my view is simple:

memory still remains the core of the current AI trade

but after this move, the more interesting opportunity may shift toward the second-order beneficiaries: neoclouds, interconnects, memory pooling, CXL and the broader infrastructure layer that makes inference scale possible

these are likely the specific constraints the market has not fully priced yet

the AI trade is also becoming more macro-sensitive

this is no longer just about earnings and capex. it is also about real yields, credit spreads, dollar liquidity, power costs and the long end of the curve.

AI infrastructure is capital intensive:

- if real yields rise, multiples compress. if credit spreads widen, financing becomes harder. if power costs rise, inference economics deteriorate
- but if inflation cools, long-end yields stabilize and credit remains open, the AI buildout can continue

that is where inference matters

every agent, coding workflow, enterprise automation and real-time AI interaction consumes compute. training was the first compute shock, but inference is recurring

if inference can be monetized profitably, AI infrastructure demand becomes recurring too

so no, i do not think this is the end of the AI trade

i think the market is moving into phase two

phase one was about obvious scarcity

phase two is about the next bottleneck: memory bandwidth, interconnects, neocloud capacity, CXL, power and compute utilization

the trade is not over, only becoming more selective

https://x.com/arndxt_xo/status/2071153352252272865