SoftBank records ¥597.2bn in losses after revising AI power-equipment plans

For the six months ended September 30, it recognized ¥597.248bn in losses connected with revised procurement of power-generation equipment for US AI data-center projects. Of that total, ¥457.030bn came from non-refundable advances and ¥140.218bn from contract modifications and settlements. [1]

These are costs of revising procurement, rather than a measure of electricity consumed by AI. The difference matters: companies can lose money securing future capacity before that capacity contributes useful computation. The matched six-month figures also should not be mistaken for SoftBank’s quarterly net result, which contains many other items.

The strategic lesson is about the sequence of commitments. For an infrastructure buyer, securing scarce equipment early can protect a schedule while making later design changes expensive. A sound capacity plan needs explicit milestones, cancellation costs and a way to revise equipment needs before large deposits become irreversible. The filing makes that trade-off unusually concrete.