James van Geelen, founder of Citrini Research, stated that he has significantly reduced his semiconductor holdings and holds some short positions, citing that AI infrastructure financing has shifted from cloud vendors' own cash flow to equity and debt markets, making market sentiment a part of the fundamentals. He believes that cyclical stocks always peak when they appear "cheapest" based on next year's expected earnings.

In a podcast conversation on October 21, James van Geelen pointed out that tech giants' cash flow has dried up, and almost all incremental investment in AI infrastructure comes from equity or debt financing. This means a downturn in market sentiment could directly cut off funding, impacting actual construction spending. He estimates that supporting current infrastructure valuations requires AI labs' annual recurring revenue (ARR) to reach approximately $1 trillion by the end of 2028. However, consumer demand is nearing saturation, price reductions have not proportionally increased usage, and open-source models' share has risen from 1% a year ago to 18%, all of which are dragging down ARR growth.

He is currently long Meta (due to the potential monetization of its AI assistant Muse, with ARPU projected to grow fourfold by 2030) and short Google (due to DeepMind talent drain and threats to its search business). He believes the AI cycle will "end with a whimper, not a bang," meaning downward revisions in ARR expectations will drag down the entire sector. Additionally, he disclosed that AI can compress drug design time by 20%, but the real bottleneck lies in laboratories (such as monkey shortages), not hashrate. Therefore, investment opportunities are in clinical outsourcing (naming Charles River Laboratories), rather than AI concept stocks.