NVIDIA GTC 2026 — Vera Rubin Era Begins
Scenario Comparison
Vera Rubin VR200 NVL72 sampling confirmed for H2 2026, volume production Q1 2027
Jensen cites strong hyperscaler capex commitments; demand exceeding supply through 2027
Vera Rubin delayed to H1 2027 due to yield or packaging issues
All Scenarios
7Positioning Suggestions
→ AMD: Consider trimming 3-5% before keynote if VR200 on-schedule probability stays >60%. MI450 window narrowing is the primary bear catalyst from GTC.
→ ANET: Asymmetric upside if Ultra Ethernet explicitly validated. Consider small add (2-3%) as a GTC-contingent position. Risk is InfiniBand narrative strengthening.
→ APLD: Hold — GTC is a demand signal event, not an APLD-specific catalyst. Strong demand language helps but is not trade-actionable.
→ MU: Hold into GTC — HBM4 demand validation is positive but likely already in consensus. More important catalyst is MU's own earnings (March 18).
→ NVDA: No action recommended pre-event. Thesis already prices in base case; GTC upside/downside roughly symmetric.
Predictions
10NVIDIA will confirm Vera Rubin VR200 sampling in H2 2026 with volume production starting Q1 2027
If confirmed → consider trimming AMD position 3-5% (MI450 competitive window narrows)
Jensen will explicitly preview the Feynman architecture (2028) with at least one technical detail (process node or silicon photonics)
If confirmed → positive for TSM (1.6nm validation) and ASML (High-NA EUV demand)
NVIDIA will announce a new dedicated inference accelerator product (separate from Vera Rubin compute GPU)
If confirmed → negative for custom ASIC players (AVGO, MRVL custom silicon business)
Jensen will cite hyperscaler AI capex commitments exceeding $300B for 2026 (up from $200B+ in 2025)
If confirmed → positive for APLD (data center demand validated) and AMAT (equipment demand extends)
Ultra Ethernet Consortium will receive explicit endorsement or product announcement from NVIDIA at GTC
If confirmed → add ANET 2-3% (Ethernet for AI back-end validated by NVIDIA itself)
NVDA stock will move more than 5% (either direction) on the keynote day (March 17)
Jensen will announce a China-specific product (H20 successor or new export-compliant chip)
If confirmed → watch for export control reaction from US government in following weeks
At least one major customer (hyperscaler CEO or CTO) will appear on stage to announce a Vera Rubin deployment commitment
If confirmed → positive for NVDA demand visibility; check if customer is Meta/OpenAI (impacts AMD warrant pipeline)
NVIDIA will announce NVLink 7.0 or next-gen interconnect that explicitly claims superiority over Ethernet for AI training
If confirmed → negative for ANET (-3-5% expected); monitor hyperscaler reaction before acting
AMD stock will decline more than 2% on GTC keynote day (March 17) due to Vera Rubin competitive pressure
Key Questions
- Vera Rubin VR200 sampling timeline — is it H2 2026 as expected, or any hint of delay?
- Inference chip — is it a new product line or an extension of Blackwell?
- Feynman tease — any silicon photonics details or TSMC process confirmation?
- Ethernet vs InfiniBand — does Jensen explicitly endorse one for AI back-end?
- Customer deployment numbers — how many GW of AI compute are committed for 2027?
- Custom ASIC competitive commentary — any direct response to Google TPU/Amazon Trainium?
- China market — any H20 successor or new export-compliant product announcement?
- Software ecosystem — OpenClaw, CUDA roadmap, any developer adoption metrics?