SoftBank Slides to $7 Billion Net Loss on US Data Centre Charges

SoftBank Group posted a net loss of JPY1.1 trillion ($7 billion) for its fiscal Q2, its first quarterly loss in two years, driven mainly by a writedown on US data centre projects.

The Japanese conglomerate booked a JPY597.2 billion charge in the US, which it tied to a revised procurement strategy "given the evolving requirements of its existing and planned projects". The charge covers projects including a development in Ohio. The quarter covered the three months to 30 September 2026.

The net loss figure also absorbed a JPY270.1 billion loss on the value of holdings in Intel, Arm and South Korean e-commerce company Coupang. A year earlier, SoftBank recorded a net profit of JPY2.5 trillion in the same period.

Net sales rose slightly to JPY2.2 trillion, from JPY1.9 trillion a year ago.

What does this change for SoftBank's US AI plans?

The loss lands at an awkward moment. SoftBank Corp CEO Junichi Miyakawa, speaking at an earnings event before the results, said the group was preparing for a possible slowdown in US demand for AI data centres and may look to other markets for growth.

Bloomberg reported Miyakawa said recent developments in the US "are not going to work particularly to our advantage". He added that global demand for AI data centres remains strong and that SoftBank "may launch in other countries first".

The remarks cast doubt on SB Neo, a US neocloud venture SoftBank unveiled in July. SB Neo rents AI chips and cloud services to large US enterprises and hyperscalers for AI model training and inference.

The ownership structure splits between the two listed arms: SoftBank Corp holds 51%, SoftBank Group the remaining 49%. The stated plan was to add capacity in phases and eventually build infrastructure at a 10-gigawatt scale, with SoftBank's Japanese GPU cloud service — in beta since May — driving the US launch.

At the July unveiling, Miyakawa said SoftBank was making steady progress on securing 10 gigawatts of power for strong US demand. Group chairman Masayoshi Son said the whole organisation would work together to deploy world-class AI infrastructure.

Why the caution now?

SoftBank had been preparing its US neocloud operations, Bloomberg reported, but last week's midterm elections in the US and growing opposition to data centres could trigger a pullback. Opposition to mega data centre projects has surfaced in several US jurisdictions, adding political risk to power-hungry AI builds that were previously seen as straightforward growth plays.

The commercial reality sits awkwardly against the venture's ambition. SB Neo was framed as a phased build toward a 10-gigawatt footprint serving enterprise and hyperscaler demand for AI training and inference. Four months on, the group's own CEO is publicly weighing whether other countries should come first, and the procurement strategy behind at least one US project, in Ohio, has already forced a charge close to JPY600 billion.

What has not changed is the top line. Sales grew year on year, and Miyakawa himself said global demand for AI data centres remains strong. The question is geography and timing, not demand itself.

For now, SoftBank must absorb the writedown while deciding whether the US power and political climate still supports the gigawatt-scale buildout it promised in July, or whether the first SB Neo launches happen outside the United States.