On November 6, 2026, CNBC reported that Microsoft was still evaluating some of its relationships with neocloud providers and had not decided which ones to stop using. A neocloud is a specialist cloud provider that rents AI computing capacity to larger customers. The reported Nscale and IREN developments concern separate projects and different kinds of changes.
Microsoft’s spokesperson told CNBC that its infrastructure strategy combines company-owned data centers, leased facilities and partnerships with third-party providers. The spokesperson said Microsoft reviews investments against customer demand and delivery. Bloomberg had reported a day earlier that projects involving Nebius, Nscale and IREN were under scrutiny.
The reported changes concern different projects
| Project | Reported development | Site and delivery context |
| Nscale’s proposed Loughton site in Essex | Microsoft reportedly withdrew as the proposed data center’s anchor customer. | Cloud Computing News and Bloomberg reporting attributed the decision to power-connection delays that pushed expected electricity supply into the 2030s. |
| Nscale’s Texas expansion option | Microsoft reportedly did not exercise an option for about 500 MW of additional capacity. | Cloud Computing News described the option as concerning a separate Nscale-owned site. It said the decision did not affect the existing deployment, described as about 104,000 GPUs at a 240 MW site leased from Ionic Digital. |
| IREN’s Horizon deployment | A Microsoft-requested network redesign and supplier timing affected the delivery schedule. | The four facilities were designed for a combined 200 MW. IREN said the first entered service for Microsoft during the quarter ended September 30, 2026; the other three were in late-stage construction. |
CNBC also reported, citing two unnamed sources, that Microsoft opted not to renew a capacity agreement with Oracle. The report did not give the agreement’s terms or a reason for the decision.
Read each project on its own terms: an expansion option, a power-connection delay and a network redesign describe different changes to AI capacity.
IREN’s agreement and revenue target cover different measures
Cloud Computing News reported that IREN’s Microsoft GPU-cloud agreement was valued at about $9.7 billion over five years, with a 20% customer prepayment. The same outlet reported that IREN had not announced a cancellation and would not face financial penalties for delays arising from Microsoft-requested network changes.
IREN’s separate target of more than $4 billion in annualized run-rate revenue covers its wider AI-cloud operations, not Microsoft revenue alone or revenue already recognized. IREN moved that target to January 2027, with a possible extension to the end of February.
Microsoft’s spending figures include forecasts
Microsoft reported $45.2 billion in capital expenditures and finance leases for the quarter ended September 30, 2026. Cloud Computing News reported that Microsoft expected fiscal second-quarter capital expenditures to exceed $50 billion and projected about $170 billion for calendar 2026. Those last two figures are outlooks, not reported spending already completed.
