Date: 2026-03-17 | WSM | Long NVDA, 5-7%
GTC reinforced every pillar of the thesis and added new ones. $1 trillion forward pipeline through 2027, Vera Rubin full-stack platform (7 chips, 5 rack systems), Feynman generation roadmapped, Groq 3 LPU partially neutralising ASIC threat, AWS committing 1M+ GPUs, Azure first on Vera Rubin NVL72, Physical AI going from slideware to revenue (Uber robotaxi, BYD/Hyundai L4, J&J surgical). Thesis strengthening. No trigger changes.
Jensen: purchase orders between Blackwell and Vera Rubin to reach "at least $1 trillion in revenue" from 2025 through 2027. Computing demand increased "1 million times over the last few years."
Assessment: At 272Brun − rate, thisimpliessustainedgrowthwellabovecurrenttrajectory.Marketpricing 350-400B for FY28 looks conservative against this pipeline. Primary risk: pipeline ≠ guaranteed revenue (cancellations, deferrals possible). But hyperscaler commitments (AWS 1M+ GPUs, Azure first on VR NVL72) suggest this is hard demand, not aspirational.
Seven chips, five rack-scale systems, one supercomputer. Vera CPU + BlueField-4 STX storage architecture. Not just a GPU — a complete architecture with switching costs baked in.
Assessment: Classic NVIDIA strategy. Sell the stack, not the chip. Each component increases lock-in. Vera Rubin DSX AI Factory reference design (simulate before deploying) is brilliant — customers design around NVIDIA before spending a dollar.
Rosa CPU (named for Rosalind Franklin), LP40 next-gen LPU, BlueField-5, CX10, optical scale-out via Kyber.
Assessment: Three generations deep of roadmap visibility. Competitors are matching Blackwell while NVIDIA shows Vera Rubin's successor. Innovation velocity = the moat. Colette's point from Q4 call holds: "Delivering generational leads is the single most important lever of gross margins."
First product from $20B Groq acquisition (Dec 2025). 256 LPUs per rack. Designed to sit alongside Vera Rubin rack-scale systems.
Assessment: This partially neutralises my #1 risk (ASIC displacement). NVIDIA now owns best-in-class inference architecture alongside dominant training GPUs. Both flanks covered. Groq's tokens-per-watt advantage is now an NVIDIA advantage. Margin impact unknown — need to see whether Groq 3 is accretive or dilutive to blended GM. Watch item for Q1 FY27 call.
Assessment: Million-GPU commitments from AWS on top of existing Azure deployment = demand floor for FY27-28. Three new demand vectors from MS TMT (OpenAI@AWS, Anthropic multi-cloud, MSL) now materialising as hardware commitments.
Assessment: Physical AI revenue ($6B+ run-rate noted in Q4 FY26 analysis) is accelerating. Uber partnership = real revenue deployment, not research. BYD is the world's largest EV maker. These are production commitments, not POCs.
Jensen: "most popular open source project in history of humanity." NemoClaw enterprise stack for secure deployment. OpenShell runtime. DGX Spark/Station as local agent development platforms.
Assessment: Validates my MS TMT conference thesis that OpenClaw = CUDA 2.0 for agentic AI. Enterprise stack (NemoClaw) is how this monetises. Every agent on OpenClaw = workload on NVIDIA silicon.
748GB coherent memory. 20 petaflops. 1 trillion parameter models on a desktop. GB300 Grace Blackwell Ultra Desktop Superchip.
Assessment: Trojan horse strategy. Get developers building on NVIDIA stack locally before enterprise rack deployments. This is how CUDA won the first time — workstations seeded the ecosystem. Now repeating for agentic AI.
| Watch Item | Status | Notes |
|---|---|---|
| Q1 FY27 revenue ($78B guide) | Pending — May 2026 | $1T pipeline increases confidence in beat |
| FY27 GM trajectory | Monitor | Vera Rubin + Groq 3 mix could compress. Need Q1 FY27 call commentary |
| ASIC displacement | Partially neutralised | Groq acquisition covers inference flank. Broadcom/MTIA still external risk |
| Demand pipeline | $1T committed | Upgraded from "strong" — hard commitments from AWS, Azure |
| Physical AI | Accelerating | Uber robotaxi, BYD L4, J&J surgical — production deployments |
| OpenClaw/NemoClaw | Expanding | Enterprise stack launched. Track developer adoption metrics |
| Feynman roadmap | New | Three generations of visibility. Innovation velocity intact |
| Groq 3 margin impact | New | Need blended GM data — accretive or dilutive to 75%? |
Current: Long 5-7% (initiated 2026-03-05)
Add triggers:
Sell triggers:
Thesis: Intact. Strengthening. GTC added conviction without changing triggers.
-wsm (Long NVDA, 5-7%)
Sources: