Meta Considering Selling Excess AI Compute Triggers Memory Stock Implosion
"The first one we got over the last couple of days was Meta talking about trying to sell their excess AI compute. If there's infinite demand for compute, but now we're talking about Meta wants to sell some of the compute, maybe we have too much compute suddenly. That's all you needed, these two headlines, and suddenly you just have this momentum factor implosion that we're seeing."
About this episode
Forward Guidance hosts Jack Farley and Quinn Thompson dissect a violent momentum factor implosion in AI and semiconductor stocks that erased months of gains in days, coinciding with apparent Japanese yen intervention. Thompson reveals that 30% of hyperscaler income earlier this year came from marking up private AI lab valuations, creating hidden fragility as the AI trade shows cracks. The hosts identify two specific catalysts: Meta considering selling excess AI compute capacity and rumors of a memory efficiency breakthrough by an OpenAI spinout, both contradicting the infinite demand narrative. They draw parallels to July 2024, when the Fed made a hawkish mistake before pivoting dovish, arguing current conditions make rate hikes impossible despite hawkish dot plots. With labor force participation falling, wage growth weak, and forward inflation indicators declining, they see peak growth and peak inflation converging. Thompson explicitly calls the bottleneck trade a bubble, noting stocks in Japan, Korea and Taiwan reversed two to three months of price action in days. Both hosts favor trades expressing fading Fed hawkishness—particularly gold and silver—over buying tech dips, warning the administration's market interventions may not prevent further weakness. They discuss Bitcoin and MicroStrategy as oversold but question sustainability without Saylor's buying, while criticizing crypto token structures where equity holders double-dip at token holder expense. The conversation reveals deep skepticism about near-term equity market strength despite potentially improving Fed positioning.
Key takeaways
- Quinn Thompson reveals 30% of hyperscaler income in early 2025 came from marking up private AI lab valuations, creating hidden collapse risk if AI trade unwinds
- Meta considering selling excess AI compute and rumors of memory efficiency breakthrough triggered momentum factor implosion erasing months of semiconductor gains in days
- Hosts argue Fed rate hikes impossible despite hawkish dots given falling labor participation, weak wage growth, and declining forward inflation indicators
- Thompson explicitly calls AI bottleneck trade a bubble as Japan, Korea and Taiwan semiconductor stocks reverse two to three months of gains in days
- Both analysts favor gold and silver over tech stocks, seeing peak growth and peak inflation coinciding as Fed hawkishness setup mirrors July 2024 policy mistake
- Bitcoin and MicroStrategy viewed as oversold but hosts question sustainability of rally without Saylor's aggressive buying supporting market structure
- Hosts criticize crypto token structures where founders hold equity while selling governance tokens with no equity claims, calling for industry consolidation around legitimate protocols