Better AI Chip Stock: Broadcom vs. Nvidia

Broadcom's AI semiconductor revenue surged 221% year over year to $16.7 billion in fiscal Q3, representing 56% of revenue, and management projects AI chip sales of $115 billion in fiscal 2027 and $230 billion in fiscal 2028. Custom accelerators, including OpenAI's Broadcom-co-developed Jalapeño chip, could reduce Nvidia dependence for inference workloads and expand Broadcom's addressable market. Nvidia remains highly resilient, with total revenue up 106% to $96.2 billion and Data Center revenue up 117% to $89 billion, while non-hyperscaler AI cloud, industrial, and enterprise revenue rose 138%; it also trades at a lower forward P/E of 14.4x versus Broadcom's 18.8x.

Analysis

The relevant debate is not AI demand but value capture per compute dollar. Custom ASIC adoption shifts economics from Nvidia's high-margin, full-stack platform toward a lower-switching-cost silicon model; AVGO can monetize design/IP and networking attach, but its revenue is likely more concentrated, milestone-driven, and exposed to a handful of customers' deployment schedules. The market should discount management's long-dated AI targets until tape-outs, wafer commitments, and customer capex disclosures corroborate them.

NVDA's offset is that custom silicon is most viable for stable, high-volume inference workloads, not the rapidly changing training and frontier-model environment where software portability, networking, and time-to-deployment matter more than chip-level cost. The second-order winner from greater ASIC penetration is TSMC, which captures leading-edge wafer demand regardless of architecture, while AVGO's Ethernet/fronthaul attach can benefit even where its accelerator does not win. Conversely, merchant accelerator suppliers without CUDA-scale software or AVGO's custom-design capability face the clearest strategic squeeze.

Near-term, this is more likely a valuation-relative catalyst than a material share shift: investors may reward AVGO contract visibility and penalize NVDA on inference-share headlines. Over 1-3 months, monitor hyperscaler capex, disclosed ASIC volume ramps, and NVDA Data Center gross margin; a custom-chip narrative without capex growth is merely mix substitution, not incremental TAM. Over 6-18 months, the key falsifier for NVDA is sustained deceleration in hyperscaler revenue alongside material gross-margin pressure, not isolated custom-chip announcements.

Contrarian view: the market may overstate the zero-sum framing. Hyperscalers can deploy ASICs to lower inference unit costs and reinvest savings into more tokens, models, and training clusters, expanding total AI infrastructure spend. AVGO deserves a premium only if custom programs convert into recurring multi-generation platforms rather than one-off design wins with lower incremental margins.

Key Decisions for Investors

• Initiate a 3-6 month relative-value position: long NVDA / short AVGO in equal dollar amounts if AVGO outperforms on custom-ASIC headlines without corroborating customer capex or volume disclosures. Thesis is that AVGO's long-duration expectations are more execution-sensitive; exit if AVGO reports backlog/wafer commitments supporting its AI outlook or NVDA guides material Data Center gross-margin compression.
• Add TSM as a 6-18 month architecture-agnostic AI exposure on pullbacks; custom accelerators diversify, rather than reduce, leading-edge foundry demand. Size below single-name AI leaders because Taiwan/geopolitical and customer-concentration risks can dominate the semiconductor-cycle thesis.
• Use NVDA quarterly Data Center gross margin and hyperscaler growth as the decision trigger rather than headline inference-share estimates: reduce the relative long if margin declines materially for two consecutive quarters while hyperscaler growth lags enterprise/cloud growth.
• Do not underwrite AVGO's stated multi-year AI trajectory as base-case earnings until customer concentration, program timing, and incremental gross-margin data are disclosed. Treat a confirmed delay in a major custom-chip deployment or a lower AI revenue guide as a short-side catalyst over the following 1-3 months.