- AI token spending grew only 1.5% across August–September versus 42% in July, despite rising volumes; price competition and customer caution may cap provider revenue.
- The author argues “tokenmaxxing” is ending as enterprises confront weaker cost-benefit economics, reducing the expected pace of AI spending and data-center demand.
- The author now recommends materially reducing AI-trade exposure, including picks-and-shovels providers; high valuations require blockbuster beats while politics or a hyperscaler pullback could worsen the slowdown.
Investment pitches
filter:
64 80 - The author urges a significant reduction in AI exposure, especially picks-and-shovels providers, as AI spending growth is expected to slow.
- A capex slowdown could hit earnings harder than expected: hardware shortages encouraged stockpiling, while some data centers reportedly use only 35–40% of their GPUs.
- Politics may slow data-center construction, and a hyperscaler pullback could intensify the risk; after major stock gains and muted Nvidia earnings reactions, the author expects blockbuster beats to become harder to deliver.
73 - S&P 500 earnings growth is highly concentrated: Nvidia and Micron account for 34% of expected third-quarter growth, while 10 companies drive roughly two-thirds.
- The author argues circular hyperscaler spending inflates AI-related earnings while delivering no measurable financial payoff outside tech; slowing CAPEX growth could weaken the broader earnings cascade.
- UBS estimates hyperscaler AI CAPEX growth decelerates from 75% this year to 25% next year and 6% in 2028, prompting the author to expect slower earnings and a CAPEX pullback.
- With rates higher for longer and AI ROI disappointing, the author stands by reducing or eliminating exposure to AI picks-and-shovels, citing punished FCF despite earnings growth.