I Know It Sounds Crazy, But I Keep Buying This Stock

NVIDIA reported Q4 FY2026 revenue of $68.13 billion, up 73.2% year over year, with full-year revenue of $215.94 billion, non-GAAP gross margin of 75.2%, and free cash flow of $96.58 billion. Management guided Q1 FY2027 revenue to roughly $78 billion even with China data center compute revenue at zero, while the company returned $41.1 billion to shareholders and still has $58.5 billion authorized for buybacks. The main risk is export controls and lost China H20 sales, but the article argues AI demand, Blackwell/Vera Rubin adoption, and strong valuation metrics keep the long-term thesis intact.

Analysis

This is less a single-name earnings story than a claims-on-capacity trade across the AI supply stack. The market’s mistake is to anchor on NVDA as hardware beta, when the real economic moat is in system-level scarcity: networking, software lock-in, and procurement priority across hyperscalers. That means the second-order winners are the companies that help NVDA turn silicon into deployable clusters, while the losers are any OEMs and accelerator alternatives that depend on a normalization in supply or pricing to gain share.

The biggest near-term risk is not demand rollover; it is digestion. When a stock compounds this fast on already-strong expectations, the next leg up requires either faster revenue inflection or multiple expansion, and neither is guaranteed. A China reset remains the cleanest downside catalyst because it is asymmetric: it can compress sentiment quickly even if the core ex-China franchise stays intact. More subtle is the possibility that capex intensity across hyperscalers peaks before revenue monetization does, creating a temporary air pocket for adjacent names tied to buildout pace.

The contrarian miss is that the trade may be under-discounting how much of the AI value chain is becoming a landlord model. If NVDA is the toll bridge, then customers are increasingly forced into long-duration commitments, which reduces cyclicality and supports premium valuation persistence. That also argues for relative trades rather than outright shorts: the path of least resistance is likely continued outperformance of the most supply-constrained names versus the broader semis and infrastructure basket, not a clean top in the AI complex.

Key Decisions for Investors

• Stay long NVDA on pullbacks over the next 1-3 months, but express it with call spreads rather than stock to avoid multiple-compression risk; target 1.5x-2.0x upside if guidance remains above the sell-side’s implied capex curve.
• Pair trade: long NVDA / short a broad semiconductor ETF or weaker AI-enablers for 6-12 weeks to isolate the scarcity premium; expect 300-500 bps relative outperformance if AI capex remains disciplined.
• Add to TSM only on weakness and only as a hedge against supply-chain bottlenecks; if advanced-node allocation tightens further, TSM benefits, but it is a lower-beta way to express the same demand thesis.
• Use CRWV as a higher-volatility satellite long for 3-6 months if you want leverage to AI factory buildout; size small because any slowdown in hyperscaler leasing or financing will hit it first.
• Avoid chasing AMZN/META here unless the goal is to own AI spend monetization, not infrastructure scarcity; their upside depends on translation of capex into revenue, which is a slower catalyst than NVDA’s supply-led rerating.