May. 24 at 5:56 PM
To find the next
$SNDK, listen to what hyperscalers keep complaining about in earnings calls rather than headline numbers. The recurring constraints are where opportunity shows up.
In 2024 NVDA repeatedly highlighted tight HBM, packaging, and memory supply, with demand far exceeding supply.
$SK Hynix prioritized HBM for AI,
$MU was sold out into 2025, and SAMSUNG ran into thermal and power issues qualifying HBM for NVDA.
The core issue is simple: AI depends on HBM, and HBM has been structurally constrained, leading to a repricing of the memory stack.
Now the bottlenecks have expanded into power, cooling, data centers, interconnects, and storage. MSFT, AMZN, GOOGL,
$META, and
$ORCL keep repeating capacity constraints.
Capital ultimately flows into infrastructure.
