Nvidia Will Still Beat AMD Through 2028. Here's the Data Behind My Conviction.

Nvidia is positioned to remain ahead of AMD through 2028, supported by a 90% discrete-GPU market share in Q2 2026, CUDA software lock-in, and data-center revenue that rose 117% year over year to 92% of sales. Analysts project Nvidia's fiscal 2026-2029 revenue and EPS CAGR at 60%, versus AMD revenue and EPS CAGR of 52% and 95%, respectively, although AMD trades at a substantially higher 86x current-year earnings multiple versus Nvidia's 21x. The article argues AMD must execute on Helios rack-scale systems and MI450 deployments to gain meaningful share from Nvidia.

Analysis

The relevant competitive question is not discrete-GPU share but whether hyperscaler capex migrates from merchant accelerators toward internal ASICs and inference-optimized systems. NVDA is best positioned to preserve economics when customers demand full rack-scale deployment, networking, software support and rapid model-training time; this shifts competition from chip benchmark performance to platform integration. The principal second-order beneficiaries are NVDA ecosystem suppliers such as AVGO (networking/custom silicon), VRT (data-center power/cooling) and TSM, while AMD must win sufficiently large deployments to overcome customer switching and software-validation costs.

AMD's earnings sensitivity is asymmetric: incremental accelerator revenue can drive substantial operating leverage, but its valuation leaves little tolerance for a delayed rack-scale ramp, lower-than-expected gross margin, or customer concentration. NVDA's near-term risk is less AMD share loss than digestion of hyperscaler capacity after an exceptionally elevated capex cycle; a sequential deceleration in cloud capex or a reduction in lead times would pressure its premium multiple within days. Over 1-3 months, earnings guidance and disclosed supply commitments matter more than broad AI enthusiasm; over 6-18 months, inference mix, custom ASIC adoption and export restrictions determine whether NVDA's platform economics persist.

Consensus may be too focused on AMD taking share directly from NVDA. The more plausible path is bifurcation: NVDA retains frontier training and high-utilization clusters, while AMD competes for price-sensitive inference and second-source deployments—an outcome that can grow AMD revenue but still disappoint investors expecting rapid margin convergence. The article's valuation and growth assertions should not be relied upon without current consensus estimates, accelerator revenue mix, and customer-level deployment data.

Key Decisions for Investors

• Maintain or initiate a 6-12 month long NVDA / short AMD dollar-neutral pair on relative-strength pullbacks, rather than outright AMD short exposure. Thesis: NVDA's integrated systems and software attachment protect gross margin while AMD bears ramp-execution risk. Cover the pair if AMD reports material accelerator revenue upside with stable/improving gross margin, or NVDA guides data-center revenue materially below cloud-capex growth.
• Use AMD earnings as the near-term catalyst: avoid adding long exposure ahead of results unless management quantifies MI450/Helios backlog, customer deployments, delivery timing and expected gross-margin contribution. A qualitative AI demand narrative without these metrics is an alert, not a catalyst sufficient to support its multiple.
• For structural AI exposure, prefer a basket of NVDA, AVGO and VRT over a pure GPU-share trade for the next 6-18 months. This captures networking and power-density spend even if accelerator share shifts; reduce if hyperscaler capex guidance turns negative sequentially or data-center order lead times normalize sharply.
• Do not establish an INTC trade from this signal alone. Its potential relevance is as a CPU pricing and foundry competitor, but the missing variables are server-CPU share trends, foundry customer traction and capital-intensity guidance.