SoftBank Group (SBG) announced on November 10, 2026, that its consolidated results for April–September 2026 showed a net loss attributable to owners of the parent of ¥768.0 billion, a swing from a profit of ¥2,924.1 billion a year earlier. The company booked a ¥597.2 billion loss after revisiting its plan to procure power generation equipment for AI data centers in the United States. For a company pouring huge sums into AI infrastructure, securing power supply equipment ahead of time has become a major financial burden.
The loss does not mean that all the money spent on procuring power equipment has been lost. SBG has paid roughly $9.5 billion in advance in total, of which about $6.6 billion is still planned to go toward acquiring equipment; the remaining roughly $2.9 billion became a loss. A further loss of about $0.9 billion arose from contract changes and similar items, while SBG also secured discounts of up to $2.1 billion usable on future equipment purchases. The results illustrate the risks that a strategy of locking in equipment manufacturing slots early, in order to speed up AI data center construction, can carry when project requirements change.
¥1,500.9 billion in advance procurement, and why about ¥597.2 billion became a loss
The procurement was carried out by Silver Bands 7 LLC (SB7), a subsidiary funded by SBG and SoftBank Vision Fund 2 (SVF2). To supply power to AI data centers in the U.S., including the "PORTS-Pike Technology Campus" in Pike County, Ohio, the company had entered into contracts to secure manufacturing capacity for power generation equipment, as well as equipment purchase agreements. According to SBG's financial results summary for the second quarter of the fiscal year ending March 2027, cumulative payments from fiscal 2025 through the end of September 2026 reached ¥1,500.886 billion (about $9.48 billion).
Behind this advance procurement is a tight supply of power generation equipment amid the surge in AI data centers. Large power generation equipment takes a long time to go from order to delivery, so even if an order is placed after a data center plan is finalized, the equipment may not arrive when it is needed. SBG therefore adopted a strategy of securing manufacturing slots and equipment before individual projects were finalized, then allocating them as each project took shape. The aim was to reduce, through front-loaded spending, the risk that equipment shortages would delay data center operations.
The tight supply is also visible in the results of power equipment makers. According to GE Vernova's third-quarter 2026 results, announced on October 28, its combined backlog and slot reservations for gas power equipment reached 122 GW, up from 116 GW in the previous quarter. The company expects to expand this to at least 130 GW by the end of 2026. The figures show strong demand for power equipment manufacturing capacity, consistent with the market conditions that led SBG to prioritize early procurement. However, GE Vernova's overall backlog cannot be equated with SB7's procurement contracts.
The problem was that the data center side's plans changed after the equipment had been secured. SBG explained that requirements changed for existing and planned projects, prompting it to revisit its equipment procurement strategy. If the specifications, quantity, or timing of required equipment differ from original assumptions, manufacturing slots and purchase contracts already secured cannot necessarily be used as they are. The disclosure does not reveal which equipment or which requirements changed, but the change to the procurement plan led to losses on non-refundable advance payments and contractual obligations.
The investor presentation published on November 10 (accounting section, page 8) explains in more detail how the cumulative payments of about $9.5 billion were treated. About $6.6 billion will go toward acquiring equipment related to the PORTS-Pike power generation project and others, while a loss was recognized on about $2.9 billion because there is no longer any prospect of acquiring the equipment. Adding about $0.9 billion in losses from contract changes and settlements gives the total loss of about $3.7 billion, or ¥597.248 billion in yen terms.
| Breakdown of procurement loss | Amount (¥ million) | In ¥100 million | Share |
|---|---|---|---|
| Loss on non-refundable advance payments | 457,030 | ¥457.0 billion | 76.5% |
| Loss from contract changes and settlements | 140,218 | ¥140.2 billion | 23.5% |
| Total | 597,248 | ¥597.2 billion | 100.0% |
Shares were calculated from the figures in millions of yen given in the results summary and rounded to one decimal place. The ¥457.0 billion loss on non-refundable advance payments accounts for about three-quarters of the total procurement-related loss. What matters is that this is not simply a decline in the market value of power equipment; it reflects the fact that there is no longer any prospect of acquiring equipment corresponding to funds already paid. The way the loss arose differs from a case in which a company holds physical equipment whose value has fallen.
It is also essential to distinguish between cash movements and accounting losses. The roughly $2.9 billion in advance payments had already been made, so no new cash of the same amount left the company when the loss was recognized. By contrast, SBG says it plans to pay the roughly $0.9 billion in costs from contract changes and settlements from the third quarter of fiscal 2026 onward. In other words, the ¥597.2 billion includes both a portion recognizing past payments as a loss and a portion for which cash outlays are yet to come.
The PORTS-Pike plan itself has not been canceled
A key point in understanding this loss is to distinguish the procurement plan revision from the data center development plan itself. In its investor presentation, SBG explicitly states that the roughly $3.7 billion loss is "unrelated to PORTS-Pike." This separates the roughly $6.6 billion in procurement funds allocated to PORTS-Pike and other projects from the loss on equipment that is no longer expected to be acquired. The ¥597.2 billion cannot be interpreted as a cancellation loss on the PORTS-Pike project.
The PORTS-Pike Technology Campus is a large AI infrastructure site planned in Ohio, involving SBG, SB Energy, OpenAI, NVIDIA, and others. According to the partnership announced by NVIDIA on August 17, SB Energy will build, own, and operate the data center, and OpenAI will use it under a 20-year lease. NVIDIA will be the exclusive supplier of AI computing infrastructure and will also provide credit support related to land, power, and building preparation. The plan initially secures 4.25 GW of IT capacity, with an option to add 3.75 GW.
The project's official website describes an initial phase of a data center of about 800 MW, coming online in stages from 2028. It also outlines a plan to build a total of 10 GW of new generation capacity to meet demand at the site and in the surrounding area, of which 9.2 GW is natural gas-fired power. This 9.2 GW buildout is also covered by a strategic investment initiative based on an agreement between the Japanese and U.S. governments.
It is worth noting that GW, which indicates the output of power generation equipment, and IT-GW, which indicates the power consumption of IT equipment such as AI servers, are different measures. A data center as a whole consumes power not only for IT equipment but also for cooling, power conversion, and distribution, and stable operation also requires grid capacity and supply reliability. When securing generation equipment, therefore, not just raw output but also equipment configuration, timing of introduction, and compatibility with power supply infrastructure matter.
However, SBG's disclosure alone does not identify what specific technical requirement changes occurred. Nor have the specific models, delivery schedules by equipment, or final placement of the roughly $6.6 billion worth of equipment already secured been fully disclosed. That PORTS-Pike construction plans continue is a separate matter from whether power equipment will be delivered and brought into operation as scheduled. An important thing to watch going forward is whether the procured equipment is actually allocated to real projects and leads to the buildout of generation capacity.
How much can the up-to-$2.1 billion purchase discount offset the loss?
As part of the revision to the procurement contracts, SB7 also obtained a new right from the supplier: the ability to receive discounts of up to $2.1 billion on future equipment purchases over the next 10 years, provided certain conditions are met. This can be seen as a mechanism for partly easing the burden caused by the procurement plan change through improved future purchase terms. However, the amount is not fixed, and as of the end of September 2026 SBG has not recognized this right as an asset.
The results summary states that the inability to determine the discount amount is the reason for not recognizing an asset. The investor presentation likewise cites the uncertainty of the timing and amount of future equipment purchases. Applying the discount requires actual equipment purchases and the satisfaction of specified conditions, and there is no guarantee that the full $2.1 billion can be used. If purchases are smaller than expected or the conditions are not met, the usable discount could fall below the cap.
Therefore, it is not appropriate to simply subtract $2.1 billion from the roughly $3.7 billion procurement loss and calculate an effective loss of about $1.6 billion. The former is a loss recognized in April–September 2026, while the latter is a conditional right that may reduce future equipment purchase costs. The two differ in both the timing of accounting recognition and the conditions required for realization. How much economic value the discount ultimately has will become clear through future equipment purchase contracts and their performance.
This procurement review highlights both the advantages and disadvantages of a strategy of securing equipment early. In periods of persistent supply shortages, locking in manufacturing slots may help prevent project delays. On the other hand, if data center designs or construction schedules change, non-refundable advance payments and contract change costs can become a heavy burden. The discount secured this time leaves room to lower future procurement costs, but it does not reverse the loss already booked.
¥1,725.8 billion in investment gains, yet a net loss: the financial burden of AI infrastructure grows
In SBG's April–September 2026 results, net income came in at a loss of ¥768.0 billion even though investment gains reached ¥1,725.8 billion. The investment gains themselves fell 56.0% from ¥3,926.7 billion a year earlier, but selling, general and administrative expenses, financial costs, and derivative-related losses weighed on results even more. The power equipment procurement loss is a particularly large item among these rising costs.
| Main income items | April–Sept. 2026 | Year earlier |
|---|---|---|
| Net sales | ¥4,235.0 billion | ¥3,736.8 billion |
| Investment gains | ¥1,725.8 billion | ¥3,926.7 billion |
| SG&A expenses | ¥2,368.8 billion | ¥1,616.9 billion |
| Financial costs | ¥790.7 billion | ¥340.0 billion |
| Derivative-related gains/losses | ▲¥433.3 billion | ¥208.1 billion |
| Other gains/losses | ▲¥663.8 billion | ¥145.8 billion |
| Net income attributable to owners of the parent | ▲¥768.0 billion | ¥2,924.1 billion |
Amounts rounded to the nearest ¥100 million. Expense items show costs as positive figures; ▲ in income items indicates a loss.
In particular, financial costs more than doubled, from about ¥340.0 billion a year earlier to about ¥790.7 billion. According to SBG, factors included growth in outstanding borrowings and bonds, higher funding rates, and the one-time write-off of financing costs from the early repayment of a bridge loan. In March 2026, it signed a $40 billion bridge facility agreement mainly to fund additional investment in OpenAI, borrowed $30 billion of it, and repaid the full amount early by September 15. To secure the repayment funds, it combined long-term borrowing, bond issuance, and financing backed by its shareholdings.
Separately from the power equipment procurement loss, there was also an impairment related to a manufacturing site in Lordstown, Ohio. For buildings and other assets acquired to establish a manufacturing base for data center-related equipment, SBG reviewed the recoverable amount in light of future use policies and plans and booked an impairment loss of ¥72.543 billion. The recoverable amount is calculated based on fair value less costs of disposal. This loss was recorded separately from the ¥597.2 billion tied to the power equipment procurement review; combined, the two total about ¥669.8 billion. However, since the assets involved and the reasons for recognizing the losses differ, they should not be treated as a single contract loss.
Meanwhile, for OpenAI, the center of its AI investments, the fair value as of the end of September 2026 was $99.6 billion, cumulative investment including indirect investment was $54.6 billion, and cumulative investment gains were $45.0 billion. SBG says OpenAI's valuation is unchanged from the end of the previous period, and no investment gain from revaluing OpenAI shares was recorded in April–September. The net loss therefore cannot be explained as the result of a decline in OpenAI's corporate value. Investment gains include valuation gains on Intel shares and other items, so changes in the value of held assets need to be considered separately from costs associated with the infrastructure business.
Funding to OpenAI itself continues. On October 1, 2026, SBG invested an additional $10 billion, completing the $30 billion additional investment it committed to in February. This brought cumulative investment to $64.6 billion and its ownership stake to about 13%, but the October investment falls outside the April–September reporting period. As the burden of capital spending comes into view while investment in AI-related companies continues, attention will focus not only on the valuation of held assets but also on funding and actual cash recovery.
The next question: how much of the pre-procured equipment can be put to use
What the results make clear is that rising demand for AI infrastructure does not by itself guarantee recovery of capital spending. Securing power equipment ahead of time in anticipation of shortages can be a rational choice for getting projects up and running quickly. But if the conditions for introducing the equipment change, manufacturing slots and contracts already secured can turn into a burden. The ¥597.2 billion loss SBG recognized shows that risk of advance procurement actually materializing.
The focus going forward is whether the roughly $6.6 billion earmarked for projects such as PORTS-Pike actually translates into acquired equipment. It is also important when and how much of the conditionally secured purchase discount of up to $2.1 billion can be used. Once equipment delivery timing, the start of power plant operations, and any additional contract change costs become clear, the post-revision plan can be assessed more concretely.
In its disclosure on business risks as of June 2026, SBG explained that the AI infrastructure business requires large amounts of capital and that, if external funding cannot be secured at the planned timing, scale, and terms, plans may not proceed as intended. The power equipment procurement review showed, in addition to this funding risk, the difficulty of matching equipment procurement terms to the requirements of individual projects. Going forward, not just the scale of investment and planned generation capacity but the track record from equipment acquisition through operation to cash recovery will be an important basis for evaluating SBG's AI infrastructure strategy.