SoftBank 3 7B Power Equipment Writedown

SoftBank swung to a $7 billion quarterly loss, its first in almost two years. The biggest driver was a $3.7 billion writedown on power equipment for its AI data centers.

The cause was nonrefundable advance payments for equipment, with procurement plans changing afterward. Cumulative payments for power-generation equipment across its US AI projects reached $9.48 billion by the end of September.

In the same week, SpaceX announced more than $74 billion in new AI campuses. One company is writing down the cost of locking in power early. Another is committing tens of billions to do the same.

AI infrastructure is now priced on a forecast nobody can verify: how much electricity AI will actually consume, and when. Chips can be resold. Generators come with deposits and long lead times, and the deposits do not come back.

For anyone building an AI roadmap, the lesson is to separate reversible commitments from irreversible ones. Capacity you can rent or scale down is a different risk from capacity you have prepaid.

The boom may well continue. But the cost of being early is now showing up as a number on a balance sheet.

Which AI commitments in your organization can you reverse, and which are already locked in?