The water is so high people are kayaking to the train station in Queens.

WHO GETS PAID IN THE AI CLOUD STACK AI clouds are becoming the middle layer between scarce AI infrastructure and companies that actually need compute: Who supplies the AI clouds • $NVDA sits at center of compute layer by supplying the GPUs that determine how much capacity these platforms can actually bring online. • $MU & $SKHY supply HBM required to keep those GPUs fed as model sizes and inference workloads continue scaling • $DLR, $EQIX, $CORZ & $APLD provide physical data center infrastructure underneath the cloud layer giving AI clouds another path to scale without owning every building themselves. Who turns that infrastructure into compute • $CRWV, $NBIS & $IREN sit directly in middle of the stack by combining GPUs, power, networking and software into usable AI compute that customers can rent. Who buys the compute • $MSFT, $META & $GOOGL are unique because they are both customers and competitors by renting external capacity when internal supply is constrained while continuing to build their own • $SHOP & $CRWD show where next leg can come from as enterprise demand broadens customer base beyond a handful of hyperscalers and AI labs. • OpenAI & Anthropic represent AI lab customer base where demand can scale really quickly as training and inference requirements grow.

Met with a team today whose margins went from very negative to very positive within a few weeks just by switching to open weight models. Honestly wild