Microsoft’s Capital Spending Ran $5 billion Below Plan After Data Center Delay

A delay at an Iren data center in Texas left Microsoft’s third-quarter capital spending about $5 billion below its previous plan, showing how construction and delivery constraints can slow AI infrastructure spending. At the same time, Super Micro Computer’s higher revenue outlook points to strong server sales, while the capacity to use those systems depends on infrastructure being delivered.

The delayed facility is an Iren data center in Childress, Texas. Microsoft had planned to spend $9.7 billion over five years leasing Nvidia GPU-equipped servers from Iren. The delay pushed some of that planned spending out of the quarter, leaving Microsoft about $5 billion below its earlier capital spending plan.

Microsoft’s data center needs also extend beyond that site. In October, the company withdrew from a planned arrangement to rent capacity at a UK data center project with Nscale after delays involving grid access. Microsoft is also renting computing capacity from several cloud providers because its own construction has not expanded quickly enough to meet demand.

Super Micro Computer reported fiscal first-quarter revenue of $19.95 billion, exceeding analysts’ estimate of $16.39 billion. It forecast second-quarter revenue of $19.5 billion to $21.5 billion. The company raised its fiscal 2027 revenue outlook to $72 billion to $82 billion, up from a previous range of $65 billion to $72 billion.

Higher server sales do not show how much computing capacity has been put into service. Contract values and planned megawatts alone cannot establish how much AI compute will come online. Cancellation provisions, contract length and deployment schedules are factors needed to assess what capacity will be delivered and when.

Delivery timing can further complicate comparisons. Some data centers are expected to come online in stages over several months or years, so their full planned capacity may not be ready at once. A chip purchase agreement and a lease for data center space can also cover the same computing capacity; counting both as separate demand would inflate the apparent total.

The developments capture different parts of the AI infrastructure cycle: Microsoft’s spending reflects a delay in facility delivery, while Super Micro Computer’s sales show revenue from servers. The contrast underscores why sales, project commitments and working computing capacity are separate measures.