My read: the ceiling is not compute demand, it is supply chain throughput.
For NVIDIA: • Hyperscalers are shifting from pilot spending to infrastructure-scale deployment. This is a multi-year order book, not a one-quarter burst.
• Industrial AI demand is broadening beyond cloud. Energy, manufacturing, simulation and digital twins are now meaningful buyers, which widens NVIDIA’s TAM materially.
• The true bottlenecks are HBM memory, advanced packaging (CoWoS), power, and datacentre buildouts, not customer appetite.
NVIDIA roadmap: • Near term: $260 to $280 if next guidance lifts again
• Bull case: $300+ becomes realistic if Rubin ramp + networking attach rates remain strong
• Ceiling? Still unclear. Demand looks capacity-constrained, not end-market-constrained.
For TSMC: • The US$52B to US$56B capex is effectively management saying AI demand is durable.
• If TSMC solves packaging and keeps 2nm / advanced-node lead, it strengthens its moat.
• Foundry dominance remains intact because rivals still lag on scale, yield and ecosystem.
The bigger risk: If everyone enters “hyperdrive”, shortages in HBM + packaging + electricity may become the hard ceiling before chip demand does.
In short: **NVIDIA’s ceiling is supply, TSMC’s ceiling is execution. Both ceilings are still moving higher.**
Comments