Microsoft signs a five-year, ~$9.7B deal to buy AI compute capacity from Sydney-based IREN, giving Microsoft access to Nvidia's GB300 in IREN's Texas facility
Microsoft Corp. has signed a roughly $9.7 billion deal to buy artificial intelligence computing capacity from IREN Ltd., becoming the Australian company's largest customer.
Context & Ripple Effects
Microsoft has already been widening its supply of external AI capacity, with more than $33 billion committed to neocloud providers, including a reported $19.4 billion Nebius capacity agreement. This IREN contract extends that procurement model to a Texas-hosted deployment using Nvidia’s GB300 systems.
The move also follows Microsoft’s earlier use of partner infrastructure: its multi-year Oracle GPU-access agreement showed that meeting AI-service demand need not rely solely on Azure-owned hardware. IREN becomes a material new supplier in that broader capacity portfolio.
First-order effects
- Microsoft secures five years of AI compute capacity from IREN’s Texas facility, including access to Nvidia GB300 systems, while IREN gains Microsoft as its largest customer.
- The contract turns a large share of IREN’s near-term business toward a single hyperscale buyer and provides a long-duration demand commitment for its AI infrastructure.
Second-order effects
- Other neoclouds and infrastructure operators face stronger pressure to offer hyperscalers financeable, long-term GPU capacity rather than compete only for shorter-term workloads.
- Nvidia benefits indirectly as a named hardware platform in another large capacity commitment, while Microsoft can diversify supply beyond capacity it builds or accesses through other partners.
Third-order effects
- If similar contracts continue, AI compute is likely to be procured increasingly through long-duration capacity agreements, separating the ownership and operation of GPU facilities from the cloud services sold to end customers.
- That model can deepen customer concentration among infrastructure providers: large contracts improve project visibility but also make providers more exposed to a small number of hyperscale counterparties.
The trend: Hyperscalers are assembling AI capacity through a diversified network of specialist operators, using multi-year commitments to secure scarce high-end GPU infrastructure.