The equipment was ordered for a power plant whose design was still moving. On Nov. 10, SoftBank Group disclosed that reworking how it buys generation equipment for its American AI data centers had left it with a 597.2 billion yen charge, about $3.7 billion, and that its fiscal second quarter closed with a net loss of 1.12 trillion yen.
The reaction outside SoftBank landed on GE Vernova, the equipment supplier, whose shares fell almost 10% the same day. SoftBank said the Ohio project must be re-evaluated in full.
A charge of that size belongs to the company that books it. The share move belongs to the company that still has to ship against a schedule nobody can now describe.
Power for data centers is bought years before it is needed. Grid connections, manufacturing slots for heavy turbines and long-lead transformers are the items that decide when a campus can switch on, and developers have learned to reserve them before the building design is frozen. Reservations are paid in advance. An early order is therefore a balance-sheet commitment as much as a procurement decision, and the two only come apart when the plan changes.
That is where the loss came from. When a design shifts after the purchase order is signed, prepaid equipment stops matching the project it was bought for. Accounting handles the gap as an impairment: the asset sits on the books at a value the plan no longer supports until someone writes it down. SoftBank wrote it down, and the size of the number says how much of the Ohio plan had been committed in hardware before the plan was settled.
For GE Vernova the arithmetic is different and less clean. The order does not necessarily disappear. What changes is the timing and the configuration, and with them the confidence an investor can place in the backlog that has been carrying the stock.
Backlogs have become the way this market values the electrification of computing. A supplier’s order book is read as evidence that demand is contracted rather than projected, and the multiple attached to that evidence is generous because the buyers are the largest companies in the world and the equipment takes years to build. A customer that reopens its design forces a discount on the proof, because the contract now carries a condition that nobody outside the negotiation can see.
The supply side is thin. Only a handful of companies build the large turbines and the grid equipment a campus of this kind requires, and their lead times are measured in years rather than quarters. Being one of the few sellers confers pricing power. It also makes each of them a read-through instrument. When a single developer re-scopes a project, investors treat it as evidence about everybody else’s queue, whether or not the other buyers have changed anything.
Analysts spent the day on exactly that question. A write-down is a known number, disclosed and closed. A schedule change is an unknown one that propagates through revenue recognition at the vendor, through the developer’s cost per megawatt and through the utilities that were told to plan around a load that may now arrive later. Known losses get priced once. Unknown ones get priced until the next disclosure.
The quarter SoftBank reported was larger than the charge. The 1.12 trillion yen loss between July and September is roughly twice the impairment, which means the re-evaluation of the Ohio equipment accounts for about half of what happened to earnings in the period and leaves the rest to be explained elsewhere in the group.
What the disclosure establishes is narrower than the headline. SoftBank has not abandoned the project. It has said the project must be re-evaluated, which is the language companies use when scope, timetable or cost has to be renegotiated with parties who are not in the room: the contractor, the utility, the state, the supplier holding the order.
For the vendor, that language is the risk. Equipment makers book orders when they are signed and adjust when they are changed, and the adjustment runs through revenue timing rather than through a single dramatic line. A project that slips a year does not appear as a loss at the supplier. It appears as a flat quarter, then as another flat quarter, and then as a guidance cut that arrives without an obvious cause.
There is a second-order effect on price. A supplier that knows its customer’s design is unsettled can reprice the next order, add cancellation terms, or require firmer commitments before it reserves a slot. Every developer in the queue pays for the discipline that one renegotiation introduces, which is how a single project’s rethink becomes an industry-wide cost of capital.
The pattern is not unique to Ohio. Reserving power before the engineering is final has become standard practice because the alternative is waiting years for a connection that may not be granted. That trade is rational for each developer and expensive for the group, since a queue full of speculative reservations is a queue that misleads everyone building behind it.
GE Vernova’s nearly 10% decline is a judgment about how much of its order book rests on projects still in that condition. The company will be asked to characterize it when it next reports, and the answer will matter beyond its own shareholders, because the same question applies to every supplier selling into the AI build-out.
SoftBank, for its part, has taken the charge and closed the quarter. The Ohio project now has to be rebuilt on paper before it is rebuilt on the ground.
The charge is booked. The schedule is not, and the equipment already paid for is sitting somewhere waiting to learn what it is for.
