tailwagdog

$IREN's Unique IaaS Proposition: Infrastructure-Heavy, Hardware-Light Yesterday during the Orange Group panel at Pubkey in NYC, @mikealfred explained that one of IREN’s unique advantages compared to other neoclouds is that, by leasing the chips instead of buying them, they are able to return the chips within the lease period and effectively replace them with a newer model. This allows IREN to keep their fleet next-gen or current-gen without having to invest in a completely new fleet, while others are depreciating a previous-gen fleet that’s burdened with an interest-accruing financing instrument but is no longer receiving a premium per GPU hour. Ironically, as a vertically integrated AI infrastructure company, this resembles an asset-light model, where the chips never really have to be owned, yet they can bring returns that—minus interest expenses—yield a profile as if they were owned. IREN prefers optionality and flexibility, but by being vertically integrated and responsible for the data center shell, there is a certain degree of inflexibility when it comes to designing the shell—or is there? There are certain components of a data center shell that can remain in place as GPU generations come and go. Think of electrical redundancy, networking equipment, and—for example—a liquid-cooled plant, UPSs, backup generators, and even substations. In a perfect world, IREN would be able to swap their GPUs every 18 months to keep up with the newest cycles, while having very little retrofit expense, basically no downtime, and effectively a hot-swap to the next-gen chip that’s delivered to them as a preferred partner. But the world isn’t perfect… yet. We are getting there fast. When $NVDA first spoke of “AI factories,” the world was unsure what this really meant. At first, this sounded like a big data center campus that operates a very large cluster of GPUs that all work in sync and power the AI world of tomorrow. Directionally this is correct, but how does it look in practice? Can IREN build and operate an AI factory that has a massive fleet of leased GPUs that can be exchanged for a newer generation every 18 months? To enable the rollout of their newest chips, NVIDIA is advancing in the infrastructure layer to bring not only complete servers but also pod-level systems/reference designs that can be used as a prefab solution for data center halls. These pods can help infrastructure companies such as IREN easily design and create a data center shell that is optimized for rapid deployment of these GPU-ready pods. By tackling the connection between the shell and the chip with a completely integrated pod solution, NVIDIA makes it easier for data center companies to scale faster and deploy chips more rapidly, greatly improving time to market (TTM). So IREN is potentially getting help from NVIDIA to scale faster with a more prefab solution that promotes modularity and general-purpose design, effectively lowering the bar for making a powered shell GPU-ready and reducing the delivery time of a turnkey data center. But what does all this mean for IREN and their prospective clients? We now know that IREN has the ability to capture most of the value chain from raw power all the way to compute. By being vertically integrated, the ownership of land, power, and data center infrastructure brings a high degree of control, while prefab and modular solutions give IREN the optionality to remain flexible in their internal design. Within this structure, IREN brings a GPU lease structure that provides them with the flexibility to return their GPUs and effectively swap them for a newer generation—charging the flywheel with the next-generation chip, which commands the highest premium per GPU hour. This keeps the entire AI cloud operation nimble but highly profitable, while retaining ownership of the assets that do keep their value over time. But this is not just good for IREN internally. By being infrastructure-heavy but “hardware-light,” IREN is a premium choice for clients that are looking for a long-term Infrastructure-as-a-Service (IaaS) deal but want to retain the flexibility of having the latest chips. This optionality is a massive differentiator for clients that are looking for a long-term deal—especially when there is a large amount of power involved—but are seeking a counterparty that has the means to deliver. In an IaaS deal, the capital expenditure of the provider typically consists mostly of GPUs, with infrastructure build costs a much smaller part. But that’s just looking at costs. When we look at the $NBIS deal with $MSFT, we can see that there are clauses that give Microsoft the right to terminate the agreement or terminate tranches of the $17.4B maximum contract value. The Nebius 6-K reads “subject to delivery and availability,” which means that delivery is in tranches and must be on time. Moreover, the entire deal is contingent on Nebius having the money lined up. Obviously, with a deal like this, there will be lenders lined up to facilitate financing to Nebius, so I’m not worried about this coming to fruition. That said, when it comes to IREN, they are a much smaller company than Nebius in market cap (at least currently), and even now Nebius is mentioned as a small company compared to Microsoft. So what does this mean for IREN compared to a hyperscaler, and how does this impact concerns about financing from this smaller infrastructure player? This is where the GPU lease structure becomes IREN’s distinctive USP. With GPUs being ~50–70% of total capex for an IaaS provider, removing much of this as a deal liability by having a lease finance structure for the chips provides IREN with a unique position to attract clients that want to go big and wide but retain the optionality to change the chips over the course of the deal period. So here we are, at the forefront of a perfect world: IREN builds and owns the valuable power and infrastructure assets, with the ability to scale fast utilizing prefab pods enabled by NVIDIA to rapidly deploy large clusters of financed GPUs in a lease structure, bringing AI factories with the newest GPU models yearly, in a vertically integrated business model, to prospective clients looking for large amounts of compute but wanting to avoid execution and financing risks. IREN is ready to deliver: power, land, and—soon—data centers will be available. If you had any doubts that IREN may not monetize their entire power portfolio, I request you to reconsider. This is not just about IREN looking for a client; it’s the client that will need to realize what a unique opportunity this deal will be for them.