They wrote off the entire Chinese data center market and still guided $78 billion.
Thesis: NVIDIA transformed from a silicon vendor to a systemic bottleneck. They own the entire data center architecture through NVLink and CUDA. The street fixates on custom silicon. They miss that 50x performance per watt gains on Blackwell inference lock hyperscalers into the ecosystem. This is a sustained pricing power monopoly masquerading as a hardware cycle.
Verdict: LONG — Conviction: HIGH
Catalyst: Vera Rubin production shipments in the second half of 2026 demonstrating 10x inference cost reduction without gross margin compression.
Key Risk: TSMC CoWoS-L packaging capacity and physical data center power limits creating a hard ceiling on revenue growth regardless of end-market demand.
The Tell: Colette Kress admitted "it's too early yet to determine how much" Vera Rubin will ramp in the second half. This directly contrasts Jensen's absolute certainty that "every single customer" will buy it. The transition cadence remains an execution risk.
Friction Level: MODERATE_FRICTION — The sustainability of hyperscaler CapEx. The bull sees a structural shift where compute equals revenue via inference monetization. The bear sees a cyclical top masked as secular growth, bounded by TSMC packaging and power limits.
Report not found
The report data is no longer available. Please return to the archive.