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

The article argues Nvidia remains a buy despite a nearly 1,500% gain since the start of 2023, citing its dominant AI-GPU ecosystem and expected multiyear data-center buildout. Nvidia estimates global data-center capital expenditures could reach $3 trillion to $4 trillion by 2030, while the five largest AI hyperscalers are projected to spend nearly $800 billion this year. The bullish valuation case rests on Nvidia trading at roughly 29x trailing earnings and below 15x next-year earnings, which the author contends leaves little future growth priced into the shares.

Analysis

This is retail-bullish commentary rather than a new fundamental catalyst, so it should not independently change positioning. The valuation argument hinges on consensus earnings holding or rising; the relevant variable is not the headline forward multiple but whether NVIDIA can preserve accelerator gross margin and revenue per deployed compute as hyperscalers shift a larger share of steady-state inference toward internally designed ASICs. A sustained deceleration in data-center revenue growth without offsetting networking, enterprise, or sovereign demand would produce multiple compression even if absolute capex remains elevated.

The second-order opportunity is in the split between flexible training capacity and lower-cost inference capacity. NVDA remains best positioned where workloads change quickly and software portability matters, while AVGO benefits when customers commit to high-volume, stable workloads that justify custom silicon design cycles. This makes the AI spend cycle less a zero-sum demand question than a mix and pricing question: rising ASIC penetration can expand total compute deployment while reducing the portion of spend addressable by premium general-purpose GPUs.

Over the next 1-3 months, the key catalyst is earnings-estimate revision breadth after hyperscaler capex guidance and NVDA's next results, not another broad AI narrative rerating. Over 6-18 months, power availability, rack deployment lead times, and inference monetization become the binding constraints; if cloud customers cannot translate GPU fleets into external revenue or internal cost savings, procurement discipline will arrive before aggregate AI infrastructure demand disappears. The consensus likely underweights this utilization/ROI risk while overreacting to isolated custom-chip headlines.

Key Decisions for Investors

• Maintain or initiate a measured NVDA long only on post-results evidence that forward data-center revenue estimates are being revised higher; use a 6-12 month horizon. Size for a 15-20% drawdown because the thesis is vulnerable to a gross-margin guide-down or a material reduction in hyperscaler capex plans.
• Express the workload-mix thesis as a 6-18 month pair: long AVGO / short an equal beta-adjusted basket of less differentiated AI hardware exposure, rather than short NVDA outright. AVGO's custom-silicon backlog and incremental ASIC program wins are the required confirmation; exit if ASIC revenue guidance misses or NVDA demonstrates sustained share gains in inference.
• Do not chase near-term NVDA upside on this article alone. Set an event-driven alert for consensus next-twelve-month EPS revisions: upgrade the long if revisions remain positive following the next two major hyperscaler reporting cycles; reduce exposure if revenue growth decelerates materially while gross margin falls, signaling price or mix pressure rather than normal product-transition noise.
• Monitor MSFT, AMZN, GOOGL, META, and ORCL capex guidance alongside AI-service revenue disclosures. A widening gap between infrastructure spend and disclosed monetization is a 1-3 quarter warning signal for procurement rationalization and would favor trimming high-multiple AI infrastructure exposures.