Captain $MARA

I don't think the main reason for $NBIS bulls to claim their deal was better than the one $IREN has, was because of software. The thing is, they never really gave an actual fundamental reason, other than: "we got a better deal because $NBIS is better", "you can see $NBIS even got a better deal with $MSFT than $IREN". The main take from $NBIS bulls is/was just that: Topline is higher so deal = better. But that's far from the truth, mainly because of the lack of details. I want to give an honorable mention to @jiahanjimliu who is not a $NBIS bull, but thought it may be because of uptime track record. I have never susbcribed to any of the above theories, but I can give you the real explanation in a fairly simple manner. $IREN has a fixed rack count in each Horizon data center. The design of the building and external cooling components, came out to be roughly 200kW of power at each rack, and sufficient cooling for that amount of power. When IREN signed Microsoft, they had to concede a timeline that was in line with $MSFT, and accept an older GPU model, the GB300. Since this model only draws 140kW inside a rack, and $IREN's data center is designed for a fixed amount of racks, the total available capacity can not be 100% utilized, because GB300 can not draw 200MW IT load across 12 Horizon buildings with a fixed rack amount. This means that there was a physical limit in the amount of GPUs IREN could lease to Microsoft in the Horizon 1-4 footprint, which resulted in the 76k GPUs across the 12 buildings. But 76k GPUs only draw ~110MW~115MW, and thus the 200MW is undersubscribed, and under-utilized. This is the reason that the topline is lower, because $NBIS on the other hand, had nothing build when they signed microsoft, and the design they have adopted, was never specifically for VR200 or a 200kW rack density. Rather than that, they gave DataOne the relatively easy task, to build 140kW rack density data centers, but just build more of those in terms of footprint, to get to 200MW it load. This results in a higher GPU count across the same allocated MW of IT load, and thus a higher topline for the contract. But that's only topline, and the reason Nebius hasn't disclosed their GPU count is because if you look at the actual TCO, the $/GPU is actually very similar to IREN (I estimate >120k GB300s in 200MW it load), and the same counts for the ROIC. But when you look at the margin profile, and the asset base, it becomes a whole different story. Because $NBIS also pays colocation fees to DataOne, which eats directly into the $/GPU hour revenue, and more notably, after 5 years, $IREN owns the data center, where $NBIS doesn't. Long story short, leaving out the details of the contract was the reason the deal got misunderstood. Zooming in explains why their topline is higher, but also why their bottomline is lower, and their unit economics are terrible. So software was not the reason why $NBIS bulls were gloating, but the lack of details around the GPU count which lead to misunderstood economics, was the main reason of their comparisons in favor of NBIS. This is also a big reason why I dislike Nebius as a company, because they called the MSFT contract a "fundraiser", even though they are not making any money from it, so it was just to pump their shareprice and run an ATM and convertible notes. Not disclosing details was in no commercial interest, it was just a way to hide their terrible unit economics and margin profile. This is ugly and uninvestible to me. Earnings will speak, and I hope they will get the reward they deserve for misleading their investors.

@BitcoinAIGuy FEW 😂

Thank you to @BTC_Bella69420 and @Buhlahkay. Real investor relations is built with trusted partners and disciplined messaging. Noise isn't strategy. Execution is.