I've Changed My Mind on AMD. The AI Supercycle Has Room for More Than Just Nvidia.

AMD's Q2 data-center revenue rose 107% year over year to $6.7 billion, helping total revenue increase 50% to a record $11.5 billion as AI infrastructure demand expands. Management expects the data-center CPU TAM to reach $220 billion by 2030 and forecasts server-CPU revenue growth of 70% in 2027, driven by a shift in GPU-to-CPU server configurations toward 1:1 from 4:1 to 8:1 historically. The growth outlook is offset by valuation risk: AMD trades at roughly 40x forward earnings versus a technology-sector average near 21x after a 284% one-year share-price gain.

Analysis

The investable implication is less “CPUs displace GPUs” than a rising compute-per-rack and power-management opportunity. Agentic inference creates more latency-sensitive orchestration, retrieval, and tool-calling work, but that can be served by AMD EPYC, Intel Xeon, ARM-based hyperscaler silicon, or offload architectures; AMD must demonstrate share gains, not merely category growth. The cleanest second-order beneficiaries are memory and networking suppliers—MU, AVGO, MRVL and ANET—because more heterogeneous racks raise DRAM, interconnect, and switching content regardless of CPU vendor.

AMD’s valuation leaves little room for a normalizing growth rate or gross-margin dilution from a CPU-heavy mix. The key earnings question over the next 1-3 months is whether data-center growth converts into accelerating EPYC unit share and gross margin, rather than lower-priced CPU volume masking a less favorable accelerator mix. Treat management’s long-dated TAM and revenue-growth framing as directional until hyperscaler design wins, server OEM backlog, and segment-level mix data validate it.

Consensus may be underestimating Intel’s optionality if the CPU demand upswing broadens beyond AMD’s installed base. INTC’s server franchise is impaired but operating leverage is substantial: even modest Xeon stabilization can constrain AMD’s pricing and reduce the share-capture narrative that supports its premium multiple. Over 6-18 months, custom ARM silicon from AMZN, GOOGL, and MSFT is the more structural risk to merchant x86 CPU TAM; agentic workloads do not inherently require x86.

The near-term risk/reward is therefore asymmetric after a strong AMD move: positive AI-capex prints can support the stock, but any guide that implies weaker accelerator supply, declining data-center gross margin, or slower EPYC share gains could trigger multiple compression disproportionate to an otherwise solid revenue result. Falsify the cautious view with two consecutive quarters of data-center growth above guidance alongside expanding gross margin and explicit hyperscaler CPU share gains.

Key Decisions for Investors

• Do not chase AMD outright ahead of the next earnings print; initiate only on a 10-15% pullback or after verified guidance implies sustained data-center growth with gross-margin expansion. Position sizing should reflect that valuation, rather than revenue growth, is the principal downside risk over 1-3 months.
• Express heterogeneous-AI-rack growth through a basket long MU, ANET, and MRVL rather than AMD alone over 6-12 months. These names monetize memory and network intensity across AMD, NVDA, Intel, and custom-silicon deployments; exit if hyperscaler capex guidance turns negative or networking lead times normalize sharply.
• Monitor a tactical long AMD / short INTC pair only after Intel reports continued server-share losses and AMD confirms EPYC share gains. Avoid entering on CPU-TAM headlines alone; the pair is vulnerable to a Xeon product-cycle recovery or government/foundry support improving INTC sentiment.
• For existing AMD longs, buy 3-6 month downside puts or trim into strength before earnings if implied volatility is not already elevated. The hedge is justified by the risk that a CPU-led mix shift supports revenue but disappoints on gross-margin progression.