Forget Nvidia

AMD is up about 100% YTD, supported by Q1 revenue growth of 38% and a 57% surge in Data Center revenue. The article argues the next leg could come from agentic AI, which may shift the GPU-to-CPU ratio from 1:6 toward 1:1 and materially expand demand for AMD's EPYC CPUs. Lisa Su also doubled the server CPU TAM outlook to 35%+ annual growth, targeting a $120B market by 2030, with AMD server CPU revenue expected to grow 70%+.

Analysis

The market is still pricing AMD as a cyclical GPU beneficiary, but the more important setup is a mix-shift in compute architecture that could make CPUs the next scarcity node. If agentic workloads proliferate as inference-heavy, memory- and orchestration-intensive systems, the bottleneck shifts from raw FLOPs to thread scheduling, data movement, and system coordination — areas where server CPUs can reclaim pricing power. That matters because it extends AMD’s growth runway beyond a single product cycle and broadens the profit pool from accelerators into the higher-volume base-layer of the data center stack.

Second-order winners include memory, networking, and server ODMs that participate in denser rack-level deployments, while the relative losers are companies whose roadmap assumes GPU-only intensity or whose CPU franchises are structurally weaker. Intel is the obvious competitive pressure point: if the enterprise refresh cycle accelerates around agentic inference rather than training, share loss can compound faster because CPU attach rates tend to follow platform wins for several quarters, not weeks. Watch for spillover into HBM-adjacent and Ethernet/optical names as architecture shifts increase total nodes deployed, even if GPU unit growth normalizes.

The risk is not demand disappearance but expectation compression. AMD now needs to convert a narrative upgrade into sustained server share gains and margin durability over the next 2-3 quarters; any pause in hyperscaler capex, mix dilution from lower-ASP products, or signs that customers are experimenting rather than standardizing would hit the stock hard. The move is likely underappreciated if investors still anchor to GPU multiples, but it becomes overdone if the market extrapolates 70%+ server CPU growth through 2030 without asking whether competitive response or supply constraints cap near-term realization.

Key Decisions for Investors

• Stay long AMD on a 3-6 month horizon, but size it as a re-rating trade rather than a straight-line growth compounder; risk/reward is best on pullbacks toward prior breakout levels, with a thesis that CPU mix can support the multiple even if GPU enthusiasm cools.
• Pair trade: long AMD / short INTC for the next 2-4 quarters to express share-shift in server CPUs; the trade works if platform wins translate into follow-on enterprise refresh, but the short leg should be kept tight because policy/turnaround headlines can squeeze.
• Buy AMD Jan-2026 call spreads to capture the market underpricing of the CPU TAM expansion while limiting downside if hyperscaler spending pauses; best entry is after any post-earnings consolidation rather than into momentum highs.
• Add a basket long in networking/memory beneficiaries versus semiconductor index if data center rack density rises faster than unit GPU demand; this expresses the second-order buildout without betting solely on AMD execution.
• Trim or hedge AMD if channel checks show agentic adoption remains pilot-stage over the next 1-2 quarters, because the stock will likely discount the future well before the revenue inflects.