My original wording, cleaned up by Claude
Key points this past week:
- AI rout that started in the previous week over OAI ARR of $50B vs. $70B worsened a lot more this week. CRWV CDS rose, pricing in significant default risk. Anthropic also showed a slowdown in ARR growth, which did not help.
- I was watching previous week 10/8 and 10/9 action and noted how NVDA and ORCL both did not reclaim the 10/8 high, which sounded an alarm.
- I wrote this in my personal note on 10/9: “10/8 there was a FT news about OAI ARR being 50B, short of the 70B estimate earlier. This turned out to be a miscommunication/misunderstanding of sorts, with the 70B number being concocted by investors to match numbers and essentially using different accounting methods.
What is interesting is that 10/9, as this got cleared up, NVDA, AMD, INTC, MU and SMH/SOXX broadly continued to sell-off in a rejection of the bounce. While ORCL rallied (but did not reverse the sell-off). AMZN, MSFT much stronger, GOOG higher but weak.“
- Oil was also higher on the week.
- CPI and PPI both came in hotter than expected, but yields shrugged off the inflation. Warsh’s IMF fireside chat showed him as less hawkish than he could have been, which dampened odds of an October hike. More importantly, the bid in treasuries, and to a lesser extent gold, is likely haven demand, showing that growth risk is now greater than inflation risk.
The wave:
- Initial trigger perhaps 9/16 Fed 25bps hike
- 10/8: OAI surprise
- 10/9: NVDA, ORCL and CRWV were not able to regain the losses, which was alarming
- 10/14: CPI caused an initial wave of selling with CRWV CDS rising. However, bonds regained their CPI losses later in the day, presumably from haven demand. There was also a piece from Gartner about how 80% of neoclouds may go belly up essentially by 2030, and that a lot of AI infra spending could be at risk of impairment, which did not help sentiments
- 10/15: Rebound as NVDA promised to shore up CRWV, but the strength lasted only one day
- 10/16: Everything broke down, and this time even hyperscalers, which fared better in the 10/8 and 10/14 selloffs, sold off and started to crack, showing spreading concerns over the entire AI capex chain
Implication: This may be a shift, or the beginning of a shift, in the AI theme investing cycle. Focus is moving away from chip demand, backlog and hopes and dreams in general toward requiring actual proof of profitability. The September hike and rising hike odds from persistent inflation may be the very first trigger of this shift.
The chip weakness is likely a repricing of 2027 and beyond AI-demand visibility, not 2026 earnings. Focusing only on the backlogs and current earnings, especially for the chip companies might no longer be a wise move. Diversified monetizers or hyperscalers should hold up better than pure play companies such as neoclouds and chip makers, but they could still suffer if a broad-based deleveraging happens.
I will write a separate piece about my recent thinking and some maths on the economics of AI capex, focusing on the value chain not the supply chain, in the coming days.
