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, supported by TPU and custom-chip programs for Alphabet, Anthropic, Meta, and OpenAI. AMD is also positioned for strong inference and agentic-AI growth, with partnerships cited as potentially generating about $100 billion each from OpenAI and Meta over several years. Analysts forecast AMD revenue growth of 73% in 2027 and 38% in 2028, versus Broadcom growth of 64% and 57%, but Broadcom's roughly 18x fiscal-2027 forward P/E versus AMD's 35x underpins the article's preference.

Analysis

The investable distinction is not simply GPU versus ASIC: AVGO monetizes hyperscaler capex even when workloads migrate away from merchant accelerators, while AMD must win workload share against NVDA and internally designed silicon. Custom-chip adoption is most damaging to NVDA’s unit economics at the mature, predictable inference layer; it is less disruptive in frontier training, where flexibility and software tooling retain value. AVGO therefore has a structurally broader share-of-wallet claim, but its AI revenue is also unusually concentrated in a small number of customers, making program delays or changes in customer make/buy decisions material to estimates.

AMD’s upside is highly convex to proof that its software stack can sustain production inference deployments, not merely secure design wins. The relevant KPI over the next 1-3 quarters is accelerator revenue growth and gross-margin progression after HBM and packaging costs, alongside evidence that large customers are expanding deployments rather than qualifying a second source. If AMD converts inference demand without meaningful pricing concessions, its earnings revision cycle could outperform; if it requires discounts to overcome ecosystem friction, revenue growth may not translate into multiple support.

Consensus may be underestimating that hyperscaler ASIC programs are a capex-efficiency trade, not incremental spending: stronger TPU/custom-silicon adoption can pressure the addressable market and pricing umbrella for both NVDA and AMD. Conversely, the market may be over-extrapolating AVGO’s long-dated management targets before verifying foundry capacity, customer purchase commitments, and the margin mix between high-value IP/networking content and lower-margin semiconductor pass-through. Near-term article impact is low; the next earnings prints and customer capex guidance are the catalysts.

Key Decisions for Investors

• Initiate a 3-6 month AVGO/AMD relative-value position, long AVGO and short AMD in beta-adjusted dollars, only if the forward valuation discount remains material after normalizing fiscal-year conventions. Target 15-20% relative return from AVGO estimate durability and AMD execution-risk repricing; exit if AVGO discloses a major custom-chip program delay or AMD reports accelerator revenue and gross margin materially above consensus.
• Maintain NVDA exposure but hedge 10-20% of AI-semiconductor beta through a modest long AVGO overlay rather than treating ASICs as a clean NVDA short. Custom silicon is most likely to cap incremental inference economics over 6-18 months, while NVDA remains better protected in training; a broad NVDA short lacks a clear near-term catalyst.
• Set an earnings alert on AMD for data-center segment growth, AI GPU revenue, and gross margin versus consensus. Do not add outright AMD until management demonstrates sequential deployment expansion at named cloud customers and stable/improving gross margin; these data are required to distinguish durable inference share from low-margin qualification volume.
• For AVGO, trim or avoid chasing a sharp pre-earnings rally unless customer concentration and backlog conversion are disclosed with sufficient detail. The key falsifier is reduced visibility on the largest ASIC programs or weaker networking attach rates, either of which would challenge the premium assigned to long-duration AI revenue.