Jensen just put Nvidia’s Blackwell and Vera Rubin sales projections into the $1 trillion stratosphere

Nvidia CEO Jensen Huang projected $1 trillion of orders for Blackwell and Rubin chips through 2027, up from about $500 billion in demand cited through 2026. Rubin is touted as the next-gen architecture at 3.5x training and 5x inference performance versus Blackwell, reaching up to 50 petaflops, with production ramp expected in H2. The large demand projection implies significant upside to Nvidia's AI hardware revenue and is broadly constructive for Nvidia and the AI/semiconductor sector.

Analysis

The hardware demand trajectory implied by recent guidance compresses a classic bottleneck: compute demand is outstripping advanced packaging, HBM supply and EUV-capable wafer slots. Expect a multi-quarter cadence where pricing power accrues to the scarce nodes in the supply chain (advanced foundry slots, CoWoS/EMIB packaging and HBM stacks) rather than to commodity memory makers; that will magnify margin dispersion across suppliers and create outsized FCF for those with capacity in the next 6–18 months.

Cloud customers will react asymmetrically: hyperscalers that locked supply early will preserve gross margins, while others will either pay up for allocation or re-architect workloads toward cheaper inference options — a dynamic that can transiently elevate cloud instance pricing and materially compress SaaS unit economics for AI-first vendors. This bifurcation creates a two-tier market for AI services over the next 3–9 months where access to premium accelerators confers a sustained competitive moat.

Key reversal risks are concentrated and fast-moving: a foundry yield slip, a sudden HBM capacity add, or a model-efficiency breakthrough that reduces required compute per token can shave off consensus upside in a single quarter. Geopolitical/export-control developments and customer channel inventory behavior (pull-forward followed by a normalization hangover) create a 'double-hump' revenue pattern risk over 6–24 months.

Consensus is pricing a near-perfect supply ramp and 100% conversion of demand into recognized revenue; that is asymmetric. The smarter exposure is to isolate capacity winners and play relative share shifts rather than owning the headline hardware name outright without hedges, because order backlog is not the same as near-term cashflow and can flip if ramp or customer pacing changes.

Key Decisions for Investors

• Long NVDA via a 6-month call spread: buy 15% OTM calls and sell 40% OTM calls, size 1.5% of portfolio. Rationale: captures re-rate if sell-through and ASPs surprise on the upside while limiting premium paid; target 3x+ return if NVDA re-prices within 6 months. Risk: capped upside and full premium loss if sentiment stalls.
• Buy TSM (TSM) or ASML (ASML) 12-month exposure (equity or long-dated calls), size 1–2%: these are direct beneficiaries of foundry/ASML capacity tightness and enjoy more stable cashflow conversion. Target 20–30% upside in 12 months if capacity remains constrained; downside is regulatory/China demand shock or cyclical capex pullback.
• Relative pair: long NVDA / short AMD (AMD) equal notional, 12-month horizon. Rationale: isolates Nvidia’s share-gain and pricing power in premium accelerators while hedging sector beta. Target relative outperformance of 200–400 bps; risk is product/performance surprise from AMD or underestimated AMD supply response.
• Protective hedge for AI exposure: buy 3–6 month put spreads on major cloud providers (AMZN, GOOGL) or purchase NVDA 3-month puts if long stock — size 0.5–1% as insurance. Rationale: hedges a sudden margin squeeze from higher instance pricing or a demand softening event; cost limited via put spreads.