Not AMD. Not Marvell. Here's Which Company Nvidia Needs to Be Worried About.

Broadcom's AI semiconductor revenue reached $16.7 billion in Q2, versus Marvell's AI data center revenue of $2.1 billion. The article estimates Broadcom's AI segment revenue at about $58 billion this year, rising to $115 billion in 2027 and $230 billion in 2028, and argues custom chips could help it gain share from Nvidia without replacing GPUs. It notes Broadcom trades at 43 times trailing earnings and 17 times next fiscal year's estimates, and presents the stock as a buy; these are the article's estimates and analysis.

Analysis

The investable distinction is workload mix, not a clean Nvidia displacement story. Custom accelerators can divert repetitive, predictable inference workloads while GPUs retain value where flexibility, software compatibility, and rapid model changes matter. If total AI compute demand keeps expanding, Broadcom can grow without a corresponding decline in Nvidia revenue; treating every ASIC dollar as a lost GPU dollar overstates the substitution risk.

Broadcom’s upside case also carries a less visible execution and concentration risk: custom-chip programs depend on a small number of large customers’ architectures, deployment schedules, and willingness to fund dedicated infrastructure. A delayed customer ramp or a shift back toward general-purpose GPUs could make forward earnings estimates—and the implied valuation argument—fragile. Verify the fiscal periods, customer-level program ramps, and whether quoted revenue expectations are company guidance or external estimates before underwriting them.

Over the next 1–3 months, watch Broadcom’s AI-related guidance and evidence that customer programs are converting into shipments, alongside Nvidia’s data-center growth and commentary on inference demand. Over 6–18 months, networking and accelerator integration may let Broadcom capture more of customers’ AI infrastructure spend, but Alphabet’s external TPU availability could also intensify competition among custom-chip suppliers rather than guarantee Broadcom’s economics. The article’s bullish valuation case is conditional on estimates and a higher earnings multiple; it does not establish either outcome.

Key Decisions for Investors

• Consider a measured AVGO-over-NVDA relative-value position rather than an outright NVDA short: the thesis is incremental ASIC share, not GPU obsolescence. Scale only after Broadcom confirms program ramps in reported results or guidance.
• Keep exposure sized for customer concentration and estimate risk. Reassess if Broadcom’s AI guidance or reported shipments fail to support the ramp, or if major customers indicate deployment delays.
• Monitor NVDA data-center growth and demand commentary as the key falsifier of the substitution thesis. Continued strong growth would suggest expanding AI demand is offsetting any workload migration.
• Treat MRVL as a watch item, not an automatic short: the article’s revenue comparisons do not establish comparable periods, product mix, or margins. Verify customer wins and profitability before taking a relative position.