Microsoft exits Oracle, Nscale, IREN neocloud deals for in-house data centers

Microsoft's move away from neocloud capacity sits alongside a much wider software and cloud services footprint: the company leans on its US$4.0 trillion scale and global reach in productivity tools, business applications and developer platforms to influence where enterprise AI workloads physically run. The pullback reduces exposure to third-party operators whose delays, power constraints and older chips can make AI capex less predictable.

The Narrative investors have been using already centers on one requirement: a huge AI data center build and US$175b to US$190b of 2026 capex must translate into durable revenue and cash flow. Tightening control over where that infrastructure sits, and reducing exposure to neocloud cost overruns or regulatory snags, aligns more with that Narrative's execution-risk side than with its demand story.

The next test is financial rather than operational. As neocloud contracts roll off, the question is whether net debt and free cash flow hold steady relative to AI infrastructure spend — a sign, if they do, that the pivot toward owned capacity is preserving Microsoft's financial flexibility rather than stretching the balance sheet.