Nvidia Is Worth $5 Trillion Once Again. Here's Why It Could Become the First $10 Trillion Stock Within the Next 3 Years.

Nvidia is positioned for continued AI-driven growth, with earnings expected to rise 75% in the current fiscal year to $8.34 per share and management now targeting a combined $1 trillion in Blackwell and Vera Rubin data center sales in 2026-2027. The article argues Vera Rubin could cut inference costs by 90% versus Blackwell GB200, supporting stronger demand as AI shifts from training to inference. It also suggests Nvidia could reach a $10 trillion market cap within three years if EPS reaches $13.60 in fiscal 2029 and the stock trades near 33x earnings.

Analysis

The key second-order signal is not “Nvidia stays dominant,” but that the inference cycle is likely to broaden the spend base beyond frontier-model training. As customers optimize for cost per token and latency, the mix shifts toward higher-volume deployment, which should extend the AI capex runway even if training budgets normalize. That is constructive for the whole AI infrastructure stack, but most of the incremental economics accrue to the vendor that can bundle silicon, networking, and software into a switching-cost moat.

The main competitive pressure is on smaller accelerator vendors and custom-chip efforts that rely on the premise that inference is a commodity. If Nvidia’s next architecture truly compresses inference cost by an order of magnitude, many “good enough” alternatives lose the only advantage they had: unit economics. The implication is a likely re-acceleration in hyperscaler procurement, but with a heavier bar for second-source designs that may now be confined to a few hyperscale-specific workloads.

The market is still underpricing the duration of the spend cycle, but it may be overpricing linearity. The risk is not a demand cliff; it is a digestion phase if customers pull forward orders ahead of the next platform transition and then pause to integrate software and deploy inference workloads. On a 3-6 month horizon, any headline on capex moderation, export friction, or customer concentration can compress the multiple faster than earnings can catch up.

The contrarian read is that Nvidia is less a “single-stock AI trade” now and more a toll collector on a broader deployment phase that could favor adjacent platform owners and application-layer beneficiaries. Meta, for example, benefits if inference economics improve enough to expand product-level AI features, but it remains a secondary beneficiary relative to infrastructure. Long-term, the more interesting mispricing may be in names leveraged to AI deployment monetization rather than AI model buildout.

Key Decisions for Investors

• Stay long NVDA on 3-6 month horizon, but use pullbacks to add rather than chase; the setup favors earnings durability, though the risk is multiple compression if capex digestion headlines emerge
• Pair trade: long NVDA / short a basket of smaller AI accelerator or custom-silicon proxies for 1-2 quarters; thesis is that Nvidia’s cost/performance step-up extends share at the expense of second-tier alternatives
• Tactically overweight META over standalone AI infra beneficiaries for 6-12 months; if inference costs fall materially, Meta should monetize the productivity gain through higher engagement and ad load, with less valuation risk than semis
• Avoid paying up for adjacent AI suppliers that depend on a stable training-era mix; if the inference transition accelerates, their pricing power is likely to compress over the next 2-4 quarters
• For option expression, consider NVDA call spreads rather than outright calls over the next earnings cycle to capture upside from continued estimate revisions while limiting damage from any digestion-driven drawdown