Elon Musk Is Going All In on Nvidia, but Are These Two Chip Stocks Better Buys?

AI inference spending is projected to grow at a 32% CAGR through 2032 to $1.3 trillion, roughly double the projected $658 billion AI-training market, creating a potentially larger opportunity for AMD and Broadcom than for Nvidia. AMD is positioned through lower-cost, memory-focused GPUs and reported multiyear $100 billion deals with OpenAI and Meta, while Broadcom expects AI revenue to double to $115 billion next fiscal year and to $230 billion in fiscal 2028. The article argues that Nvidia remains dominant in training, but AMD and Broadcom could outperform due to their inference exposure and growth from smaller revenue bases.

Analysis

The investable issue is not whether inference grows, but where economics accrue as workloads migrate from centralized model development to high-volume serving. That transition favors silicon with lower total cost of ownership—memory bandwidth, networking, power and utilization—not simply lower chip price. AVGO is positioned to monetize the design-to-volume bottleneck through custom ASIC programs and associated networking, while AMD's opportunity depends more directly on software maturity and availability of complete rack-scale systems; these are materially different risk profiles despite the same inference narrative.

Custom silicon is a second-order negative for NVDA's incremental inference mix, but not necessarily for its aggregate earnings near term: hyperscalers typically deploy ASICs for stable, high-volume internal workloads while retaining GPUs for rapidly changing models and third-party capacity. The more immediate spillover is toward AVGO's optical/interconnect franchise and potentially ANET, because distributed inference and custom accelerators increase network bandwidth requirements. Conversely, merchant GPU pricing pressure would hit AMD before NVDA if customers use custom silicon principally to replace lower-cost GPU deployments.

The article's revenue and customer assertions should be treated as promotional until reconciled with 10-Q disclosures, foundry capacity commitments, HBM supply, and customer capex guidance. Over the next 1-3 months, AI capex commentary from GOOG, META and MSFT is the primary sentiment catalyst; over 6-18 months, disclosed ASIC production ramps and inference gross-margin trends determine whether AVGO's premium multiple is earned. Falsification for the AVGO thesis is a material reduction in hyperscaler capex or evidence that TPU/custom-chip deployment remains captive and does not translate into external volume; for AMD, failure to convert design wins into reported data-center revenue and sustained ROCm adoption would invalidate the catch-up case.

Consensus may be underestimating the duration of NVDA's software and systems advantage, particularly as inference moves to agentic workloads with variable models and long-context memory demands. ASIC economics improve most when workload specifications are stable; rapid model iteration can preserve GPU flexibility longer than the headline inference TAM implies. Therefore, a broad "inference equals NVDA loss" trade is too simplistic; relative exposure should favor AVGO over AMD rather than reflexively shorting NVDA.

Key Decisions for Investors

• Initiate a 6-12 month long AVGO / short AMD dollar-neutral pair; target 15-20% relative upside with a 7-10% stop on relative underperformance. AVGO has the clearer monetization path if custom-chip programs reach volume, while AMD bears higher execution risk around software, supply and customer conversion.
• Maintain core NVDA exposure rather than shorting it against the inference thesis; hedge concentrated NVDA longs with 3-6 month downside puts only if hyperscaler capex revisions turn negative. A break in aggregate GOOG, META and MSFT AI capex guidance is the trigger for reducing the semiconductor complex, not inference TAM projections.
• Add ANET on pullbacks as a second-order beneficiary of accelerator cluster and inference-network scaling; validate with order growth and management commentary on AI back-end demand. Exit if cloud order lead times normalize without corresponding revenue acceleration.
• Set an earnings-season watch item for AVGO: require evidence of funded customer programs, advanced-packaging/foundry capacity, and AI revenue conversion rather than accepting long-range company targets. If these disclosures do not improve over the next two reporting cycles, avoid adding despite favorable narrative momentum.