Kivorent News

Wednesday, November 18, 2026

Tech and AINews

SoftBank writes off $3.7 billion of power equipment it prepaid for US AI data centers

SoftBank Group took a ¥597.2 billion charge after changing orders for power generation gear it had paid for in advance, pushing it to a $7.1 billion quarterly loss.

Kiln, covering chips, data centers and power
, 3 min read

$3.7BSoftBank charge on US data center power equipment

What to know

  • SoftBank disclosed a ¥597.2 billion ($3.7 billion) charge on US data center power equipment with its fiscal second-quarter results on November 10.
  • The charge combines ¥457 billion of non-refundable advance payments and ¥140.2 billion of contract changes and settlements, according to Capacity.
  • SoftBank had paid about $9.5 billion in advances for the equipment by the end of September.

SoftBank Group wrote off about $3.7 billion of power generation equipment it had paid for in advance to supply its US AI data centers, after it changed its plans for the gear. The ¥597.2 billion charge was disclosed with the company's fiscal second-quarter results on November 10, according to Capacity, and was the largest single item behind a quarterly net loss of ¥1.12 trillion ($7.1 billion), The Next Web reported.

What SoftBank wrote off

Capacity broke the charge into two parts: ¥457 billion of advance payments that SoftBank cannot get back, and ¥140.2 billion tied to contract changes and settlements. By the end of September, SoftBank's advance payments for the equipment had reached about $9.5 billion, according to Reuters and CNBC figures cited by both outlets.

SoftBank's chief accounting officer, Kazuko Kimiwada, said the company could not recover fees on the orders it cancelled, The Next Web reported. SoftBank can instead claim rebates of up to $2.1 billion on future purchases over ten years.

An executive on the earnings call said SoftBank had made "bridge investments" to secure generation equipment ahead of an expected shortage, according to Capacity.

Turbines reserved, then cancelled

The Next Web, citing The Wall Street Journal, reported that SoftBank reserved more than 13 gigawatts (GW) of gas turbines, mostly from GE Vernova, between late 2025 and early 2026, then cancelled reservations it could not use. GE Vernova told the Journal the changes had no negative or positive impact on it.

Chief financial officer Yoshimitsu Goto said the company needed a thorough reassessment of its spending, according to The Next Web.

SoftBank has not said which data center projects the written-off orders were meant for. One of its main US projects is the PORTS-Pike Technology Campus in Pike County, Ohio, where it has a 20-year lease with OpenAI for 8 GW of IT capacity, according to Capacity. SoftBank says it has no plans to shrink that project.

The rest of the quarter

SoftBank's interest expense for the six months to September more than doubled from a year earlier to ¥676 billion, The Next Web reported, citing Bloomberg. The company raised $56.5 billion in the quarter, including an $11.87 billion term loan and a $10 billion dollar bond. Its cumulative investment in OpenAI stood at $54.6 billion, carried at a fair value of $99.6 billion.

The same week, SoftBank and Grab signed binding agreements with Sarawak's state energy company PETROS covering $1.85 billion of seed capital for two data center projects in Malaysia, $1 billion of it from SoftBank, according to Capacity. No customer, site or completion date has been named.

Why it matters

The AI building boom has made gas turbines and generators some of the scarcest parts of a data center, and companies have been paying years ahead to lock in supply. Anthropic's reported takeover of 1.7 GW of Caterpillar generators through OnePWR (/stories/anthropic-onepwr-gas-generators) is one example; SoftBank's write-off shows the other side of that rush, where a buyer reserves more power gear than its projects can use and pays to unwind it. For utilities and equipment makers, it is a sign that some reserved demand for AI power may not turn into installed capacity. For background on why the industry needs so much power, see /stories/why-data-centers-use-so-much-power-and-water, and for other recent AI power and data center deals, /stories/ppl-blackstone-qts-pennsylvania-power-deal and /stories/openai-project-camellia-rico-lawsuit.

Sources

AI chips and data centersOpenAI

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.