Forget AMD: This AI Hardware Stock Is the Smarter Bet Right Now

Broadcom's AI semiconductor revenue reached $16.7 billion in fiscal Q3 2027, up 221% year over year, versus AMD's $6.7 billion data-center revenue, up 107%. Broadcom raised its projected AI semiconductor revenue for fiscal 2028 to $115 billion from $100 billion and expects $230 billion in 2029, underpinning the article's bullish case that Broadcom is undervalued relative to AMD. The analysis argues that hyperscaler demand for lower-cost custom AI chips could allow Broadcom to outperform AMD over the next several years.

Analysis

The relevant debate is not AVGO versus AMD headline AI growth, but whether hyperscaler ASIC programs convert from design wins into durable, high-volume deployments without eroding Broadcom’s economics. AVGO captures high-value silicon, networking, and software-adjacent spend around custom accelerators, while AMD remains more exposed to the still-unproven ability of a second merchant GPU vendor to win sustained share against NVDA. If ASIC adoption broadens, the more consequential loser is likely merchant GPU pricing power: NVDA and AMD may need to accept lower accelerator ASPs or richer customer financing to defend workloads that can be standardized.

The reported long-range AI revenue outlook should be treated as a customer-capex scenario rather than a backlog equivalent. Over the next 1-3 months, AVGO’s multiple can expand if management discloses additional funded programs, rising custom-chip content per customer, or networking attach rates; AMD needs MI-series unit/ASP evidence and gross-margin expansion, not merely data-center revenue growth, to close the relative valuation gap. A 6-18 month risk to AVGO is customer concentration: a delay, architecture change, or in-house silicon success at one hyperscaler could create an abrupt revenue air pocket because ASIC ramps are lumpy and have long qualification cycles.

Consensus may be too linear on ASIC substitution. Custom chips are superior for stable, mature inference workloads, but frontier model training and rapidly changing architectures preserve GPU flexibility; this limits near-term displacement and can leave AVGO vulnerable if investors capitalize aspirational 2028 revenue too early. The cleaner thesis is relative execution: AVGO can outperform AMD even if overall AI spending moderates, but only if its incremental AI revenue converts at margins consistent with current semiconductor profitability rather than being bought through customer-specific pricing concessions.

Key Decisions for Investors

• Initiate a 6-12 month beta-neutral long AVGO / short AMD pair, sized 1:1 on semiconductor-factor beta rather than dollars. Target 15-25% relative return if AVGO sustains AI margin and AMD fails to demonstrate material GPU-share gains; exit if AVGO reports AI semiconductor growth below 50% year-over-year for two quarters or guides segment gross margin materially lower.
• Use NVDA as a read-through hedge rather than a directional short: retain NVDA exposure against the AVGO long if hyperscaler capex remains strong, since custom silicon adoption can coexist with expanding total compute demand. Reassess if NVDA data-center gross margin compresses by more than 300 bps or management identifies ASICs as a material demand displacement factor.
• Before adding outright AVGO, wait for independently verifiable disclosure of customer commitments, shipment timing, and incremental gross-margin contribution. If those details are absent at the next earnings release, treat the long-range revenue target as promotional guidance and keep exposure limited to the relative-value pair.
• Monitor hyperscaler capex guidance and networking demand over the next two earnings cycles. Broad-based capex reductions, not AMD-specific execution, are the principal near-term falsifier of the AVGO thesis and would favor reducing both AI hardware legs rather than rotating further into AVGO.