Mag 7 starting to underperform [pdf]

Original Article ↗ Hacker News Discussion ↗

Quality and Intent of the Apollo Deck

  • Many find the slide deck shallow: mostly charts, public price tables, and API-doc screenshots, with little real modeling or causal analysis.
  • Some suspect it’s marketing or short-term sentiment shaping rather than neutral research.
  • Others argue Apollo’s scale and “skin in the game” give it more credibility than random online opinions, but AUM is not proof of forecasting skill.

AI Capex, Free Cash Flow, and Hyperscalers

  • Core concern: hyperscalers’ AI capex (hundreds of billions annually, possibly >$1T soon) is consuming or exceeding free cash flow; some are borrowing and issuing equity.
  • Slide on hyperscaler free cash flow shows recent “evaporation,” with Oracle negative and Amazon dropping sharply; omission of Apple is questioned.
  • View 1: this is dangerously capital‑intensive, telecom‑like “massive demand, bad returns.”
  • View 2: some firms (e.g., Meta) could simply cut AI build‑out and quickly restore strong FCF.

Valuation, Bubble Risk, and Market Concentration

  • Several see AI‑driven tech valuations as bubble‑like, with cyclically adjusted P/E near all‑time highs and heavy dependence on a few mega‑caps.
  • Others note that recent Mag7 underperformance over a few months is too short a window to draw firm conclusions.
  • Some think a significant correction is “likely and coming,” potentially magnified by leveraged AI bets and heavy corporate and sovereign debt.

Historical Performance, Mean Reversion, and Diversification

  • Cited research: top past performers tend to underperform the market over the following decade; most individual stocks underperform T‑bills, with a tiny minority driving total wealth creation.
  • Debate over whether this is a true financial effect vs a tautological consequence of averages.
  • Broad theme: concentration in recent winners is risky; diversification (across sectors, geographies, factors) is stressed.

Company‑Specific Themes

  • Apple: relatively low AI capex, huge cash/FCF; seen as both prudently avoiding the arms race and “missing AI.”
  • Google: seen by some as over‑rewarded for AI (higher costs to deliver similar services); others point to strong profits, cloud/AI growth, and diversified “optionality.”
  • Meta: history of large, possibly wasteful bets (metaverse, now AI); concern about hidden/leverage‑like obligations.
  • Oracle: negative FCF and AI‑related debt viewed as a potential weak link.
  • Tesla: some argue it never belonged in Mag7; its TAM narratives are compared to “fusion‑like” promises.

AI Economics and LLM Usage

  • Critique of using OpenRouter token graphs as representative of overall AI usage; direct APIs and cloud marketplaces dominate for leading labs.
  • Concern that if “good enough” open models and cheap inference win, returns to frontier model training and massive capex could disappoint.
  • Counterpoint: hyperscalers still profit as the primary providers of compute and infra, even if model moats narrow.

Index Dynamics, Buybacks, and Flows

  • Discussion on how index funds, dividends, and buybacks might structurally favor large caps, though some argue this effect is overstated in market‑cap‑weighted indices.
  • Observation that constant inflows into index and target‑date funds provide a persistent bid for mega‑caps, possibly amplifying Mag7 moves.

Data Center Build‑Out and Systemic Risk

  • Data centers under construction/planned in the US up ~60% vs existing, with likely larger average size — seen as evidence of a huge AI infrastructure boom.
  • Some fear that when AI capex slows, upstream hardware/memory stocks could crash hard.
  • Unclear whether this build‑out will ultimately be justified by sustainable AI‑driven revenues or prove to be overcapacity.