements and project timelines evolved, a substantial portion of those advances could not be refunded. The recognised loss comprised roughly $2.85 billion of non-refundable advances plus approximately $0.88 billion of contract-modification and settlement costs, totalling near $3.7 billion. Reuters
The charge was recorded in the fiscal second quarter and formed part of the wider ¥1.12 trillion ($7.1 billion) net loss. Bloomberg
What Volume of Turbines and Capacity Had SoftBank Reserved?
Contemporary reporting, led by The Wall Street Journal, indicated that SoftBank had reserved more than 13 gigawatts of gas-turbine capacity between late 2025 and early 2026. That volume equates to the output of more than six Hoover Dams and exceeded the immediate needs of projects then under firm development. The majority of the reservations were placed with GE Vernova. Wall Street Journal
Which Entity and Projects Were Involved?
The procurement vehicle has been identified in reporting as Silver Bands 7 LLC (or a similarly named SoftBank- and Vision Fund II-backed entity). The equipment was destined for multiple US AI data-center developments, including a large-scale project in Ohio linked to broader Japan–US energy and technology arrangements. SoftBank later clarified that the specific $3.7 billion charge was not solely attributable to any single named project. Reuters
How Does This Relate to AI Data-Center Power Demand?
Training and inference clusters for frontier AI models consume extraordinary amounts of electricity. Developers have competed aggressively to secure both grid connections and on-site generation equipment. SoftBank’s large early reservations were an attempt to guarantee supply in a market where turbine lead times had lengthened dramatically and reservation fees had risen. The subsequent write-down demonstrates that final engineering specifications and construction timelines can diverge materially from the assumptions used when capacity was first locked in. Underlying demand for power has not disappeared; the loss reflects a mismatch of scale and timing. Bloomberg
What Did Management Say About the “Default” Allegations?
On 6 November 2026 Elon Musk posted that SoftBank was “defaulting on their turbine orders.” During the subsequent earnings briefing, Chief Financial Officer Yoshimitsu Goto stated that the claim was “simply not correct at all” and that no default had occurred. Management characterised the actions as a deliberate review and resizing of procurement commitments as project requirements evolved, not a failure to honour contractual obligations. Reuters
Are There Any Offsetting Recoveries or Future Rebates?
SoftBank retains contractual rights to conditional rebates of up to $2.1 billion on future equipment purchases over ten years. These potential rebates have not been recognised as assets in the current financial statements because their realisation depends on subsequent order volumes and the satisfaction of specified conditions. Any future recovery would therefore appear in later periods. Reuters
What Does the Loss Signal for the Broader AI Infrastructure Build-Out?
The charge is a concrete data point on execution risk inside the AI power supply chain. Turbine manufacturers have enjoyed strong pricing power; customers that place oversized, non-refundable reservations can face stranded costs when plans change. Other hyperscalers, data-center developers and infrastructure investors are likely to re-examine the cancellation terms, phasing and sizing of their own advance commitments. At the same time, the episode does not imply that the fundamental shortage of power for AI workloads has eased.
How Have Markets and Analysts Reacted?
The earnings miss and the turbine-related charge added to existing investor caution about the pace, cost and return profile of AI infrastructure spending. SoftBank’s shares and those of related AI-exposed names experienced volatility around the release. Analysts have focused on whether the remaining reservations and any future rebates can improve the economics of the US projects, and on the wider implications for power-equipment pricing and availability.
Approximate Composition of the $3.7 Billion Charge
| Component | Approximate Amount | Accounting Treatment |
|---|---|---|
| Non-refundable advance payments | $2.85 billion | Written down |
| Contract modifications and settlements | $0.88 billion | Expensed |
| Total recognised loss | ~$3.7 billion | Included in Q2 results |
| Potential future rebates | Up to $2.1 billion | Contingent, not recognised |
Key Takeaways
- SoftBank recorded a roughly $3.7 billion loss on scaled-back reservations of power-generation equipment for US AI data centers.
- The charge helped drive a quarterly net loss of about $7 billion.
- Reservations reportedly exceeded 13 GW—well above the needs of then-finalised projects.
- Management explicitly denied any default on turbine orders.
- Conditional rebates of up to $2.1 billion on future purchases exist but remain unbooked.
- The episode highlights both the scarcity of gas turbines and the financial risk of over-ordering before specifications are locked.
- Underlying AI-related electricity demand remains robust; the loss is one of timing and scale.
Frequently Asked Questions
Did SoftBank default on its gas-turbine orders?
No. Chief Financial Officer Yoshimitsu Goto stated clearly that no default occurred and that the claim was incorrect.
How much of the advance payments can still be recovered?
SoftBank holds rights to conditional rebates of up to $2.1 billion over ten years; realisation is uncertain and has not been recognised.
Which US projects were linked to the equipment?
Multiple AI data-center developments, including frequent references to a large Ohio site. SoftBank has clarified that the charge is not tied exclusively to any single project.
Does the write-down mean demand for AI data-center power is weakening?
No. It reflects an over-reservation relative to finalised specifications, not a collapse in electricity demand.
Who was the main turbine supplier?
Reporting has identified GE Vernova as the primary counterparty for the bulk of the reservations.
How did Elon Musk’s comment affect the story?
Musk’s public claim of default drew additional attention; SoftBank’s immediate and categorical denial on the earnings call became part of the official record.
What should other data-center developers learn from this?
Advance payments for long-lead power equipment carry material cancellation or resizing risk. Phased commitments, flexible contractual terms and tighter alignment between reservations and final engineering are prudent.
Will the loss affect SoftBank’s ability to fund future AI projects?
The company retains substantial resources and continues to pursue large-scale infrastructure investments; the charge is significant but not described as existential.
Conclusion
SoftBank’s $3.7 billion turbine-related loss is among the most tangible financial consequences yet recorded of the global race to power artificial-intelligence data centers. For finance, energy and technology audiences the episode demonstrates both the extreme scarcity value of gas turbines and the real cost of committing capital at scale before project specifications are final. While management has rejected any suggestion of default and retains the possibility of future rebates, the charge stands as a cautionary marker for every participant in the AI infrastructure build-out. Subsequent reporting on rebate realisation, the status of remaining reservations and the progress of the underlying US projects will determine whether the loss remains an isolated mis-calculation or becomes a broader reference point for execution risk in the sector.