Background – surging AI compute demand and GPU shortages
The boom in generative AI has produced insatiable demand for compute power. Language‑model inference and training jobs require massive clusters of advanced GPUs and high‑speed networking. Hyperscalers such as Microsoft, Amazon and Google are already spending tens of billions of dollars per quarter on chips and data‑centre capacity. In Microsoft’s July‑September 2025 quarter alone the company spent nearly US$35 billion on capital expenditures, with roughly half earmarked for computer chips and the remainder for data‑centre real‑estateabcnews.go.com. The surge in AI spending has exposed capacity constraints that are slowing the rollout of new services. A Reuters report on 3 November 2025 noted that Microsoft’s ability to cash in on the AI boom is limited by a “computing crunch” and capacity shortagesreuters.com. These shortages have pushed hyperscalers to seek alternative sources of high‑density GPUs (“neocloud” providers) rather than relying solely on their own infrastructurereuters.com.
GIGABYTE GeForce RTX 3050 WINDFORCE OC V2 6G Graphics Card, 2X WINDFORCE Fans, 6GB GDDR6 96-bit GDDR6, GV-N3050WF2OCV2-6GD Graphics Card
NVIDIA Ampere Streaming Multiprocessors
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
IREN’s evolution – from bitcoin mining to AI cloud
IREN Limited (formerly Iris Energy) is emblematic of how energy‑rich crypto miners are repurposing their infrastructure for AI. Originally a bitcoin‑mining company, IREN owns more than 2.9 GW of secured power capacity across North America and operates large campuses in Texas and Canadareuters.com. In March 2025 the firm announced that once its bitcoin mining capacity reached 52 EH/s it would pause further mining expansion and redirect capital to AI data‑centre projects, citing the superior economics of AI hostingtheminermag.com. Its first AI facility, “Horizon 1,” at the Childress campus in Texas is a 75 MW liquid‑cooled data centre designed to support 200 kW per rack using direct‑to‑chip cooling. The 420‑acre campus totals 750 MW, with around 450 MW already energiseddatacenterdynamics.com.
Cold Heading Roughing Finishing Ai Server Nvqdocp Data Center Connector Cooling Liquid Water-Cooled Connector
Broad Compatibility; Features standard interface design for seamless connection with various common liquid cooling systems and components, facilitating…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Microsoft–IREN contract
Contract terms and financials
On 3 November 2025, IREN announced a five‑year GPU cloud‑services contract with Microsoft. The agreement grants Microsoft access to next‑generation NVIDIA GB300 GPUs on IREN’s AI cloud platform and has a total value of about US$9.7 billion, including a 20 % prepaymentiren.gcs-web.comabcnews.go.com. IREN simultaneously signed a supply agreement with Dell Technologies to purchase the GB300 GPUs, servers and ancillary equipment for approximately US$5.8 billioniren.gcs-web.com. Microsoft’s prepayment will fund part of these capital expendituresreuters.com.
IREN will phase deployment of the GPUs through 2026 at its 750 MW Childress campus in Texasiren.gcs-web.com. The campus will add liquid‑cooled data centres delivering about 200 MW of critical IT load across “Horizon 1” through “Horizon 4”iren.gcs-web.com. The contract can be terminated if IREN fails to meet delivery timelinesreuters.com, highlighting execution risk. According to TechCrunch, IREN’s CEO expects the Microsoft contract to occupy only about 10 % of the company’s total capacity and to generate roughly US$1.94 billion in annualised revenuetechcrunch.com.
Table 1 – Key contract parameters
| Parameter | Value / description | Evidence |
|---|---|---|
| Contract parties | Microsoft and IREN Limited (GPU cloud services) | Official IREN releaseiren.gcs-web.com |
| Term | Five years | IREN releaseiren.gcs-web.com, TechCrunchtechcrunch.com |
| Total value | ~US$9.7 billion with 20 % prepayment | IREN releaseiren.gcs-web.com |
| Hardware supplier | Dell Technologies will supply GPUs, servers and ancillary equipment (about US$5.8 billion) | IREN releaseiren.gcs-web.com |
| Technology | NVIDIA GB300 GPUs (Blackwell Ultra) housed in Dell integrated racks | IREN releaseiren.gcs-web.com, Dell blogdell.com |
| Deployment location | Childress campus (Texas), 750 MW site with 200 MW liquid‑cooled AI capacity | IREN releaseiren.gcs-web.com |
| Deployment schedule | Phased deployment through 2026 | IREN releaseiren.gcs-web.com |
| Purpose | Provide Microsoft with additional AI computing capacity without building new data centres | Reuters reportreuters.com |
Technology stack – NVIDIA GB300 NVL72 and Dell integrated rack
The contract will use NVIDIA GB300 NVL72 systems built on the Blackwell Ultra architecture. Each rack integrates 72 Blackwell Ultra GPUs and 36 Arm‑based Grace CPUs, acting as a single massive GPU designed for test‑time reasoning. NVIDIA states that the GB300 NVL72 platform delivers 1.5× more AI performance than the prior GB200 generation and can provide more than one exaflop of dense AI performance and up to 40 TB of high‑bandwidth memory per rackinvestor.nvidia.comdell.com. The racks use NVLink switch trays and high‑speed InfiniBand/Ethernet networking to support ultra‑low‑latency, high‑bandwidth communication among GPUspress.asus.cominvestor.nvidia.com. To handle the roughly 200 kW per rack heat load, the design employs liquid cooling and manifold‑based direct‑to‑chip coolingpress.asus.comdatacenterdynamics.com.
Dell’s Integrated Rack Scalable System for the GB300 NVL72 includes PowerEdge XE9712 servers with 72 NVIDIA Blackwell Ultra GPUs and 36 Grace CPUs, plus NVLink switch traysdell.com. Dell reports that each rack provides more than one exaflop of dense AI performance and around 40 TB of fast memory, and that the rack is fully assembled, liquid‑cooled and tested before deploymentdell.com. The order from IREN (roughly US$5.8 billion) underscores Dell’s growing role as a primary supplier of rack‑scale AI infrastructure.
4U Server Cabinet Case – Rackmount Server Chassis with 7 PCI Slots, Lockable with Key
Server Cabinet Case:The 4u server cabinet case adopts a combined internal architecture.With 7 x PCI slot, providing additional…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for Microsoft
- Securing scarce GPU capacity: By contracting with IREN, Microsoft gains access to cutting‑edge Nvidia GB300 GPUs without having to build its own new data centres or secure additional power, thereby addressing two major bottlenecks in meeting surging AI demandreuters.com. The partnership allows Microsoft to continue scaling generative‑AI services such as ChatGPT and Copilot while capitalising on IREN’s power‑rich sites.
- Capital expenditure optimisation: Accessing GPUs via a services contract sidesteps some of the heavy capital expenditures associated with owning high‑end chips that depreciate quicklyreuters.com. Given Microsoft’s enormous capex (US$35 billion in a recent quarterabcnews.go.com), spreading risk through service contracts can improve financial flexibility.
- Complementing other neocloud deals: The IREN contract is part of a broader strategy in which Microsoft procures capacity from specialist providers. Reuters notes that demand for AI compute has propelled “neocloud” companies like CoreWeave and Nebius ahead in the AI race; Microsoft recently signed a US$17.4 billion deal with Nebius for infrastructure capacity and also announced agreements with Lambda and Nscalereuters.comtechcrunch.com. Together, these contracts diversify Microsoft’s supply chain and reduce reliance on a single supplier or region.
NVIDIA RTX PRO 6000 Blackwell Server Edition
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for IREN
- Validation of AI cloud strategy: The deal validates IREN’s pivot from bitcoin mining to AI infrastructure. With a five‑year commitment worth US$9.7 billion, IREN locks in long‑term revenue and positions itself as a trusted hyperscale partner. The company’s management noted that the agreement opens a new customer segment among global hyperscalers and reinforces IREN’s position as a leading AI cloud service provideriren.gcs-web.com.
- Financial impact and capacity utilisation: CEO Daniel Roberts expects the Microsoft contract to utilise only about 10 % of IREN’s total capacity but to produce roughly US$1.94 billion in annualised revenuetechcrunch.com, implying margins high enough to fund expansion. The 20 % prepayment from Microsoft helps finance the US$5.8 billion hardware purchase from Delliren.gcs-web.com and reduces upfront capital requirements.
- Execution risk: The contract can be terminated if IREN misses delivery milestonesreuters.com. IREN must therefore complete construction of multiple liquid‑cooled data‑centre phases by 2026, integrate tens of thousands of GPUs and maintain reliability. The company’s ability to fund additional phases may depend on accessing debt and equity markets or securing further prepaymentsiren.gcs-web.com.
- Strategic transformation: IREN’s pivot reflects a broader trend among miners to repurpose power‑dense sites for AI. Analysts have noted that bitcoin mining revenues are declining due to halving events, whereas AI hosting offers higher, more stable returnstech.co. The contract therefore signifies not only a financial win for IREN but also a strategic repositioning toward the data‑centre market.
Implications for Dell
- Revenue opportunity: IREN’s US$5.8 billion order for GB300 racks is one of the largest commitments for Dell’s AI infrastructure business. The company recently announced that it was the first to deliver GB300 NVL72 racks to CoreWeavedell.com. Dell’s integrated approach – combining compute, networking, storage and cooling – positions it as a leading supplier of rack‑scale AI systems, and high‑profile wins with CoreWeave and IREN provide proof of execution.
- Ecosystem leverage: Dell is co‑developing the AI Factory with NVIDIA, which bundles servers, networking, software and services. By supplying fully integrated, liquid‑cooled racks, Dell differentiates itself from competitors that provide only components. The IREN deal, along with other hyperscale deployments, will likely accelerate the adoption of Dell’s PowerEdge XE series and drive service revenues.
Implications for Nvidia
- Reinforcing GPU dominance: The contract relies on NVIDIA GB300 GPUs, part of the Blackwell Ultra family announced in March 2025. Each GB300 NVL72 connects 72 GPUs and 36 Grace CPUs and offers 1.5× more AI performance than the previous generationinvestor.nvidia.com. These systems are purpose‑built for AI reasoning and agentic workloads, emphasising inference scaling and high‑bandwidth memoryinvestor.nvidia.com.
- Demand visibility: Hyperscalers and neocloud providers are placing multi‑billion‑dollar orders years in advance, providing NVIDIA with revenue visibility and strengthening its bargaining power. NVIDIA’s press release lists Dell, Cisco, HPE, Lenovo and other OEMs as partners expected to deliver Blackwell Ultra‑based servers starting in H2 2025investor.nvidia.com. Contracts like Microsoft–IREN demonstrate that customers are willing to commit long term to secure supply.
- Ecosystem expansion: Nvidia’s dominance extends beyond hardware. The Blackwell Ultra platform integrates high‑speed networking (Spectrum‑X and Quantum‑X) and software like NVIDIA Dynamo for inference orchestrationinvestor.nvidia.com. The IREN deal will deploy these technologies at scale, fostering adoption of Nvidia’s full AI factory stack.
Broader AI infrastructure market
- Rise of neocloud providers: Supply constraints have enabled specialist infrastructure providers (CoreWeave, Nebius, Nscale, Lambda and IREN) to secure large pre‑committed contracts. Reuters notes that such “neocloud” firms selling cloud services built on Nvidia processors are gaining prominencereuters.com. Microsoft’s multiple deals (US$17.4 billion with Nebius and similar arrangements with Nscale and Lambda) suggest that hyperscalers are diversifying supply chains to mitigate GPU shortagesreuters.comtechcrunch.com.
- Pivot of crypto miners: As bitcoin halving reduces mining profitability, miners are repurposing their power‑dense sites for AI compute. Tech.co notes that miners such as IREN, Riot, TeraWulf, CleanSpark and Cipher Mining are converting facilities to host AI workloadstech.co. IREN’s decision to pause mining expansion at 52 EH/s and invest in AI data centres exemplifies this shifttheminermag.com.
- High‑density, liquid‑cooled data centres: To accommodate GB300 racks, data centres must support power densities around 200 kW per rack and rely on direct‑to‑chip liquid coolingdatacenterdynamics.compress.asus.com. IREN’s Horizon 1 facility is designed for such densities and will cost US$300–350 milliondatacenterdynamics.com. This highlights a broader trend towards liquid‑cooled, modular AI facilities capable of rapid deployment.
- Investment scale: Goldman Sachs and other analysts estimate that global AI‑related infrastructure spending could reach US$3 trillion–US$4 trillion by 2030 (noted in broader industry reports). The Microsoft–IREN contract, along with similar deals, indicates that hyperscalers are willing to commit tens of billions to secure compute capacity years ahead of need.
Conclusion and outlook
The Microsoft–IREN AI cloud contract underscores how the AI revolution is reshaping technology supply chains. Facing acute GPU shortages and skyrocketing demand for reasoning models, Microsoft has chosen to partner with specialist infrastructure providers rather than building all capacity in‑house. IREN, a former bitcoin miner, has repositioned itself as an AI cloud service provider, leveraging its energy‑rich assets to secure a multi‑billion‑dollar hyperscale contract. Dell and Nvidia emerge as key beneficiaries, supplying the high‑density, liquid‑cooled racks and advanced GPUs that make such deals possible. The broader market is witnessing the rise of neocloud providers and the conversion of crypto mining infrastructure to AI data centres, with investments measured in tens of billions of dollars and rack densities approaching 200 kW.
Looking ahead, the success of this contract will depend on execution – IREN must complete its phased data‑centre build‑out and integrate tens of thousands of GPUs. If successful, the partnership could serve as a template for future hyperscale collaborations and accelerate the transition of power‑intensive industries into AI infrastructure. It also signals that high‑density, liquid‑cooled data centres are becoming the standard for next‑generation AI workloads. Ultimately, the Microsoft–IREN deal is not just a response to current shortages but a strategic move to shape the future of AI infrastructure.
