Nvidia's Networking Revenue Just Grew 263%. The AI Trade Is No Longer Just About GPUs.

Nvidia's networking revenue surged 263% YoY to $11 billion last quarter while data-center revenue grew 75% YoY, underscoring robust AI-infrastructure demand. Arista posted record 2025 revenue of $9 billion (+29% YoY) with AI networking sales of $1.5 billion and management expects this to more than double to $3.2 billion in 2026. Vertiv reported Q4 organic orders +252% YoY, a backlog of $15 billion, and is guiding ~28% organic net sales growth for 2026, though valuation and a potential slowdown in data-center spending remain key risks.

Analysis

The AI fabric shift is becoming a platform race, not just a chip race — hyperscalers and large AI customers are paying a premium for predictable, high‑bandwidth fabrics plus software that reduces ops friction. That creates durable margins for vendors who can sell both silicon and sticky software/maintenance (high attach rates), and it increases the value of companies that own orchestration layers and telemetry engines rather than raw ASICs alone.

A large, underappreciated second‑order effect is the multiplier into physical infrastructure: optics, pluggable transceivers, liquid cooling modules, UPS and transformer upgrades, and modular construction become multi‑year revenue streams with multi‑quarter lead times. Suppliers with constrained capacity will see pricing power and backlog visibility; conversely, any knee in AI capex triggers inventory destocking and a sharp revenue re‑acceleration in the opposite direction.

Key macro and technical catalysts to watch are bookings/backlog trends at hyperscalers, optical lead‑time dispersion, and standardization moves (PCIe/CXL/ethernet RDMA vs proprietary fabrics) over the next 6–24 months. Tail risks that could reverse the current cycle include rapid per‑model efficiency gains (reducing GPU density per cluster), export controls tightening GPU/network silicon flow, or a sudden enterprise/cloud capex pullback.

From a positioning standpoint, this market favors concentrated, conviction bets sized to liquidity and optionality — favor exposure to software‑attached networking and exposed yet optional power/cooling providers while using spreads and pairs to control asymmetric valuation risk.

Key Decisions for Investors

• Long ANET equity (6–12 months), size 1.5–2% portfolio: thesis is software attach and customer consolidation; set a stop-loss at 18% downside and target 30–50% upside if bookings accelerate. Hedge with a 6–9 month 10–15% OTM put if conviction is higher than conviction to hold through volatility.
• Long VRT via 18–30 month call spread (buy LEAP call / sell higher strike) to capture multi‑year power/cooling adoption while capping premium; allocate 1% notional. Reward scenario: 2–4x on spread if backlog converts; risk limited to premium paid if AI capex stalls.
• Pair trade (6–12 months): Long ANET / Short CSCO equal notional (or short legacy switch exposure) — capitalizes on EOS attach and faster refresh cycles. Target asymmetric return: 30%+ on the pair if Arista takes share, with pair stop if both report synchronized order weakness.
• Protective hedge on NVDA: buy 3–6 month puts (or implement collars on existing positions) to guard against a rapid demand shock or regulatory export action. Cost of hedge should be treated as insurance — acceptable if it prevents a >20% tail loss in portfolio NVDA exposure.