Understanding IREN isnt that hard. For IREN, HBM improving efficiency has at least three layers. First is the customer math. More HBM capacity, higher memory bandwidth, can run a bigger model. What the customer cares about is how many more tokens they get for the same hour of money. Second is pricing power. Who has the lower total cost for the same job. Comparing $NBIS, $IREN, and $CRWV on hourly rent alone isnt enough. Look at price divided by effective token throughput. Third is the one I watch. Revenue per MW. Every data center is capped by power. That flows through. More tokens per MW → higher customer value → higher dollars per MW → shorter payback. IREN recent 3-year contracts are already above $20M per MW a year. Company puts payback at about 2 years. So the way I read it now, whats scarce is how many useful tokens a MW can keep producing. $MU moving, HBM getting more expensive, does lift IRENs server cost. If the server goes from 100 to 110, and the new HBM lifts token throughput by more than that, the customers cost per token can still come down. Thats what Im hoping for. Then IREN has a shot at higher utilization, pricing power, and better revenue per MW at the same time.

$NKE might of actually bottomed today, like for real.