2 Growth Stocks Are the Smarter Plays in the AI Supercycle Than Nvidia Right Now

The article argues that the AI build-out is entering a second phase centered on inference and agentic AI, which could broaden leadership beyond Nvidia. Broadcom is highlighted as a major beneficiary of custom AI ASIC demand, with management indicating visibility to over $100 billion in AI ASIC revenue by fiscal 2027. AMD is presented as another winner via inference-optimized GPUs and a growing CPU role in AI servers, with two $100 billion GPU deals and a potential Anthropic contract cited.

Analysis

The market is likely underpricing how much the AI supply chain shifts from scarce FLOPS to scarce power-efficiency and deployment tooling. That is structurally better for AVGO and AMD than for pure training winners, because custom ASICs and higher-memory inference parts monetize the phase change without requiring the same open-ended capex intensity as frontier training. The second-order effect is margin pressure on general-purpose GPU ecosystems: as hyperscalers internalize more inference workload, pricing power migrates away from merchant accelerators and toward design wins, networking, and software-defined orchestration.

AVGO is the cleaner multiyear expression because its AI exposure is becoming less dependent on a single customer or single chip cycle and more tied to platformization at hyperscalers. The risk is not demand—it is timing and concentration: a few large programs can create lumpy revenue recognition, and any delay in custom silicon tape-outs can make estimates too aggressive over the next 2-3 quarters. But if power constraints tighten into 2026, the economic case for ASICs strengthens nonlinearly, which should extend the runway beyond the current consensus window.

AMD is more of a “catch-up plus optionality” story than a clean leader. The hidden upside is that if inference clusters move toward more CPU-heavy agentic workflows, AMD gets a second bite at the apple through server CPUs, which can raise its attach rate inside the same data-center budget. The key risk is execution: software adoption and developer friction can still slow share gains, so the next 6-12 months should be judged on gross margin stability and real deployment cadence, not headline partnership value.

The contrarian read is that NVDA is not losing the AI cycle, but its beta to incremental upside is lower from here because training remains the most defended part of the stack. That creates a relative-value setup: the crowd will continue paying for the leader, while the better risk/reward may sit in the beneficiaries of inference economics and custom silicon adoption. GOOGL and META matter as demand validators, but the more important tell is whether their capex mix shifts toward proprietary silicon and networking, which would confirm that the AI spend pool is broadening rather than concentrating.

Key Decisions for Investors

• Long AVGO vs short NVDA on a 3-6 month horizon: express the view that custom silicon and networking capture more incremental AI economics than merchant GPU training; target ~15-20% relative outperformance if hyperscaler ASIC adoption stays on schedule.
• Buy AMD on pullbacks with a 6-12 month view: use it as a higher-beta inference and agentic-AI catch-up trade, but size smaller than AVGO because execution risk is materially higher; favorable if the stock re-rates on evidence of sustained software traction.
• Pair long AVGO / long AMD against short INTC: the thesis is that AI server spend is migrating toward performance-per-watt and memory density, leaving legacy CPU share at risk unless the AI agent CPU transition materializes faster than expected.
• Sell out-of-the-money NVDA calls 3-6 months out against a long core position or use call spreads: monetize the idea that the market is paying up for dominance that is already well-owned, while capping upside if inference competition compresses multiple expansion.
• Watch GOOGL and META capex disclosures over the next 1-2 quarters; if custom chip/networking spend rises as a share of total AI capex, add to AVGO and AMD because that is the best confirmation that AI is moving from model training to industrial-scale deployment.