Microsoft has committed over $60 billion to neocloud providers to meet rising AI compute demand

Microsoft has now poured more than $60 billion into a swarm of neocloud infrastructure companies, in what looks like an all-out effort to claw back enough power and chips to meet its overwhelming AI compute demand.

According to a report from Bloomberg, this is a direct response to Microsoft’s own team (and its customers) needing way more capacity than its traditional data centers can handle.

The biggest deal so far is a $23 billion commitment to Nscale, a startup based in the UK. That agreement gives Microsoft control of around 200,000 Nvidia GB300 GPUs, spread across new or existing locations in the United Kingdom, Norway, Portugal, and Texas.

That single deal already puts Microsoft leagues ahead in terms of AI infrastructure control, especially as most rivals are still fumbling around with internal buildouts.

Microsoft signs $10B in fresh neocloud deals with IREN and Lambda

Just this week, Microsoft added two more massive deals to its stack. One is with IREN Ltd., an Australian infrastructure firm that will rent out access to Nvidia GPUs for $9.7 billion over five years.

That agreement includes a 20% prepayment, and the chips will be installed gradually through 2026 at IREN’s 750 megawatt Childress campus in Texas. The hardware will be supplied by Dell Technologies in a $5.8 billion procurement deal.

On the same day, Microsoft also reached a multibillion-dollar agreement with Lambda, though the exact number wasn’t made public.

Many of these neocloud contracts are locked in for five-year terms, giving Microsoft a longer runway to absorb increasing usage from AI services like Azure OpenAI and Copilot.

A Microsoft spokesperson allegedly said the company’s “global infrastructure approach is built on flexibility and optionality, based on the near-term and long-term demand signals we see from customers.” That means they’re now mixing in leased capacity, third-party providers, and owned data centers to stay ahead of compute bottlenecks.

CEO Satya Nadella said last week that the real problem isn’t chips anymore, it’s all about power now. “It’s actually power that’s the bottleneck,” he said on a podcast.

That’s why Microsoft is leaning so hard into partners like Nscale and IREN, who already cleared power permitting and supply chain hurdles. All Microsoft has to do now is flip the switch and plug in its AI stack.

Analysts say OpenAI deal and capex surge show no signs of slowing down

While Amazon hasn’t made any public moves on neoclouds, Google and Meta have quietly been renting capacity from CoreWeave. Reuters reported that Google used CoreWeave to support its OpenAI-related workloads.

Meta also leased compute from CoreWeave in recent months. Still, Microsoft is way out in front when it comes to locking down long-term AI infrastructure.

That momentum was reinforced when Microsoft hit a $4 trillion market cap, boosted by OpenAI’s internal restructuring. Microsoft owns 27% of OpenAI, a stake now worth $135 billion.

That comes alongside a new agreement that pushes $250 billion of spend into Azure over the next several years, with exclusive AI access still intact.

Brad Sills, an analyst at Bank of America, responded to the update by lifting his capital expenditures forecast for Microsoft’s fiscal 2026 from $125 billion to $140 billion, or 43% of revenue.

He said this level of spend should result in 80% more AI capacity year-over-year. “The AI infrastructure story is real,” Brad said in a research note to TheStreet, where he also reiterated a buy rating and confirmed his $640 price target.

On last week’s earnings call, Microsoft told investors it had spent $35 billion in the most recent quarter, most of it on servers, GPUs, and leases. Anurag Rana, an analyst at Bloomberg Intelligence, wrote that the new wave of deals “reinforces our view of a severe capacity crunch across the industry, driven by surging demand for AI workloads.”

Microsoft’s partnerships with these smaller, hyper-scaled providers are clearly designed to avoid delays from power shortages, zoning delays, and chip scarcity. And for now, no one else is doing it at this size.

Sharpen your strategy with mentorship + daily ideas - 30 days free access to our trading program