I'm Not Buying Nvidia Right Now. These 2 Growth Stocks Are the Smarter AI Supercycle Play.

The article argues that AMD and Broadcom are better AI infrastructure plays than Nvidia, citing AMD’s growing role in inference and AI-agent-related CPU demand, and Broadcom’s leadership in custom ASIC/TPU chips. AMD’s recent partnerships with OpenAI and Meta and Broadcom’s expanded TPU business with Alphabet and Anthropic are highlighted as key growth drivers. The piece is opinion-driven rather than event-driven, so the likely market impact is limited despite a constructive view on both stocks.

Analysis

The market is beginning to bifurcate AI capex into two pools: programmable training silicon and lower-cost inference silicon. That matters because inference is where utilization can scale fastest, but it is also where hyperscalers will be most aggressive about margin compression, pushing workload-specific hardware into the stack and away from general-purpose GPUs. In that regime, AMD and AVGO benefit not just from share gains, but from customers’ desire to diversify away from a single-vendor bottleneck and negotiate down total cost per token.

The second-order effect is that the real bottleneck is shifting from accelerator supply to system orchestration: CPUs, networking, interconnect, and power efficiency. AMD’s CPU upside is underappreciated because agentic workloads create more control-plane chatter, more session state, and more non-parallel logic than simple inference benchmarks imply. If AI agents become persistent rather than episodic, server CPU demand can compound faster than accelerator unit growth, which is why the incremental upside to AMD may come from the platform mix, not just GPU share.

Broadcom’s edge is that custom silicon becomes more attractive precisely when inference moves from experimentation to production. Once a workload is stable enough to hardwire, customers are trading flexibility for better watts-per-token economics, which tends to extend design win lifetimes and improve visibility. The risk, however, is concentration: a handful of hyperscalers can reprice or slow-roll deployments if ROI slips, and AVGO’s implied growth path requires a very high conversion rate from design wins into shipped volume over the next 12-24 months.

Consensus may be underestimating how much of the AI spend mix migrates away from Nvidia rather than how much total spend grows. That is constructive for AMD/AVGO relative to NVDA, but the trade is not risk-free: if model efficiency improves faster than token demand, capex could pause before these second-order beneficiaries fully monetize. The best setup is a medium-duration rotation trade, not a blind beta chase.

Key Decisions for Investors

• Add AMD on pullbacks over the next 2-6 weeks; favor a staged entry because the stock is likely to remain headline-sensitive until inference and CPU share gains show up in guidance. Risk/reward is attractive if CPU scarcity persists, but near-term execution risk remains high.
• Initiate a long AVGO / short NVDA pair trade for a 3-9 month horizon. Thesis: custom silicon and networking capture a larger share of post-training AI capex while NVDA remains exposed to any deceleration in generalized accelerator demand.
• Buy AVGO calls or call spreads out 6-12 months to express upside from design-win conversion without paying full multiple expansion. Use a spread to limit premium decay if hyperscaler monetization slips.
• Use AMD versus INTC as a structural relative-value long. AMD has clearer leverage to both AI server CPUs and GPU inference, while INTC’s benefit is more indirect and likely slower to monetize.
• If AVGO outperforms sharply on TPU/custom-chip headlines, trim into strength rather than chasing; this is a multi-quarter story, but crowded positioning can create 5-10% air pockets on any capex pause.