CNBC reported on 6 November 2026 that Microsoft is evaluating a reduction in the number of neocloud providers it uses for AI capacity, while putting more emphasis on its own datacentre build-out. The company has not decided which providers it may stop using. For Azure customers, the issue is capacity availability, delivery timing and the reliability of third-party GPU supply.
Microsoft has worked with CoreWeave, Nebius, Chronoscale, IREN, Lambda and Nscale, and with Oracle, as demand for AI infrastructure increased. Two people familiar with the matter said Microsoft had opted not to renew a deal with Oracle.
Microsoft said its infrastructure strategy is "built on flexibility and choice" and combines Microsoft-owned datacentres, leased facilities and third-party providers. The company also said it reviews investments against customer demand, including delivery progress and changes in plans.
More emphasis on Microsoft's own fleet
Microsoft's use of neocloud providers increased after the launch of ChatGPT in 2022, when demand for graphics processing units rose sharply. At the time of ChatGPT's release, OpenAI used only Azure. OpenAI's cloud suppliers now include Amazon, Cerebras, CoreWeave, Google and Oracle.
CoreWeave became one of the more visible examples of that shift. When CoreWeave went public, it said 62% of its 2024 revenue came from Microsoft. That concentration shows why changes in Microsoft demand matter to neocloud providers and to customers watching the supply chain behind Azure AI capacity.
Microsoft's own fiscal first-quarter call in October 2026 already showed pressure in capacity planning. Chief financial officer Amy Hood said some cloud contracts involving GPUs were delayed until the first half of 2027, affecting Azure growth. Microsoft reported USD 45.2 billion in capital expenditures and finance leases for the latest quarter.
Hood also said Microsoft expects capital expenditure of more than USD 50 billion in the second quarter and approximately USD 170 billion for the calendar year, partly because third-party capacity deliveries moved later. She said customer demand continues to exceed available supply. Satya Nadella said Microsoft added 24 new datacentres across five continents during the quarter.
Delivery risk moves up the agenda
The neocloud market is also under financial pressure. Shares of CoreWeave, Nebius and IREN were all down by roughly half over the previous month, even after a Friday rally. Rising borrowing costs and debt levels have increased scrutiny of companies that are building expensive AI capacity ahead of long-term demand.
A recent industry review counted 323 neoclouds, and experts expect many not to survive. IREN said it handed over datacentre capacity to Microsoft in August, but Microsoft requested a networking equipment upgrade. IREN said the infrastructure will come online in January or February 2027 rather than in 2026.
IREN co-chief executive Daniel Roberts said there was a "disconnect between what we're seeing in public market and what we're seeing at the customer level" after the company said AI revenue more than doubled from the prior quarter to USD 200 million. He also said capital markets had become more selective, with questions moving from contract size to whether a contract "can be delivered, financed, and run at a profit".
Microsoft has already reduced work with Nscale in the UK. Darren Hardman, Microsoft's CEO for the UK and Ireland, said in October that the decision to walk away from a planned site in England was related to Nscale's access to electricity.
Azure customers should treat capacity as a project risk
For enterprise teams, the practical question is which GPU capacity will be available, under which commercial arrangement and on which date. Microsoft may keep using some neocloud partners, expand its own fleet and use leased sites at the same time. The mix can still change when supplier delivery, networking upgrades, financing or power access change.
That matters for projects built around Azure AI Foundry, Copilot extensions, model hosting and large-scale inference. A workload may be designed correctly and still be affected by capacity allocation. Procurement, architecture and finance teams need to read capacity commitments as operational dependencies, not only as cloud consumption forecasts.
Microsoft has not named the neoclouds it may stop using. Until it does, customers with GPU-heavy workloads should assume that delivery schedules and regional capacity are part of the risk model for AI deployments.
Whether AI capacity stays inside Microsoft's own datacentres, shifts to leased sites or depends on fewer neocloud partners is the kind of planning problem ECS's Infrastructure topic addresses, alongside Azure architecture, hybrid cloud, cost and platform engineering. Sessions and workshops give attendees room to compare Azure designs, capacity assumptions and cost controls with people running these platforms in production. Bring the GPU scheduling, supplier and budget questions that are hard to settle from a roadmap slide alone.
This article was written by the ECS editorial team with the assistance of Amal Agent, run.events' AI assistant, which supported the research, drafting and proofreading.