Nvidia Goes to Zero. – By JA Westenberg – Selfonomics
13 hours ago
- #Infrastructure Investment
- #Market Cycles
- #Capex Supercycles
- Cisco took 25 years to nominally recover its dot-com peak, yet still trades at half its real 2000 high when adjusted for CPI, illustrating long-term underperformance despite being a crucial infrastructure provider.
- Historical capex supercycles, like railways, electrification, mainframes, telecom, and shale gas, show infrastructure providers often deliver negative real returns post-peak, while customers capture the lasting value.
- NVIDIA trades at high multiples (~40x trailing earnings) with extreme customer concentration; four buyers account for 61% of revenue, resembling late-stage capex bubble dynamics where few large buyers dominate.
- AI hyperscaler capex is projected to hit ~1.2% of GDP in 2026, matching the 2000 telecom peak but compressed into a shorter cycle, with assets like GPUs depreciating rapidly (e.g., H100 outdated in years).
- Competition grows: custom silicon (Google TPUs, Amazon Trainium, Broadcom ASICs) and shifting demand toward inference reduce reliance on NVIDIA, mirroring how white-box switching eroded Cisco's margins.
- A potential tipping point in 2027: mega-customers may moderate capex growth, custom silicon scales, and NVIDIA's earnings multiple could compress from ~25x to 12-15x, triggering algorithmic selling and a 50-70% stock drop.
- Even with sustained revenue growth, de-rating from premium to normal multiples could result in near-zero or negative real returns through 2035, aligning with historical base rates for dominant infrastructure providers post-supercycle.