2 AI Stocks That Survived the March Sell-Off -- and Look Stronger Because of It

Arm projects a $100 billion data-center CPU market by 2031 and aims to capture $15 billion of that, targeting $25 billion in total revenue for the period. AMD generated $16.6 billion in data-center revenue last year (CPUs + GPUs), is launching its Venice chiplet-based CPUs to increase core counts for agentic AI, and has two GPU partnerships each valued at over $100 billion. The article argues agentic AI will materially increase CPU demand (sequential decision-making, tool-calling), creating a sizable growth runway for both Arm and AMD and implying potential upside for their stocks.

Analysis

The market is repricing a structural shift: CPUs are moving from an incidental utility to an active orchestration layer for agentic workloads, which creates asymmetric value for vendors that control architecture, packaging and software stacks simultaneously. That favors players who can combine high core counts, aggressive power-efficiency and fast time-to-market via foundry/pakaging partnerships; it also puts pressure on TTL suppliers (board vendors, power delivery, cooling) and memory/interconnect vendors to redesign for far higher ratios of CPU-to-GPU coordination traffic.

Second-order winners are the ecosystem facilitators — OS and hypervisor vendors, compiler toolchains, and server integrators — because early benchmarks and TCO studies will drive procurement cycles at hyperscalers; conversely, incumbent x86 specialists that cannot match silicon-software co-design risk margin compression and share loss. Key bottlenecks to watch are foundry capacity and advanced packaging throughput: a missed wafer or substrate cycle can delay multi-quarter ramps and transfer demand to competitors with spare node capacity.

Tail risks are binary and timing-sensitive: a delayed or underwhelming first silicon disclosure, poor power/perf at scale, or hyperscaler vertical integration would reverse upside quickly; alternatively, clear benchmark leadership and a couple of hyperscaler design wins would accelerate adoption within 12–24 months. Practically, expect volatility around tapeout windows, customer announcements, and quarterly guidance — these are the natural catalytic moments that will reprice winners/losers materially over 3–18 months.

Key Decisions for Investors

• Initiate a 12–18 month pair: Long AMD (equal-$ basis) / Short INTC (equal-$ basis). Rationale: capture continued share shift to AMD if agentic-CPU demand materializes while hedging market/systematic risk. Target relative outperformance 30–60%; set stop if pair underperforms by 15% vs S&P over any 90-day window.
• Buy a limited-risk AMD call spread (9–12 month expiry) sized for 2–4% portfolio exposure. Entry on a post-earnings pullback or on a tapeout/benchmark beat. Reward scenario: 40–100% upside if AMD converts design wins; max loss = premium paid.
• Build a small, event-driven long in ARM (3–5% portfolio): scale into first public server-CPU benchmark and a confirmed hyperscaler design win within 6–12 months. Use a 20% hard stop or hedge with out-of-the-money puts; upside scenario is 2x+ on a clean technical lead, downside is a >30% re-rating on execution/royalty concerns.
• Tactical hedge/express short on INTC: buy 12-month puts (or a short stock position) to express risk from pricing pressure and execution gaps. Position size should offset gross long CPU exposure; reward = levered downside if Intel fails to arrest share loss, cost = put premium or borrow/financing on short.