AMD Is Trading at 97 Times Forward Earnings While Nvidia Trades at 21 Times -- Someone's Valuation Model Is Wrong. Here's Who I Think It Is.

Nvidia and AMD both remain positioned to benefit from AI-driven data center demand, but the article argues Nvidia is still undervalued at 21x projected fiscal 2027 EPS versus AMD at 97x forward earnings. Nvidia’s revenue grew 85% year over year in Q1 fiscal 2027 to $81.6 billion, while AMD’s Q1 2026 revenue rose 38% to $10.25 billion, both modestly beating consensus. The piece is bullish on Nvidia relative to AMD, citing repeated analyst underestimation of Nvidia and higher execution risk embedded in AMD’s valuation.

Analysis

The market is still pricing NVDA and AMD as if their end-markets are independent, when the bigger second-order effect is that they are both becoming a proxy for hyperscaler capex intensity. If AI infrastructure spend elongates beyond the current buildout cycle, NVDA keeps the operating leverage, but AMD becomes the cleaner “beta with optionality” beneficiary because it can take share on price-performance while Nvidia monetizes the ecosystem. That makes the real loser not Intel in headline share terms, but any adjacent supplier or customer segment with fixed budgets, because every dollar diverted to AI accelerators is a dollar not flowing to legacy server refresh, networking, or enterprise software budgets.

The consensus miss is that “cheap” for NVDA and “expensive” for AMD are both incomplete without estimating estimate revision velocity. NVDA’s multiple can stay deceptively low if earnings are still being revised up every quarter; the real issue is not valuation compression but duration risk if the cadence of beats slows. AMD, by contrast, is priced for a much cleaner share-gain trajectory than the competitive landscape likely allows; the market is implicitly underwriting both material CUDA substitution and meaningful server CPU share gains, which is a high bar given switching costs, software lock-in, and customer qualification cycles that usually run in multi-year rather than quarterly increments.

Near term, the catalyst path is asymmetric: NVDA has more upside if the next platform transition re-accelerates orders, while AMD has more downside if MI adoption proves lumpy or if server CPU gains stall. The biggest tail risk for both is not demand, but supply normalization and capex digestion: if hyperscalers finish front-loading infrastructure, the market can rapidly re-rate the group from “scarcity growth” to “cyclical growth” over 3-6 months. In that scenario, AMD’s premium multiple is much more vulnerable than NVDA’s, because it has less margin of safety and less evidence of sustained economic moat capture.

Key Decisions for Investors

• Long NVDA / short AMD on a 6-12 month horizon; the trade benefits if hyperscaler capex remains strong but share gains disappoint. Risk/reward favors NVDA because it has revision momentum and a lower earnings multiple, while AMD’s premium valuation is highly sensitive to any execution miss.
• Buy NVDA on pullbacks after earnings or platform-update commentary, not into strength; use 3-6 month horizons. Best entry is on any post-print digestion where the market focuses on guide quality rather than top-line size. Upside comes from repeated upward revisions, not multiple expansion.
• Use AMD puts or put spreads for 3-6 months as a hedge against AI semis exposure. The cleanest thesis is that the stock already discounts too much MI400/MI500 share capture and server CPU progress; downside accelerates if qualification cycles slip or gross margin mix disappoints.
• For a lower-risk expression, pair long NVDA with short INTC over 6-12 months. This isolates AI infrastructure winners versus legacy server CPU underpressure, with NVDA’s moat and software lock-in offering better earnings durability than a cyclical recovery bet.
• If already long both names, trim AMD into strength and keep NVDA as the core holding. The relative risk/reward is materially better in NVDA because the market can continue to underestimate duration, whereas AMD needs a near-perfect adoption curve to justify its multiple.