A credible path to roughly doubling in five years if AI networking/custom silicon keeps scaling and infrastructure software renewals remain stable enough to fund reinvestment. The main debate is whether VMware churn and hyperscaler in-sourcing cap the multiple.

Broadcom compounds as a scarce “AI cluster bill-of-materials” choke point (Ethernet switching/optics + custom accelerators) while VMware-driven infrastructure software cash flows fund faster product cadence, supply commitments, and capital returns; upside convexity comes from packaging networking/infra into SLA + verification-priced offers that reduce price-only procurement without triggering VMware exits.

It sells the shovels for AI scale (networking + custom silicon) and also owns an embedded enterprise control surface (VMware). Main obsolescence vectors are hyperscaler in-sourcing and VMware migration pressure from pricing/packaging changes.

The non-linear opportunity is Broadcom moving up the stack from “parts supplier” to “assured AI infrastructure”: validated Ethernet-fabric designs with paid telemetry/SLA layers, plus VMware-based control/chargeback for private AI workloads. That shift matters because it can turn volatile hardware cycles into more durable, contract-like economics while keeping capex low (fabless + software). The base of the model remains a two-engine business: AI-driven semiconductor growth (custom silicon + switching/optics content per cluster) plus infrastructure software that remains sticky enough to finance aggressive reinvestment and shareholder returns. The key constraint is not demand creation but conversion: outsourced leading-edge wafers/advanced packaging availability and customer concentration can make growth lumpy, which is why the multiple assumption bakes in some de-rating versus today.

The dominant risks are (1) hyperscaler bargaining power and vertical integration compressing Broadcom’s content per AI cluster, (2) VMware renewals weakening over multiple cycles as customers execute phased exits, (3) outsourced leading-edge/advanced-packaging capacity gating near-term revenue conversion, and (4) premium valuation amplifying drawdowns if any of the above show up in guidance.

They control choke points that AI data centers must buy—high-speed networking silicon and custom compute-adjacent chips—and they also sit inside enterprises through VMware renewals. The risk is powerful buyers designing around them and customers steadily migrating off VMware if pricing/workflow friction stays high.

$421.34