Nvidia Started the AI Boom. These 2 Stocks Could Fuel Its Next Chapter.

Marvell reported record fiscal Q1 revenue of $2.42 billion, up 28% year over year, with data center revenue reaching $1.83 billion and management guiding fiscal Q2 revenue to about $2.7 billion, or roughly 35% growth at the midpoint. Broadcom's most recent quarter showed $10.8 billion in AI semiconductor revenue, up 143%, underscoring strong demand for custom AI chips and networking hardware. The article argues that the AI infrastructure shift toward ASICs could benefit Marvell and Broadcom even as Nvidia remains dominant in GPUs.

Analysis

The market implication is not “GPU demand is weakening,” but that AI capex is fragmenting into two profit pools: training compute and cluster plumbing. That matters because custom silicon typically carries lower unit volatility and more durable design-in economics than accelerator ASPs, so the better long-duration trade may be the picks-and-shovels layer that benefits from every incremental rack deployed, regardless of who wins the compute socket. In that setup, Broadcom has the cleaner second-order exposure because networking content per dollar of AI capex rises as cluster size scales.

Marvell looks like a higher-beta version of the same thesis, but its key sensitivity is customer concentration and program timing rather than end-demand. The stock can re-rate hard on design-win momentum, yet it is also more vulnerable to any single hyperscaler pausing a custom program or elongating qualification cycles. That makes MRVL more of a quarterly execution trade, while AVGO is closer to a multi-year compounder with multiple shots on goal across silicon, interconnect, and software-adjacent pricing power.

For Nvidia, the strategic read-through is mixed but not bearish enough for outright structural shorting. A larger ASIC share reduces the rate of GPU penetration in certain inference workloads, but it also expands the total addressable infrastructure market and preserves NVIDIA’s dominance in the frontier training layer. The real risk is margin mix: if hyperscalers succeed in shifting enough inference to custom chips, GPU utilization could normalize faster than consensus expects, compressing forward growth even if unit demand remains strong.

Contrarianly, the market may be underestimating how much of the value shifts from chips to connectivity and system integration once AI clusters get large enough. If power and rack efficiency become the binding constraints, networking vendors and chip designers with deep platform relationships can capture economic rent without needing to “beat” Nvidia head-on. That makes the winner set broader than the headline suggests, and it argues for owning the enablers of AI scale rather than betting solely on the compute leader.

Key Decisions for Investors

• Long AVGO over a 6-12 month horizon; best risk/reward in the group because AI networking content should compound even if custom accelerator growth slows. Use pullbacks to add, with downside protected by broad diversification across end-markets.
• Tactically long MRVL for a 1-3 month catalyst trade into the next print or guidance update; upside is higher on incremental design-win commentary, but trim aggressively if management language implies any customer timing slippage.
• Pair trade: long AVGO / short NVDA for 3-6 months as a relative-value hedge against AI capex rotation from general-purpose accelerators to custom silicon and cluster interconnect. The pair benefits if inference spending outpaces frontier-training spending.
• Avoid shorting NVDA outright unless evidence emerges that utilization or order backlog is rolling over; the better expression is a covered-call or collar structure to monetize implied volatility while retaining upside.
• Watch for hyperscaler capex commentary over the next 1-2 quarters; if cloud budgets stay firm but GPU mix softens, rotate incrementally from NVDA into AVGO and MRVL rather than reducing AI exposure wholesale.