Forget AMD. Here's Who Nvidia Really Needs to Be Worried About.

Broadcom reported fiscal Q3 AI semiconductor revenue of $16.7B, up 221% year over year, and projects this business could double to $115B in 2027 and again to $230B in 2028. The article argues Broadcom's custom-chip partnerships with Alphabet, Meta, OpenAI and Anthropic make it a more significant competitive risk to Nvidia than AMD, whose $6.7B data-center revenue grew 107% versus Nvidia's 117% growth to $89B. Broadcom's expanding custom AI-chip opportunity could shift AI-compute market share away from GPU-only deployments, although Nvidia remains substantially larger.

Analysis

The investable distinction is not GPU-versus-ASIC unit share; it is where incremental hyperscaler capex earns the highest return. Custom silicon is most disruptive to NVDA in mature, predictable inference workloads, where utilization and power efficiency matter more than programmability. Training, frontier-model experimentation, and rapidly changing architectures should remain GPU-heavy, so ASIC adoption is more likely to slow NVDA’s mix/price realization before it materially impairs absolute revenue.

AVGO’s upside depends on converting design wins into recurring production ramps without absorbing margin pressure from a concentrated customer base. The market should demand evidence in customer-funded commitments, advanced-packaging capacity, and AI semiconductor guidance rather than extrapolate aspirational multi-year targets; a shortfall would be punished disproportionately given the stock’s AI premium. Second-order beneficiaries of a broad ASIC ramp include TSMC (TSM), advanced-packaging equipment and high-speed networking suppliers, while HBM availability may become a binding constraint for both custom accelerators and GPUs.

Consensus is too quick to frame this as a zero-sum AVGO/NVDA contest. Hyperscalers can deploy ASICs for steady-state inference while retaining NVDA systems for training and workload flexibility, expanding total accelerator spend rather than reallocating a fixed budget. Over the next 1-3 months, the key catalyst is whether GOOG and META capex commentary identifies incremental internal-silicon deployments; over 6-18 months, the decisive metric is ASIC cost-per-token improvement after networking, memory, software, and utilization are included. The thesis fails if customer ramps slip, packaging constraints persist, or NVDA’s platform economics preserve a sufficiently large total-cost-of-ownership advantage.

Key Decisions for Investors

• Initiate a 3-6 month long AVGO / short AMD pair at equal dollar exposure. AVGO has direct custom-silicon optionality, whereas AMD remains exposed to a more difficult GPU share-gain and software-adoption path; reassess if AMD data-center guidance materially outgrows the market or AVGO reduces AI-related outlook.
• Maintain core NVDA exposure but hedge the inference-share risk with a smaller AVGO overweight rather than rotating entirely out of NVDA. A 60/40 NVDA/AVGO basket captures aggregate accelerator-capex growth while reducing exposure to a single architecture; trim the AVGO overweight if disclosed AI backlog or customer production schedules fail to support the next fiscal-year ramp.
• Use GOOG and META earnings as a catalyst watch rather than a directional trade: add to AVGO only if capex growth is accompanied by explicit evidence of higher internal-accelerator deployment. Capex increases without ASIC deployment confirmation remain more supportive of NVDA’s broader platform demand.
• Monitor TSM advanced-packaging utilization and HBM supply commentary as a bottleneck signal. Tightness supports near-term revenue visibility for accelerator suppliers but raises delivery-risk for AVGO; evidence of packaging-related shipment deferrals would be a reason to reduce the AVGO/NVDA semiconductor basket.