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

AMD expects its data-center CPU total addressable market to reach $220 billion by 2030 as AI-agent workloads drive server CPU/GPU configurations toward a 1:1 ratio from prior GPU-heavy mixes of 4:1 to 8:1. The company reported a 107% increase in second-quarter data-center revenue to $6.7 billion and 50% total-sales growth to a record $11.5 billion, while forecasting 70% CPU-server revenue growth in 2027. The outlook is tempered by AMD's 40x forward P/E versus roughly 21x for the technology sector and a 284% share-price gain over the past year.

Analysis

The investable implication is not a zero-sum GPU-to-CPU rotation: higher CPU content per AI cluster expands platform spend and favors AMD’s ability to bundle EPYC CPUs with Instinct accelerators. The more important competitive casualty is Intel (INTC), whose server CPU installed base is increasingly vulnerable where hyperscalers value power efficiency and a single-vendor validation path; TSMC (TSM) is a quieter beneficiary through higher leading-edge wafer intensity. NVDA is not directly impaired unless CPU attach materially displaces GPU budget, but AMD’s stronger CPU footprint improves its access to enterprise accounts and can lower the commercial friction of GPU share gains over 6-18 months.

The key debate is whether agentic workloads genuinely require a sustained 1:1 CPU/GPU server mix or whether this reflects an early deployment phase that will be optimized away through inference accelerators, DPUs and custom silicon from AVGO-linked hyperscaler programs. A high earnings multiple leaves little tolerance for a deceleration in data-center growth, gross-margin dilution from CPU mix, or another large AI-GPU customer concentration disclosure. Near term, the stock needs independently verifiable evidence in backlog, cloud-instance launches and EPYC unit growth—not merely TAM expansion—to sustain relative performance.

Consensus is likely underestimating the margin distinction between revenue growth and value creation: server CPUs are strategically valuable for account control but generally do not command accelerator-like economics. The upside case is therefore AMD proving cross-sell economics and improving accelerator mix; the downside is that CPU strength becomes a lower-margin offset to slower GPU adoption. Watch the next two earnings cycles for data-center gross margin, disclosed MI-series revenue trajectory, and management’s 2027 CPU growth guide; any guide reset below roughly 50% would challenge the premium valuation quickly.

Key Decisions for Investors

• Maintain a tactical long AMD only on post-earnings confirmation of data-center growth and stable-to-up segment gross margin; use a 3-6 month horizon and size modestly given valuation sensitivity. Exit on a data-center guide reduction or evidence that CPU growth is being purchased through material pricing concessions.
• Preferred relative-value expression: long AMD / short INTC over 6-12 months, sized beta-neutral. The trade captures ongoing server-share migration while reducing broad AI-capex risk; reassess if Intel demonstrates sustained foundry/process execution plus two consecutive quarters of server-share stabilization.
• Do not initiate a directional short NVDA from this CPU theme alone. Instead, monitor whether NVDA’s networking and rack-scale attach rate offsets any CPU-content increase; a deterioration in those attachment metrics would be the prerequisite for a more credible AMD-versus-NVDA pair trade.
• Add TSM to the AI infrastructure watchlist rather than chase AMD at elevated multiples: confirmation that AMD’s CPU and accelerator volumes are translating into leading-edge wafer demand provides a lower single-product-risk way to express the same 12-18 month buildout thesis.