Hyperscaler compute monetization breaks AI infrastructure's scarcity premium, memory holds.
When a hyperscaler with existing GPU capacity begins selling compute externally, the pricing assumptions baked into every neocloudco revenue model require a rethink — and that repricing reached the semiconductor stack unevenly today.
Financials led a split tape on July 1, with seven of twelve sectors positive, while semiconductor and AI-infrastructure names sold off sharply as Meta's reported move to monetize existing GPU capacity externally threatened the compute-scarcity pricing models underpinning neocloudcos like CoreWeave and Nebius.
Since the last briefing
Yesterday the committee called for energy-sector avoidance over single-name quality bets within refiners, paired with maintaining AI-infrastructure exposure where confirmed multi-week momentum and institutional accumulation align — specifically NVDA over PBF
What drove the session and how it transmitted, the exposure table by persistence grade, where the committee disagreed, its verdict and weighting, and the falsifiers that would prove it wrong.