It's been a while since I've put out any meaningful posts due to my life outside of AI miners. Here is a couple of things that caught my eye from the latest $IREN earnings call and presentations deck. This is the debut of Fireworks AI (@FireworksAI_HQ) in any of the $IREN published literature as a customer. Notice that the customers that IREN has featured on their deck in $MSFT, Together AI, FluidStack, and Fireworks AI, are all PaaS providers. This just reiterates how $IREN views it's market positioning going forward in AI compute, they will not seek to go beyond IaaS, as they feel there is simply no need for them to do so. Given their strength is on the infrastructure side and they have partnered themselves with some of the best players that live at the PaaS level and above. During the earnings call, this is what Co-CEO Dan Roberts and CCO Kent Draper had to say: Kent Draper: To date, the vast majority of our customers have required a bare metal offering and that is their preference. You know, these are all highly advanced AI or software companies like a Microsoft, they have significant experience in the space and they want the raw compute and the performance benefits that brings having access to a bare metal. Dan Roberts: But for a business like ours, that is pursuing scale and monetizing a platform that we've spent the last seven years building. It's very hard to see how you get scale by focusing on software, which is I think everyone generally accepts going to be commoditized anyway in coming years, as compared to just selling through the bare metal and letting these guys do their thing on it. This clearly presents IREN's goal and positioning as being the provider of the highest quality of AI infrastructure at the best possible price. This will allow their software specialist partners to lay their platforms on top of the quality IREN AI infrastructure. This is the segment that I fully expect $IREN to dominate in, I don't think you can name me a peer at this moment for this segment. The other slide that caught my attention was the one with the mentioning of the "100 MW superclusters". As Jensen Huang has said: horizontal scaling is easy, vertical scaling is extremely hard. Former IREN ally, Eiso Kant, also had a chat with IREN CTO Denis Skrinnikoff at RAISE Paris regarding the difficulties of vertical scaling beyond 10k GPUs. Being able to build 100 MW superclusters is validation of the AI infrastructure building prowess of IREN, there are not too many entities outside of the current hypescalers that can do this right now. IREN has stated that each 50 MW Horizon DC can house about 19k GB300 GPUs, so a 100 MW critical IT load using Horizon DCs would be a supercluster of around 38k GB300 GPUs. An interesting note is that a 38k GB300 cluster would be more powerful than a 100k H100 cluster for training by 1.9 to 2.7x. This also brings up a pretty interesting question. If $IREN has the prowess to build 100 MW superclusters, and is trusted enough to be partnered with $MSFT for a $9.7bil deal to provide IaaS. Then why would a "trusted" publication such as SemiAnalysis still have them in the "underperforming" bucket. Which of those other names in that bucket is capable of building a 100 MW supercluster or have a mega deal with MSFT to provide IaaS? 🤷‍♂️ ...