Microsoft trims neocloud partners: ask where your AI runs

The biggest GPU renter in the market is getting picky. Microsoft is shrinking its list of neocloud partners, the GPU-rental companies that filled its AI capacity gap, CNBC reported on November 6. It wants to work with a few key suppliers and build more of its own data centers. If your AI tools run on someone else's rented GPUs, this changes the risk under them.

What actually happened

• The plan. CNBC's sources say Microsoft is in an evaluation stage and hasn't picked which neoclouds to stop using. Its current roster includes CoreWeave, Nebius, IREN, Lambda and Nscale, plus Oracle.
• What's already moved. Bloomberg reported that Microsoft cancelled a roughly $3 billion Nscale deal tied to a UK data center last month. It also declined to expand a Nscale deal in Texas and stepped away from a 1.35-gigawatt letter of intent in West Virginia. CNBC's sources add that Microsoft chose not to renew a deal with Oracle.
• Why. Grid power delays, construction problems and cost. Bloomberg describes cabling issues and a generator fire at IREN's Texas site. CNBC says IREN capacity that was due this year now comes online in January or February. Microsoft has sent staff to neocloud sites to check build specs.
• The market. CNBC says CoreWeave, Nebius and IREN shares are down roughly half over the past month. CNBC also notes Microsoft was 62% of CoreWeave's 2024 revenue.
• Microsoft's line. A spokesperson told CNBC that it reviews these investments against customer demand and adjusts plans as circumstances change. Microsoft didn't confirm cuts on the record. The reporting comes from unnamed people familiar with the plans.

Why it matters for your business

You probably don't rent GPUs from Nebius. But the AI startup behind your support bot, transcription tool or image generator might. Neoclouds sell capacity to Microsoft, OpenAI and a long tail of smaller AI vendors. When their biggest customer pulls back, they have two moves: chase smaller customers on price, or cut back. Both hit those vendors' pricing and uptime.

Microsoft's CFO already said some GPU cloud contracts slipped into the first half of 2027, and that slowed Azure growth. Capacity is getting harder to plan, even for the companies with the most money.

What we'd do this month:

1. Ask each AI vendor where its inference runs. Its own cloud, a hyperscaler or a neocloud. Get it in writing.
2. Ask about failover. If one GPU provider has a bad quarter, can the vendor move your workload?
3. Keep your prompts and data portable. Two providers you can switch between in a day beat one great deal.

Sources: CNBC, Bloomberg via Yahoo Finance.