Investment in power equipment for AI data centers is showing new signs of divergence. SoftBank Group disclosed on November 10th that it recognized a loss of 597.2 billion yen (approximately $3.7 billion) due to adjustments in its US AI data center power equipment procurement plan. Of this, $2.85 billion came from non-refundable prepayments, and another $880 million involved contract modifications and settlements. Reuters reported on November 11th that SoftBank had previously reserved over 13GW of gas turbine capacity, exceeding the demand of some existing projects, and is currently reducing excess orders.
Meanwhile, Indian steam turbine manufacturer Triveni Turbine announced its second-quarter results for fiscal year 2027, showing a 3% year-on-year decline in revenue, a 35.5% decrease in EBITDA, and a drop in profit margin from 26.3% to 17.5%. However, new orders for the quarter still reached a record high, increasing by 17.2% year-on-year, and order backlog increased by 10.5%. The company explained that the profit pressure was mainly due to orders placed a year ago, an unfavorable product mix, and the impact of rising raw material and logistics costs.
AI-driven investment in power infrastructure is entering a new phase: developers previously rushed to secure equipment capacity in advance, with some pre-orders potentially exceeding actual project demand; meanwhile, equipment manufacturers, even with ample orders, face pressure regarding contract prices, product structure, and costs. The market's focus is shifting from the year equipment orders are booked to whether these orders can correspond to real projects, be delivered on time, and translate into profits.
