Even Microsoft Is Betting on Owning the Metal

This week, Microsoft moved to shrink the number of specialist “neocloud” partners it uses for AI infrastructure, according to people familiar with the matter cited by CNBC. The same reporting, along with a Bloomberg account relayed by Reuters, says Microsoft is winding down an arrangement to rent AI computing from Oracle and wants to prioritize its own data center footprint. That is not a small procurement tweak. It is one of the largest platform companies in cloud choosing owned metal over a long tail of rented capacity — and it is a warning for every operator whose roadmap still depends on someone else’s partner list.

What Microsoft is changing

Microsoft has spent recent years signing deals with half a dozen of these specialist providers, including CoreWeave, Nebius, Chronoscale, IREN, Lambda, and Nscale, as well as a capacity relationship with Oracle. CNBC reports that Microsoft is now in an evaluation stage and has not decided which neoclouds it will stop using. It will keep working with some key suppliers while it builds out its own infrastructure. One of the people familiar with the talks said Microsoft wants to prioritize its own data center footprint.

Two people familiar with the matter told CNBC that Microsoft opted not to renew a deal with Oracle. Reuters, citing Bloomberg, reported that Microsoft had been turning off rented Oracle computing capacity in recent months after declining to renew the contract, and that the hardware no longer met Microsoft’s evolving needs because it used an older generation of Nvidia chips.

The pullback is already visible in individual sites. Microsoft slashed its work with U.K.-based Nscale, walking away from a site in England. Darren Hardman, Microsoft’s CEO for the U.K. and Ireland, told CNBC in October that the decision had to do with Nscale’s ability to get access to electricity. The Financial Times reported, CNBC noted, that Microsoft also decided against renewing a contract with Nscale in Texas.

A Microsoft spokesperson said the company’s infrastructure strategy is “built on flexibility and choice.” “We combine Microsoft-owned datacenters, leased facilities and strategic partnerships with third-party providers to deliver the capacity our customers need, where and when they need it,” the spokesperson said. “We regularly review these investments against customer demand. That includes tracking delivery and adjusting plans as circumstances change.”

That statement leaves room for partners. It does not change the direction of travel: own more of the footprint, depend less on a long roster of rental capacity.

Why this is bigger than one company’s partner list

It is easy to treat this as a Microsoft story, or as a market story about specialist cloud stocks. It is not only that. It is a control-of-the-stack story. When the computing your business runs on is rented from a provider whose largest customer is rewriting the roster, your roadmap is only as stable as that customer’s next review.

A recent industry review counted 323 neoclouds, CNBC reported, and experts have predicted that many of them will not survive. When CoreWeave went public last year, it said 62 percent of its 2024 revenue came from Microsoft. Concentration like that works until the large customer shortens the list. IREN said it handed over data center capacity to Microsoft in August, then Microsoft requested a networking equipment upgrade that pushed the infrastructure later — into January or February instead of this year.

Operators in that position do not get a strategy memo. They get delayed handovers, redesigned networks, and contracts that are not renewed. Teams that spent months documenting runbooks, capacity plans, and recovery against a provider they assumed would still be on the roster then inherit an engineering project they did not staff.

Own the hardware. Own the outcomes.

At InnoScale we made the ownership bet twenty-five years ago. We own our hardware. We run our own datacenters in the United States and in Europe. There is no partner program deciding whether your environment is still on next year’s shortlist. When something breaks, you reach real engineers, any hour — not a ticket queue designed to protect a rental margin. We have been doing this for twenty-five years.

That model does not make every capacity risk disappear. No honest host will claim that. What it does is put the stack you depend on under operators who answer to you, on metal that is not rented from a hyperscaler’s overflow list or gated by a specialist provider whose largest customer is consolidating suppliers. If you are planning for growth, designing for resilience, or simply want a backup that is not another name on someone else’s partner roster, that is the conversation worth having.

Have a plan that you own

CNBC’s reporting and the Bloomberg account of the Oracle wind-down are a useful snapshot of November 2026. The operational question for everyone else is older: who actually owns the machines your business runs on?

Independent cloud is not a slogan. It is the decision to control the stack. Whether you need a parallel environment, dual-region presence across the US and EU, or a host that will still be there when a larger player’s partner list gets shorter, start with a clear inventory of what you cannot afford to lose — and who owns it.

If you want to talk about running workloads on hardware we own, contact the InnoScale team. We will keep the conversation practical. We will not make your roadmap depend on someone else’s capacity review.