1 Agentic AI Chip Stock to Buy and 1 to Sell

The article favors AMD over Arm as agentic AI could drive a major expansion in server-CPU demand, with the GPU-to-CPU ratio projected to shift from 8:1 in LLM training to 1:1 in AI-agent deployments. AMD estimates the data-center CPU market could reach $220 billion and targets 50% share, while its inference-GPU opportunity includes reported $100 billion deals with OpenAI and Meta plus a sizable Anthropic agreement. Arm is viewed negatively despite its server opportunity because it trades near 140x forward earnings, faces declining smartphone volumes, and could see gross-margin pressure as it moves from IP licensing into chip production.

Analysis

The relevant read-through is not simply higher CPU unit demand: agentic workloads increase the value of memory bandwidth, networking, orchestration software and power efficiency per token. AMD is better positioned to monetize a broader system bill of materials through EPYC plus Instinct accelerators, but CPU attach-rate expansion will likely be gradual because hyperscalers optimize heterogeneous clusters rather than rebuild installed GPU-heavy fleets. The more immediate beneficiary of rising CPU intensity could be AMD's server mix and gross-margin resilience if EPYC share gains persist against Intel; it is not a near-term step-function in industry CPU TAM.

ARM's strategic risk is customer conflict, not merely handset cyclicality. Hyperscalers use Arm IP to retain control of custom silicon economics; a move into merchant server chips could encourage Amazon, Google and Nvidia to diversify architectures, negotiate royalties harder, or internalize more design. That creates a tension between higher revenue per server and a lower-multiple, inventory- and execution-sensitive hardware revenue mix. The article's large customer-contract assertions should be treated as unverified until reflected in AMD backlog, disclosed purchase commitments, or customer capex commentary.

Over 1-3 months, the key catalyst is hyperscaler capex guidance and evidence that inference spending broadens beyond NVDA-centric deployments. Over 6-18 months, AMD's upside requires both accelerator software adoption and sustained EPYC share gains; either alone is insufficient to justify a major earnings rerating. The contrarian case is that agentic inference becomes increasingly latency- and power-constrained, favoring custom ARM CPUs and ASICs at cloud providers rather than merchant x86 systems. Falsify the AMD relative thesis if EPYC revenue growth decelerates materially versus Intel's Data Center segment for two quarters, or if AMD's data-center gross margin contracts despite revenue growth.

Key Decisions for Investors

• Initiate a 3-6 month long AMD / short ARM dollar-neutral pair on strength, sized modestly: AMD has diversified CPU-plus-accelerator monetization while ARM faces customer-conflict and business-model transition risk. Target 15-20% relative return; stop if ARM demonstrates incremental disclosed hyperscaler chip design wins or AMD cuts data-center guidance.
• For cleaner CPU-share exposure, overweight AMD versus INTC over the next two earnings cycles. Monitor EPYC revenue, server OEM commentary and Intel Data Center operating losses; take profit if Intel shows two consecutive quarters of server-share stabilization and improved segment margin.
• Do not underwrite AMD on reported large AI customer commitments without verification. Add only after quarterly disclosures show data-center backlog conversion, higher accelerator revenue visibility, and no deterioration in inventory or receivables.
• Use ARM upside as a hedge rather than a standalone short: buy limited-delta ARM calls against the pair ahead of earnings if implied volatility is reasonable. A credible merchant-chip design win or materially higher royalty guidance could produce a sharp multiple-driven squeeze despite structural concerns.
• Watch AMZN, GOOG and META capex calls over the next 1-3 months. Increased custom-silicon disclosure is negative for merchant CPU assumptions but can be positive for ARM royalty economics; increased spend on general-purpose GPU/CPU clusters supports AMD's nearer-term revenue conversion.