May. 25 at 2:47 PM
AI compute is being repriced higher via long-term contracts while spot markets tighten, leading to a layered AI infrastructure stack. Hyperscalers remain the primary buyers and builders, including
$GOOGL, MSFT, AMZN, META, ORCL, BABA, locking multi-year capacity. Microsoft ~
$627B RPO (+99% YoY), Oracle
$553B+ AI commitments, Google Cloud
$460B backlog (doubling), and Amazon
$225B+ Trainium orders all reflect compute scarcity being priced in.
Neoclouds like
$CRWV, NBIS, IREN,
$WYFI rent GPU capacity at market rates, with rapidly growing backlogs, showing demand being locked years ahead and pricing power shifting to fast-scale operators.
A third layer includes HPC and energy infrastructure firms such as
$APLD,
$HUT, GLXY, WULF, RIOT, CIFR, CORZ, MARA, CLSK, HIVE, focused on power, land, grid access, and GW-scale data center execution.