The Best Stocks to Buy Right Now: Nvidia vs. AMD vs. Broadcom

The article argues Nvidia remains the AI leader, but highlights emerging growth opportunities for AMD and Broadcom as AI shifts from training to inference and agentic AI. Nvidia is cited as targeting a $200 billion opportunity and $20 billion in CPU revenue this year, while Broadcom sees more than $100 billion in ASIC revenue by fiscal 2027. The author’s preferred stock is AMD, but the piece concludes all three are attractive AI buys.

Analysis

The market is likely underappreciating how quickly AI capex is fragmenting from a single-vendor GPU story into a multi-layer stack. That is bullish for the overall AI infrastructure complex, but it changes the winner set: NVDA still owns the software moat, yet the marginal dollar of incremental spend is increasingly likely to go to networking, custom silicon, and CPUs rather than pure accelerator share. The second-order effect is that hyperscaler bargaining power improves over the next 6-18 months, which should compress headline enthusiasm for NVDA while expanding TAMs for AVGO and, to a lesser extent, AMD.

AMD’s opportunity is the most convex because it is still valued like a challenger, not a core AI platform beneficiary. If inference and agentic workloads really do shift the bottleneck toward memory and cores, AMD gets leverage from both GPU and CPU exposure without needing to win the entire training stack. The key risk is execution: software adoption and design-win conversion lag hardware cycles, so the market may overpay for near-term “optional inventory” and then demand proof in 2-3 quarters; until then, the stock can rerate on narrative faster than fundamentals.

AVGO is the cleaner cash-flow compounder, but the market may be too comfortable with the idea that custom silicon is automatically accretive. Every hyperscaler ASIC win is also a signal that customers are trying to internalize margin that would otherwise sit with NVDA, and that can cap long-run accelerator pricing across the ecosystem. The hidden loser is not NVDA alone but also smaller semiconductor vendors and ODMs tied to off-the-shelf GPU buildouts; custom silicon shifts value toward IP, networking, and system integration, not unit volume.

Contrarianly, the consensus may be too linear on “AI spend = more semis.” If inferencing economics improve materially, the total workload volume could explode, but capex intensity per unit of revenue may fall, which matters for multiple compression across the group. That makes relative positioning more attractive than outright beta: the trade is not just long AI, it is long the companies closest to workload migration and systems integration, while fading names whose economics depend on sustained scarcity pricing.

Key Decisions for Investors

• Go long AMD vs NVDA over a 3-6 month horizon: pair trade captures the re-rating potential from inference/agentic optionality while limiting downside if AI capex cools; target 15-20% relative outperformance if AMD secures additional inference design wins.
• Add AVGO on pullbacks for a 6-12 month hold: use it as the highest-quality expression of AI capex diversification, but size smaller than NVDA because custom-chip wins can eventually pressure pricing; preferred risk/reward is 2:1 to the upside with lower fundamental volatility.
• Reduce or hedge NVDA into strength over the next 1-2 quarters via covered calls or a partial short against long basket exposure: the company remains best-in-class, but consensus may be overestimating how much of incremental spend remains in discrete GPUs.
• Initiate a long AMD / short NVDA basket only if AMD confirms a second major inference/customer adoption catalyst: the trade needs proof, not speculation, but can work well as a catalyst-driven relative-value position.
• If you want pure AI infrastructure beta, prefer AVGO + AMD over equal-weight NVDA: the mix is better aligned to the next phase of workload migration and offers more upside if the market rotates from training scarcity to inference efficiency.