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

The article argues that AI infrastructure spending is shifting from training to inference and agentic AI, creating new winners beyond Nvidia. It highlights Broadcom’s potential in custom AI ASICs, including a clear line of sight to more than $100 billion in AI ASIC revenue by fiscal 2027, and AMD’s opportunities in inference and AI-agent workloads, supported by major OpenAI and Meta deals. The piece is largely an investor commentary rather than new company-specific news, so the likely market impact is modest.

Analysis

The market is quietly moving from a scarcity-of-compute regime to a monetization-of-compute regime. That transition matters because the winners are no longer just the chip vendors with the fastest silicon; the highest incremental returns should accrue to the firms that sit closest to workload specification, power efficiency, and networking bottlenecks. In that frame, custom ASICs and inference-optimized architectures can compress GPU pricing power over the next 12-24 months, especially if hyperscalers standardize around a smaller set of bespoke designs.

Broadcom looks best positioned because it monetizes the unglamorous layer where adoption becomes sticky: design wins, interconnect, and the networking fabric that scales chip clusters. The second-order effect is that every new custom accelerator deployment increases switching costs for the cloud customer while also expanding the need for high-margin networking silicon. That creates a more durable earnings stream than pure cycle exposure, and it also implies the current AI capex boom is broadening upstream into plumbing suppliers rather than just compute vendors.

AMD’s setup is more asymmetric but also more execution-sensitive. The key underappreciated point is that inference workloads are memory- and orchestration-heavy, which makes product mix and software maturity more important than raw FLOPS; that favors a company with improving software and memory-dense designs. The CPU angle is potentially larger than the GPU angle if agentic AI really increases server CPU intensity, but that thesis likely plays out over multiple budget cycles, not immediately. Near term, the market may be underestimating how much of the “AI second phase” is still a procurement story, where customer qualification timelines and supply allocation determine share gains more than technical superiority.

The contrarian risk is that consensus may be extrapolating demand curves too linearly while ignoring customer concentration. If the largest hyperscalers slow custom spend or demand better economics after initial pilots, revenue visibility can look strong while actual booked-to-bill weakens. That would hit AMD first because it is more reliant on successful platform adoption, while AVGO should hold up better due to its diversified exposure across ASICs and networking. NVDA remains the quality anchor, but the upside from multiple expansion is likely more limited than from the emerging second-derivative beneficiaries.

Key Decisions for Investors

• Overweight AVGO vs NVDA over the next 6-12 months: long AVGO / short equal-dollar NVDA as a relative-value expression of custom-chip and networking monetization. Risk/reward favors AVGO if hyperscaler ASIC adoption continues; stop if NVDA re-accelerates inference share or AVGO guidance loses visibility.
• Initiate a staged long in AMD on pullbacks over 1-3 months, with a 12-18 month horizon. Use call spreads if available to cap execution risk; thesis depends on ROCm adoption and follow-on design wins rather than near-term sentiment.
• Pair trade long AVGO / short a basket of hyperscaler capex proxies if valuation disconnect widens. The idea is to isolate supplier monetization versus customer capex intensity; the trade works if AI spend stays high but becomes more efficiency-driven.
• Buy NVDA only on weakness, not strength, as a core hold rather than a trade. Use it as the quality hedge inside the AI complex while the market tests whether inference economics cannibalize training-led multiples.
• Watch for 1-2 quarter confirmation on custom ASIC bookings and networking backlog; if either inflects down, cut exposure quickly because this is a delayed-cycle theme with sentiment already optimistic.