Microsoft signs $9.7B AI cloud deal with IREN as Oracle expands AMD-powered AI partnership

Microsoft has agreed to a five-year, approximately US$9.7 billion contract with IREN Limited (IREN) to secure large-scale GPU-based cloud capacity — a move aimed at easing its crunch for AI infrastructure. Meanwhile, Oracle Corporation is deepening its partnership with Advanced Micro Devices (AMD) to launch a publicly available “AI supercluster” starting in Q3 2026, powered by tens of thousands of AMD Instinct GPUs. Together, these deals underscore how hyperscale players are racing to lock in compute power for the next wave of AI.

Background / Context

Over the past year, the cloud-and-AI infrastructure market has become a battleground of scale, cost, and speed. Demand for GPUs and other accelerators to train large language models, generative AI systems, and inference engines has exploded. Microsoft, among others, has flagged a capacity shortage extending into 2026.

At the same time, traditional cloud firms such as Oracle are expanding beyond their legacy software and database strengths into AI infrastructure, seeking to compete with established hyperscalers by offering large-scale, optimized clusters. Their strategic tie-ups with chip firms reflect that push.

Meanwhile, new entrants are emerging. IREN, formerly associated with large-scale mining infrastructure, is pivoting into “neocloud” AI infrastructure services — illustrating how compute assets once used for cryptocurrency are being repurposed for AI workloads.

Key Developments / Details

Microsoft-IREN deal:

• Microsoft and IREN signed a five-year agreement valued at about US$9.7 billion.
• Under the contract, Microsoft will gain access to GPU-based infrastructure — specifically tied to NVIDIA’s GB300 architecture — hosted by IREN at its 750 MW campus in Childress, Texas.
• IREN has also signed a separate deal worth approximately US$5.8 billion with Dell Technologies to purchase GPUs and ancillary equipment for deployment.
• The deployment is to be phased through 2026, with IREN expecting annualized revenue of about US$1.94 billion from the deal once fully implemented.
• Microsoft reportedly made a prepayment of about 20% as part of the agreement.

Oracle-AMD partnership:

• Oracle and AMD announced that starting in calendar Q3 2026, Oracle Cloud Infrastructure (OCI) will be the first major cloud provider to offer a publicly available AI supercluster powered by 50,000 AMD Instinct MI450 Series GPUs, with further expansion planned for 2027 and beyond.
• The cluster will use AMD’s “Helios” rack architecture: MI450 GPUs, next-gen EPYC CPUs codenamed “Venice,” and Pensando advanced networking (“Vulcano”).
• AMD’s MI450 Series GPUs feature up to 432 GB of HBM4 memory and 20 TB/s bandwidth, enabling customers to train or infer models roughly 50% larger than previous generations.
• Oracle emphasized that this offering gives customers a choice beyond the dominant GPU provider, and positions OCI to scale training and inference workloads for large language and agentic models.

Technical Explanation

Think of AI workloads like ultra-marathon computers: they need massive endurance (memory), speed (bandwidth), and coordination (networking) to train and run large models.

• In the Microsoft-IREN deal, Microsoft is essentially leasing compute “lanes” from IREN’s data-centre highway rather than building the highway itself — helping the company scale faster without the full upfront cost of new infrastructure.
• For Oracle-AMD: the “supercluster” is like assembling thousands of high-performance engine blocks (GPUs) plus next-gen CPUs and high-speed interconnects into a single machine, so it can handle massive AI training jobs with minimal delays and maximum efficiency.
• The high memory and bandwidth specifications mean fewer bottlenecks — less need to split models across many chips and wait for communication. That leads to faster, more efficient training and inference.
• The rack-scale “Helios” design means many GPUs, CPUs, and networking units in one optimized chassis, reducing cost per unit of performance and power consumption — critical for AI infrastructure economics.

Implications

• For cloud customers and enterprises: These deals signal faster access to top-tier AI infrastructure. Enterprises using Azure or OCI may be able to train or deploy more complex models sooner and more cost-effectively.
• For the AI infrastructure market: Compute scarcity, especially GPU supply, is a real constraint. Firms are locking in capacity years ahead, as seen in Microsoft’s move via IREN. Oracle’s partnership with AMD highlights the growing diversification beyond NVIDIA.
• For competition in the cloud space: Oracle’s move positions it more strongly against AWS, Microsoft Azure, and Google Cloud in the high-end AI market. AMD’s involvement gives it a stronger play versus NVIDIA in the AI boom.
• For data centre and energy implications: Building large GPU clusters consumes significant power and cooling infrastructure. IREN’s Texas site and Oracle’s rack designs underscore the energy intensity of AI computing.
• For the broader economy: Multi-billion-dollar contracts like these highlight the high stakes of the AI compute race, influencing not just technology markets but also data-centre real estate and energy policy.

Challenges / Limitations

• Execution risk: Deploying thousands of GPUs and meeting strict delivery deadlines is complex. IREN’s deal success depends on timely milestone completion.
• Technology risk: AI hardware evolves quickly; today’s cutting-edge systems could be outdated within a few years.
• Supply chain and power constraints: Large GPU clusters require vast electrical power and advanced cooling, which could become bottlenecks.
• ROI uncertainty: The cost of building and operating these clusters is massive, and revenue realisation may lag demand cycles.
• Competitive risk: As more players expand capacity, AI infrastructure may become commoditised, shifting value toward software and services.

Future Outlook

• Expect more long-term capacity contracts like Microsoft’s as cloud providers secure future compute access.
• Competition among chip suppliers (AMD, NVIDIA, and others) will intensify, leading to greater choice and possibly lower costs.
• Infrastructure innovation — such as liquid cooling, optical networking, and modular racks — will accelerate to improve efficiency.
• Over the next few years, the build-out of AI superclusters will likely define which cloud platforms dominate next-generation workloads.

Conclusion / Summary

The deals between Microsoft and IREN and between Oracle and AMD mark major milestones in the global race for AI-scale cloud infrastructure. They illustrate how compute capacity and hardware partnerships are becoming the new battleground for AI dominance. For enterprises and developers, these moves promise faster, more accessible AI compute — but also signal rising stakes in cost, energy, and competition.