Had You Invested in These 2 AI Infrastructure Winners 10 Years Ago, Here's What You'd Have Now

Arista reported FY2025 revenue of $9.006B, up 28.6% YoY, and has shipped 150M cumulative ports; ANET has delivered returns of +96.1% (1Y), +556.2% (5Y) and +3,007.2% (10Y). Comfort Systems posted FY2025 net income surpassing $1.0B (+95.73% YoY) and a record year-end backlog of $11.94B (roughly doubled), with returns of +385.1% (1Y), +1,855.7% (5Y) and +4,595.4% (10Y). Key valuation/risk points: Arista trades at ~36x forward P/E and faces export/tariff or hyperscaler capex pause risk; Comfort trades at ~47x forward P/E, raised its quarterly dividend to $0.70, and risks backlog conversion pressure from labor/materials inflation.

Analysis

Two structurally different ways to play AI infrastructure are visible: one is a high-margin, software-attach networking platform whose value is realized conditional on sustained hyperscaler GPU cluster rollouts; the other is a high-visibility, project-driven contractor whose earnings are front-loaded by booked work. That means the second has higher near-term cash-flow visibility but greater sensitivity to input-cost inflation and labor markets, while the first has greater binary execution risk tied to shipment cadence, product cycles, and geopolitically driven supply constraints.

Second-order winners if the AI build continues include silicon suppliers that sit below Arista in the stack (Broadcom/Marvell-like ASIC players) and specialty prefab/modular OEMs that could win repeat orders from hyperscalers; losers include commodity subcontractors and regional HVAC/electrical firms that lack scale and pricing leverage. Crucially, a sustained squeeze on high-speed switch exports or a hyperscaler capex pause would accelerate verticalization (in-house switching fabrics) and commoditize market share, compressing multiples across the networking cohort over 12–36 months.

Short-term catalysts to watch are quarterly shipments and a rolling 12-month bookings conversion metric for the contractor; medium-term drivers are campus-network software monetization and longer-term secular outcomes hinge on architecture choices (open networking vs proprietary interconnects). The market has priced near-perfection into shares; that creates asymmetric opportunity to express convex views via defined-risk option structures and inter-company pairs that isolate execution vs backlog risk over 3–18 month windows.

Key Decisions for Investors

• Long FIX equity (3–12 months): size as a tactical overweight to capture backlog conversion and dividend carry. Target asymmetric return of 30–50% if conversion and pricing hold; downside ~25–35% if project starts slow or fixed-price margins compress. Use a 10–15% trailing stop or buy 6–9 month 20–25% OTM protective puts to cap downside.
• Long ANET via 12–24 month call spread (buy 12–24 month ATM calls, sell higher strike): play software attach and campus expansion while capping premium paid. Reward: target 50–100% on spread if AI networking demand continues; defined loss = net premium. Avoid outright long stock exposure through earnings windows — earnings have produced post-beat selloffs historically.
• Pair trade (12 months): long FIX / short ANET (equal notional) to isolate construction/backlog risk vs execution/supply-chain premium. This hedges macro capex cyclicality; if data center starts cool, both fall but ANET should fall less if software recurring revenue cushions — monitor for divergence and rebalance at 10% relative move.
• Event hedge (days–weeks around geopolitics): buy inexpensive 60–90 day puts on high-end switch supply chain proxies (ANET or a Broadcom-like ETF proxy) if new export control headlines emerge. Tail protection cost is small relative to correlated drawdown risk; unwind if no policy change within 30 days.