Nvidia vs. Broadcom: Whose AI Revenue Actually Grows Faster From Here?

Broadcom forecasts AI semiconductor revenue of roughly $58B in fiscal 2026, $115B in fiscal 2027, and $230B in fiscal 2028—implying two consecutive years of approximately 100% growth. Nvidia generated $96.2B of quarterly revenue, including $89B in data-center sales, and guided next-quarter revenue to about $108B while projecting roughly 70% fiscal-2028 growth. Broadcom offers the faster AI-growth outlook but carries greater customer-concentration risk, while Nvidia's larger and more diversified customer base supports a potentially more durable expansion.

Analysis

AVGO’s forward AI narrative is less a broad semiconductor-cycle call than a concentrated commitment from a small set of hyperscalers to internal ASIC roadmaps. The critical distinction is that custom silicon displaces portions of accelerator spend only after customers have committed software, networking, packaging, and deployment capacity; a delayed tape-out or lower utilization can defer an entire revenue tranche rather than merely reduce quarterly orders. This makes AVGO’s earnings sensitivity materially more convex than NVDA’s, despite the apparent growth advantage.

The underappreciated beneficiary of both paths is TSMC (TSM): advanced-node wafer demand, CoWoS/advanced packaging, and leading-edge design starts rise whether compute is delivered through merchant GPUs or customer-specific ASICs. Memory suppliers, particularly MU, also benefit if accelerator deployments remain supply constrained, though ASIC mix can alter HBM content and therefore the magnitude of upside. Conversely, MRVL faces a more difficult competitive setup: it must win a narrower set of custom-program sockets while AVGO’s scale and incumbency raise execution and pricing pressure.

Consensus may overstate direct substitution between NVDA and AVGO. ASICs are most compelling for stable, high-volume inference workloads, while NVDA retains an advantage in frontier training, rapidly changing models, and customers unwilling to bear multi-year design risk. Over the next 1-3 months, the decisive read-through is hyperscaler capex commentary and disclosed custom-chip production ramps; over 6-18 months, the key question is whether ASIC deployment improves customer compute economics enough to redirect incremental budgets away from GPUs rather than simply expanding total AI spend.

A reversal in the AVGO thesis would be signaled by a major customer slowing capex, an advanced-packaging bottleneck, or AI semiconductor growth falling below management’s implied ramp for two consecutive quarters. For NVDA, the more relevant falsifier is not a single customer pause but a deterioration in non-hyperscaler demand or a material gross-margin concession needed to sustain volume.

Key Decisions for Investors

• Initiate a 3-6 month relative-value position: long NVDA / short AVGO, sized beta-neutral, only if AVGO’s valuation premium to NVDA expands further on the growth narrative. The trade expresses AVGO’s higher customer-concentration and execution risk while retaining AI infrastructure exposure; cover if AVGO validates its next two custom-silicon ramp milestones without margin dilution.
• Add TSM on weakness as the lower-idiosyncratic AI compute beneficiary over a 6-18 month horizon. Track CoWoS capacity commentary, leading-edge utilization, and customer prepayments; the thesis is impaired by a broad hyperscaler capex reset rather than a shift between NVDA and AVGO.
• Maintain an underweight/short bias in MRVL versus AVGO for the next two earnings cycles, but do not add aggressively before visibility into new design wins. The risk is that MRVL announces a large incremental ASIC program or materially improves its optical interconnect mix, which would invalidate the competitive-pressure thesis.
• Set an event alert around hyperscaler earnings: any coordinated reduction in AI capex or comments favoring internally designed inference silicon should favor the NVDA/AVGO relative short leg initially; evidence that ASICs are incremental rather than substitutive should reverse that positioning and support owning both suppliers.