Microsoft signs $9.7 billion AI cloud deal with IREN

(Bloomberg image)

Microsoft Corp. has signed an approximately $9.7 billion deal to purchase AI cloud capacity from IREN Ltd., becoming the Australian company’s largest customer.

The five-year agreement will provide Microsoft access to Nvidia Corp. accelerator systems in Texas built using the GB300 architecture for AI workloads and includes a 20% prepayment, IREN said in a statement on Monday. Sydney-based IREN also said it’s agreed to purchase the requisite GPUs and related equipment for $5.8 billion from Dell Technologies Inc.

Once the deal is fully implemented, it’s expected to generate approximately $1.94 billion in annualized revenue, IREN Chief Executive Officer Daniel Roberts said via email. It will use about 10% of IREN’s total capacity, leaving room for the infrastructure provider to sign more contracts and generate additional revenue.

“We’ve always viewed the major hyperscalers as natural partners,” Roberts said. “We’ve been in discussions with several of them, and those conversations have accelerated as both their compute requirements and our AI Cloud capabilities have grown.”

IREN shares surged more than 28% in pre-market US trading after the announcement, while Dell’s stock rose about 4%. Microsoft was little changed.

IREN, along with companies like CoreWeave Inc., Nebius Group NV, Crusoe Inc. and Nscale, is a member of the group of so-called neoclouds — data center operators that specialize in AI — that are vying to provide computing power to large hyperscalers like Meta Platforms Inc. and AI companies such as OpenAI. Many of these companies, including IREN, began as Bitcoin mining operations and have diversified their computing clusters into AI.

Its NASDAQ-traded shares were already up more than 500% this year through Friday’s close, lifted by bets on the artificial intelligence boom that’s made Nvidia a $5 trillion company.

Microsoft in particular has made significant use of leasing contracts with several of these companies to shore up its own resources as it races to sell more AI services and support the needs of customers like OpenAI. The Redmond, Washington-based company said it’s struggling to secure enough cloud capacity to fully satisfy demand for its Azure services when it reported earnings last week.

In Texas, the new GB300 systems will be installed in phases through next year at IREN’s facility in Childress. It’s planned to support 750 megawatts of capacity, the company said. IREN also has 2GW at its Sweetwater Hub near Abilene in the same state, and it’s seeing “strong interest in large-scale AI infrastructure deployments” there, Roberts said.