If I Could Only Buy 1 Artificial Intelligence (AI) Stock, This Would Be It

Nvidia expects 70% revenue growth in fiscal 2028, ending January 2028, versus Wall Street expectations for materially slower expansion. The company says AI hardware demand continues to exceed supply as the five largest AI hyperscalers are projected to spend $1.3 trillion on data-center capex next year, supporting Nvidia's $3 trillion-$4 trillion estimate for annual AI data-center spending by 2030. The article argues Nvidia's 24x forward P/E—roughly in line with the S&P 500's 24.8x trailing P/E—does not reflect its projected growth, implying substantial upside if execution remains strong.

Analysis

The relevant debate is not whether NVDA can grow, but whether incremental AI capex remains GPU-intensive enough to sustain gross-margin and multiple durability as hyperscalers increasingly optimize workloads with internal ASICs. NVDA's software and networking attach rates make its earnings less exposed than a simple accelerator-unit model, but customer concentration creates a nonlinear risk: a pause by even one of MSFT, AMZN, GOOGL, META, or ORCL could produce a near-term inventory correction despite healthy aggregate AI demand. The next 1-3 months should be driven by hyperscaler capex commentary and lead-time data, not long-range TAM claims.

AVGO is the cleaner second-order beneficiary if the market shifts from general-purpose training clusters toward customer-specific inference and internal silicon. That creates a useful relative-value framework: NVDA wins when frontier-model training intensity and networking content rise; AVGO wins when large customers prioritize lower cost per token and ASIC deployment. A broad AI capex slowdown hurts both, but AVGO's custom-chip programs carry greater customer/program concentration and longer qualification cycles, making disappointments more discontinuous.

Consensus appears too willing to annualize an elevated AI infrastructure cycle while treating a market-level earnings multiple as proof of undervaluation. The proper valuation question is whether earnings estimates already embed unusually high utilization, pricing, and supply discipline through the product transition. A sustained reduction in NVDA gross margin, a material deceleration in data-center revenue growth, or hyperscaler capex guidance below current expectations would likely compress the multiple before the longer-term demand thesis is disproven. NFLX has no meaningful read-through and should not be treated as an AI-capex proxy.

Key Decisions for Investors

• Maintain a tactical long NVDA only into the next hyperscaler earnings/capex-readout window; use a 8-10% downside stop or trim if data-center revenue guidance decelerates materially versus consensus. Upside requires both estimate revisions and stable gross margin, making risk/reward favorable only after confirming demand visibility rather than chasing a bullish media narrative.
• Express workload-mix uncertainty through a balanced pair: long AVGO / short equal-dollar NVDA for 3-6 months if evidence builds that inference ASIC deployments are accelerating. Exit if NVDA networking/software attach or frontier-training demand reaccelerates; this pair is not appropriate during a broad AI-capex drawdown because correlation will rise sharply.
• Set alerts around MSFT, META, AMZN, GOOGL, and ORCL capex guidance, accelerator useful-life assumptions, and AI monetization disclosures. Any coordinated moderation in 2027 capex plans is a de-risk signal for both NVDA and AVGO, regardless of management long-term TAM targets.
• Avoid adding long-dated NVDA upside options until implied volatility is compared with the expected earnings move and post-results drift; missing this data makes a directional equity position or defined pair trade preferable to an options recommendation.