Asymmetric Investing by Travis Hoium“So, what if that operating margin goes back down to where it was in 2020? What if the price to sales multiple goes back to where it was in 2020? The stock would be down 70% from where it is today.”@ 8:55 · open at this moment on YouTube ↗
Thesis at the time
Strongly BearishNvidia 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
Counter-arguments acknowledged
- Timing uncertain — bubble could inflate further before bursting
- NVDA has had cyclical drawdowns before and recovered
- Could be 'next year or two' — not imminent
Hedges and caveats (from the video)
- 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
The call
- Date said
- Apr 15, 2026
- Timeframe
- next year or two as AI CAPEX slows
- Deadline
- Apr 15, 2028
- Price at prediction
- $198.87
- Confidence
- medium
- Specificity
- specific
How it resolved
- Status
- unverifiable
Why this resolved this way(resolution audit)
Full rules: docs/resolution-spec.md.
Confidence Reasoning
Specific magnitude (~70%) with a clear mechanism (margin + multiple mean reversion), but the trigger timing is explicitly uncertain ('timing is really uncertain'). Framed conditionally ('what if...') which softens to Tier 3, but the bear-case framework is the central thesis.