The AI Bubble Is About to Burst and Here's Why
Overall Thesis
AI has been in a self-reinforcing bubble for 3 years; we're past the top. Hyperscaler CAPEX ramp (Alphabet going $30B→$90B→$180B 2022-2026; Amazon expected $200B 2026) is now forcing debt issuance (Oracle $110B+ debt, ~2x op-cash-flow on capex). Market already rejecting: Oracle -25% QoQ, Microsoft -21% YTD. Expect AI infrastructure spending pullbacks starting late-2026 into 2027, then margin and multiple compression on chipmakers. AI beneficiary stocks could drop 50-80%+ — including Nvidia (~70% downside if margins revert to 2020 levels).
Narratives
Nvidia at a 60%+ operating margin and mid-teens price-to-sales is at the top of a cyclical peak. Historical margins run 15-20% with periodic dips negative. If hyperscaler AI CAPEX slows late-2026/2027 (already signs: Oracle -25% QoQ, debt load $110B+), Nvidia faces simultaneous margin compression (60% → 20%) and multiple compression (P/S reverting to 2020 levels). Combined, that's ~70% downside from today. Cyclical sales history: negative revenue growth in 2010, 2014, 2020 — not unprecedented.
Key Arguments
- Op margin 60%+ today vs historical 15-20% band — unsustainable peak
- Hyperscaler CAPEX unsustainable: Alphabet $30B→$180B (2022→2026), Amazon $200B expected for 2026
- Oracle already 2x operating-cash-flow in capex; $110B+ debt — the canary
- Market already rejecting: ORCL -25% QoQ, MSFT -21% YTD, AMZN pulling back
- Debt-funded capex historical parallel: dot-com bubble burst when debt carrying became untenable
Predictions (1)
Oracle is the canary in the AI CAPEX coal mine — spending roughly twice operating cash flow on capex and carrying $110B+ in debt. Stock already -25% QoQ. If AI infrastructure spending slows, Oracle takes disproportionate pain.
Key Arguments
- Capex ~2x operating cash flow — unsustainable on current cash generation
- $110B+ debt load and growing
- Already -25% QoQ — the market is sending an early signal
- Highest leverage to AI-infrastructure-slow scenario
Hedges & Caveats
- Timing is uncertain — 'next year or two' / 'in the next few years'
- Not that these companies go out of business — historical dot-com parallel had survivors
- Trough of disillusionment is the entry point, not current levels