I've Held Nvidia for Nearly 3 Years. Here's Whether I'd Still Buy Today.

The article argues Nvidia remains a compelling buy, citing its dominant AI GPU ecosystem, a valuation of roughly 29x trailing earnings and less than 15x next-year earnings. Nvidia projects global data-center capital expenditures could reach $3T-$4T by 2030, while the five largest AI hyperscalers are expected to spend nearly $800B on data centers this year. The bullish thesis is that Nvidia's forward valuation prices in little of the expected multiyear AI infrastructure build-out, though the article is opinion-based rather than reporting a new company disclosure.

Analysis

The relevant question is not whether NVDA's headline forward multiple appears inexpensive, but whether hyperscaler AI capex converts into sustained GPU revenue at current gross margins. The market is increasingly underwriting a shift from training clusters to inference, where utilization, model efficiency, and customer-designed silicon can lower accelerator intensity per unit of AI output. A modest reduction in NVDA's share of incremental spend or a mix shift toward lower-margin networking/enterprise systems could matter more to FY estimates than aggregate data-center capex growth.

Near term, bullish retail commentary is not itself a catalyst and likely reinforces crowded positioning into earnings. Over the next 1-3 months, the stock's direction will hinge on evidence that the next platform ramp is supply-constrained rather than demand-rationed: lead times, cloud-provider capex commentary, and whether NVDA can sustain elevated gross margin while absorbing new-product transition costs. The key downside asymmetry is that a single large customer deferral or lower-than-expected forward revenue visibility can compress the multiple even if absolute growth remains exceptional.

For 6-18 months, the second-order beneficiaries of continued AI infrastructure spending may offer cleaner upside from a lower expectations base: TSMC (TSM) captures leading-edge wafer and advanced-packaging content; Broadcom (AVGO) captures custom accelerator and networking substitution; Vertiv (VRT) and Eaton (ETN) monetize power-density constraints. The contrarian view is that custom ASIC adoption is not necessarily bearish for NVDA immediately—it expands total AI deployment—but it caps the terminal share and margin assumptions embedded in a pure-GPU narrative.

Thesis falsification for a constructive NVDA view: two consecutive quarterly reductions in hyperscaler capex guidance, material deterioration in NVDA data-center gross margin, or disclosed inventory/receivable growth exceeding revenue growth. Conversely, sustained inference demand with stable pricing and rising enterprise contribution would support a longer duration earnings base than the market currently credits.

Key Decisions for Investors

• Maintain NVDA as a core long only on post-earnings volatility or a 10-15% drawdown rather than chase retail-driven strength; target a 6-12 month holding period. Size against the risk that forward gross-margin guidance falls below consensus, which would likely drive multiple compression despite revenue beats.
• Express infrastructure breadth via long TSM and VRT, funded by a partial NVDA underweight rather than an outright NVDA short, over 6-18 months. This preserves AI capex exposure while reducing single-vendor GPU share and product-cycle risk.
• Initiate a 3-6 month relative-value watch: long AVGO / short NVDA only if hyperscalers explicitly raise custom-silicon deployment targets or NVDA's next-quarter data-center outlook decelerates materially. The missing confirmation is customer-level ASIC volume and associated networking attach; absent it, do not execute.
• Ahead of NVDA earnings, hedge concentrated long exposure with downside puts or put spreads dated just beyond results; monetize the hedge if management reaffirms revenue visibility and gross-margin resilience. The objective is protection against guidance-driven gap risk, not a directional bearish call.
• Monitor MSFT, AMZN, GOOGL, META and ORCL capex guidance and AI-service monetization each reporting cycle. A capex slowdown without corresponding AI revenue acceleration is the earliest actionable signal to reduce the entire AI-infrastructure basket.