Microsoft slows neocloud buildout as delays and power limits hit third parties

Microsoft has begun tightening oversight on billions of dollars in contracts with newer data centre firms, walking away from certain facilities and cutting back planned expansions as construction delays and electricity shortages bite.

Over the past two years, large software vendors leased computing capacity outside their own facilities to meet demand for generative artificial intelligence. Upstart operators known as neoclouds promised fast delivery of specialized graphics chips, securing commitments running into tens of billions of dollars across the sector.

Recent filings and disclosures show delivery is falling behind schedule. Microsoft recently passed on an option to double the size of an agreement with Nscale at a site in Texas, and previously exited a project in Britain after grid operators indicated electricity would not arrive until the 2030s. At the same time, operator IREN confirmed that a design alteration requested by Microsoft pushed its annualized revenue targets into early 2027. Microsoft has also declined to renew a rental arrangement with Oracle for older generation hardware as newer chip specifications take priority.

While Microsoft still spent more than 45 billion dollars in capital expenditure during its last reported quarter, that figure trailed earlier expectations by around 5 billion dollars due to third-party delivery holdups. Company leadership is now placing greater emphasis on owned facilities and reviewing where outside capacity offers real value.

What this means for Australian businesses

Hardware availability shapes how software providers price their services. When infrastructure builds slow down or become more expensive, software vendors tend to pass those costs onto business customers through lower usage caps or revised tier pricing.

Australian businesses can prepare by taking several practical steps:

• Check software renewal agreements for automated shifts toward metered billing, especially for services embedding workplace artificial intelligence.
• Audit your organisation's daily usage to see whether staff need premium model tiers for everyday office admin, or if smaller, cheaper models handle the job equally well.
• Avoid locking multi-year contracts onto single, proprietary cloud environments until hosting capacity and inference prices settle down.