Got $1,000? 2 Growth Stocks Building the Data Centers of the AI Infrastructure Supercycle.

The article recommends AMD and Broadcom as AI infrastructure beneficiaries, citing AMD's inference, agentic AI and physical-AI positioning alongside reported $100 billion inference deals with OpenAI and Meta. Broadcom forecasts custom-chip revenue will double to $115 billion next fiscal year and double again to $230 billion the following year, driven by hyperscaler ASIC demand and AI networking. The article also argues Broadcom is attractively valued at less than 12x estimated fiscal 2028 earnings, while portraying both stocks as long-term AI growth opportunities.

Analysis

The investable distinction is not simply AMD versus AVGO: AVGO monetizes hyperscaler AI capex twice—through custom silicon and the network fabric required to scale clusters—while AMD must prove that software maturity, memory availability, and customer deployment convert design wins into recurring accelerator revenue. If ASIC adoption broadens, NVDA faces a mix shift from premium general-purpose training GPUs, but the nearer second-order beneficiary is AVGO because each incremental custom-chip program also increases attach opportunities in switches, optics-adjacent DSPs, and SerDes. TSMC capacity and advanced packaging remain the common bottleneck; a tight supply environment could limit volume upside while supporting pricing and gross margins for the highest-priority designs.

The reported AMD customer commitments, acquisitions, and AVGO revenue trajectory require independent verification before underwriting estimates. For AMD, the key 1-3 month catalyst is evidence of production deployments and raised data-center revenue/gross-margin guidance, rather than additional partnership announcements; the 6-18 month question is whether inference workloads become sufficiently heterogeneous to reward its open architecture. For AVGO, custom-silicon revenues are likely lumpy and customer-concentrated, making any delay in a single hyperscaler tape-out or leading-edge node transition disproportionately important despite the longer-term secular setup.

Consensus likely understates Ethernet's role if scale-out inference and agent workloads drive east-west traffic faster than compute spend. But it may also overstate ASIC substitution: hyperscalers retain NVIDIA exposure where workloads change rapidly, so custom silicon is more likely to cap NVDA's incremental share than produce an abrupt GPU demand collapse. The highest-probability trade is therefore relative exposure to AI infrastructure bottlenecks, not a directional bet on every AI-chip narrative.

Key Decisions for Investors

• Initiate a 3-6 month long AVGO / short NVDA pair in equal dollar beta-adjusted size. Thesis: AVGO captures custom-silicon and networking mix gains while NVDA is more exposed to expectations compression if hyperscaler capex rotates toward ASICs; target 10-15% relative return, with a stop if NVDA's next data-center guide materially accelerates or AVGO discloses a major customer program delay.
• Maintain AMD as a watch-list long rather than add aggressively on partnership headlines. Upgrade only after independently verified production revenue and data-center gross-margin expansion at the next earnings print; use a 6-12 month horizon and invalidate if accelerator revenue conversion or MI-series supply guidance slips.
• For AI-networking exposure, prefer AVGO over a broad semiconductor ETF such as SOXX for the next two quarters, but size below core levels because the custom-chip business has concentrated-customer and tape-out timing risk. Reassess if hyperscaler capex guidance weakens or leading-edge foundry/packaging constraints extend beyond two quarters.
• Set an event alert around GOOG, META, and major AI-lab capex disclosures: confirmation that internally designed accelerators are moving from pilot to volume production is a catalyst for AVGO and a relative headwind for NVDA; evidence that flexible GPU fleets remain the preferred deployment model reverses the pair-trade thesis.