I'm Not Buying Nvidia Right Now. These 2 Growth Stocks Are the Smarter AI Supercycle Play.

The article argues AMD and Broadcom are better positioned than Nvidia for the next phase of AI, citing inference, agentic AI, and custom-chip demand as key growth drivers. It highlights AMD’s new partnerships with OpenAI and Meta, plus rising CPU demand, while Broadcom’s ASIC/TPU business could see custom AI chip revenue reach $100 billion in fiscal 2027. This is opinionated analyst commentary rather than new company-specific financial results, so the likely market impact is limited.

Analysis

The market is still pricing AI as a single-bottleneck story, but the edge is shifting from raw training FLOPS to system architecture and orchestration. That matters because inference and agentic workloads are far more sensitive to total cost of ownership, latency, and power efficiency than model-training benchmarks, which structurally expands the addressable market for both custom silicon and high-end CPUs. The second-order effect is that the value chain should broaden: hyperscalers, networking, and memory vendors can all gain even if GPU share becomes less concentrated.

AMD’s setup is less about taking share from Nvidia outright and more about becoming a credible “good-enough” platform for workloads where capex discipline matters. The bigger upside catalyst is not the headline AI GPU win rate, but the CPU scarcity created by agentic workloads running more parallel, stateful inference chains; that is a slower-burn, multi-quarter revenue tailwind if supply remains tight. The main risk is that the market is extrapolating partnership announcements into sustainable design wins before software ecosystems and deployment scale are proven.

Broadcom has the cleaner second-order trade because custom AI silicon monetizes not just chip content but the whole design-and-networking stack. If custom accelerators keep growing, the pull-through into high-speed interconnects and merchant networking should compound, while generic GPU demand becomes relatively more elastic on pricing. The key contrarian point: Broadcom may be less cyclical than the market assumes because inference workloads create recurring replacement and expansion demand, but the stock can still be vulnerable if a handful of hyperscalers delay capex by even one budget cycle.

The consensus is likely underestimating how quickly power and rack-level constraints force buyers away from “best chip” toward “best system.” That favors Broadcom most, AMD second, and Nvidia still as the default benchmark, but with diminishing exclusivity. Over the next 3-6 months, watch for any signs of CPU supply tightness or custom-silicon design wins converting into booked revenue; if those slip, the trade should de-rate faster than the narrative has risen.

Key Decisions for Investors

• Long AVGO / short NVDA as a 6-12 month relative-value pair: express the view that inference economics and custom silicon will take share from generic GPUs; target outperformance if hyperscaler capex broadens beyond training-only spend.
• Accumulate AMD on pullbacks for a 6-9 month horizon: upside is driven by CPU scarcity and inference adoption, but size modestly because execution depends on software adoption and supply continuity.
• Use call spreads in AVGO rather than outright equity for the next 2 quarters: better risk/reward if the market re-rates custom-chip recurring revenue, with defined downside if customers defer deployments.
• Hedge AI beta with a small short in high-multiple semiconductor suppliers most levered to training-only demand; the risk is a slower transition to inference, while the reward is protection if capex rotates toward efficiency-driven architectures.
• Set a tactical trigger to add on any evidence of constrained CPU availability or incremental custom-chip bookings; those are the earliest signs the market is still underappreciating the duration of the demand cycle.