AMD vs. Broadcom: The AI Chip Race Is Big Enough for Both. Here's the Better Buy.

Broadcom is presented as the preferred AI semiconductor investment, with management projecting AI revenue to double to $115 billion in fiscal 2027 and again to $230 billion in fiscal 2028, driven by TPUs and custom-chip programs for Alphabet, Anthropic, Meta, and OpenAI. AMD is also positioned for substantial inference and agentic-AI growth, with partnerships cited as potentially generating roughly $100 billion each from OpenAI and Meta over several years. The valuation case favors Broadcom at about 18x fiscal 2027 forward earnings versus roughly 35x for AMD, while analysts forecast 2027 revenue growth of 64% for Broadcom and 73% for AMD.

Analysis

The central differentiator is not headline AI demand but monetization quality. AVGO captures higher-value custom-silicon design/IP content and has embedded customer switching costs once an ASIC enters production, but its revenue is inherently concentrated, milestone-driven, and exposed to a small number of hyperscaler deployment decisions. AMD’s inference opportunity is more scalable across customers, yet software compatibility, systems integration, and realized cluster utilization—not memory bandwidth alone—will determine whether it earns durable pricing or merely fills capacity at lower gross margin.

Custom accelerators are a second-order negative for NVDA’s share of hyperscaler internal workloads, but not necessarily for total AI capex: lower cost per token can expand inference usage and preserve aggregate demand for networking, HBM, advanced packaging, and general-purpose GPUs. The nearer competitive casualty is likely MRVL, whose custom-silicon narrative faces greater customer concentration and a less differentiated software/IP ecosystem than AVGO. Watch TSMC advanced-packaging allocation and HBM contract pricing; both can constrain unit delivery while temporarily protecting supplier margins.

The cited forward valuation comparison should be treated cautiously because AVGO’s earnings mix includes software and its AI programs carry customer-specific execution risk, while AMD’s multiple embeds substantial inference share gains before they are independently visible in segment gross margin. Over the next 1-3 months, earnings commentary on booked versus deployed capacity, customer prepayments, and 2027 supply commitments matters more than broad AI demand rhetoric. The 6-18 month falsifier for AVGO is any material delay in a lead customer’s tape-out or production ramp; for AMD, it is failure to show sequential data-center GPU margin expansion and repeatable multi-customer inference wins.

Key Decisions for Investors

• Initiate a 3-6 month long AVGO / short AMD pair at equal dollar exposure following the next earnings-related volatility window. The thesis is that AVGO’s contracted design-win economics and customer lock-in deserve a relative premium; exit if AVGO reports a major custom-chip ramp delay or AMD demonstrates two consecutive quarters of data-center GPU margin expansion plus broad inference customer adoption.
• Maintain an underweight in MRVL versus AVGO for 6-12 months. Escalate the relative short only if AVGO discloses incremental custom-silicon program ramps while MRVL does not identify comparable production customers; cover on evidence of material non-concentrated custom AI revenue or a meaningful AVGO customer defection.
• Do not add directional AMD exposure solely on claimed customer partnerships. Set an alert for disclosed binding capacity commitments, GPU revenue conversion, and gross-margin contribution at the next two reports; absent those data, AMD remains a watch item rather than a fresh long despite favorable inference positioning.
• For NVDA exposure, avoid interpreting ASIC adoption as an automatic short catalyst. Hedge only if hyperscaler capex guidance shifts from incremental AI infrastructure spending to explicit GPU substitution; otherwise, custom silicon may expand token demand and sustain the broader AI supply chain.