Should You Forget Nvidia and Buy These 2 Artificial Intelligence (AI) Stocks Instead?

Nvidia's revenue has increased eightfold over three years and grew 73% last quarter, while AI data-center spending is expected to exceed $700 billion this year, underpinning continued demand for GPUs and networking. Broadcom's data-center networking revenue grew 60% in Q1 2026 and it projects AI ASIC revenue to exceed $100 billion in fiscal 2027 (more than 1.5x fiscal 2025 revenue), highlighting its position in custom chips and switches. AMD, the distant No.2 GPU vendor, has secured large GPU commitments from OpenAI and Meta and leads in data-center CPUs, positioning it to benefit from growing inference and emerging agentic AI workloads.

Analysis

The shift from one-size-fits-all GPUs to task-specific silicon and tighter integration between switching and compute is a value reallocation event: more margin and strategic leverage will accrue to IP, packaging, and networking owners than to pure-play GPU die suppliers. That reallocation magnifies the importance of bottlenecks that are normally ignored in headline GPU debates — co-packaged optics, HBM supply, and advanced substrate capacity — each creates a 6–18 month gate for customers moving designs from tapeout to at-scale deployment, and therefore confers pricing power to firms controlling those inputs.

A second-order winner dynamic is customer concentration and stickiness: hyperscalers that co-design ASICs will lock in IP and ecosystem dependencies (switch topologies, interposer designs, software stacks) that raise switching costs to rivals and create lumpy revenue episodes. Conversely, this concentration raises volatility — a single hyperscaler changing strategy can remove a multi-quarter revenue stream. Expect material stock moves around a handful of hyperscaler contract announcements and foundry/OSAT capacity updates over the next 3–12 months.

The margin and TAM implications are asymmetric. Energy- and latency-sensitive inference workloads create durable demand for hardwired solutions that lower ongoing operating expense, but R&D and upfront NRE shift costs to suppliers and early adopters; those who can monetize NRE (licensing, IP blocks, switch-chip tie-ins) will capture a higher fraction of lifetime spend. The clearest near-term catalysts to watch are (1) large-scale ASIC production ramps (6–12 months), (2) co-packaged optics volume milestones (next 12 months), and (3) public hyperscaler design wins or disclosures that reprice customer concentration risk.

Key Decisions for Investors

• Long AVGO equity or Jan-2027 LEAP calls (12–18 month horizon). Thesis: asymmetric upside from networking + IP monetization as packaging/optics bottlenecks give pricing power. Position sizing: 3–5% of risk budget. Risk/reward: target 25–40% upside if revenue cadence accelerates; max loss = premium/stock cost if ramps disappoint.
• Pairs trade — Long AVGO / Short NVDA (equal notional, 6–12 month horizon) to express reallocation from GPU-dominated valuation to ASIC + networking. Use protective options on the short leg (buy 3–6 month calls) to cap tail risk. Risk/reward: skewed to upside if ASIC deals and CPO milestones print; downside if incumbent extends software lock-in unexpectedly.
• Call spread on AMD (buy 9–12 month call, sell higher strike) to play CPU + inference coordination upside while limiting premium risk. Timeframe: 6–12 months around fiscal quarters where customer disclosures are likely. Risk/reward: limited downside (net premium) vs 2–3x upside if agentic/coordination workloads drive incremental CPU attach.
• Event hedges: buy short-dated puts on AVGO (2–3 months) around key earnings or hyperscaler trade-show dates to protect against deal cancellations; conversely, buy OTM calls on leading foundries/OSAT names around capacity announcements to play the supply-side squeeze. Keep hedge notional <20% of directional exposure.