Citi Identifies Chip Winners as AI Capex Cycle Accelerates — NVDA, AVGO, MRVL Top Picks

The Quick Take

• Citigroup's semiconductor team published a bull case note naming NVIDIA, Broadcom, and Marvell as the primary winners of the accelerating AI capex cycle.
• The note raised price targets across the group, citing evidence that hyperscaler GPU orders are running ahead of prior-quarter guidance.
• Custom silicon momentum at Broadcom and Marvell was highlighted as a structural share-gain story independent of NVIDIA's dominance.

Citigroup's semiconductor analyst team upgraded their sector view and raised price targets for NVIDIA, Broadcom, and Marvell, arguing that hyperscaler AI capital expenditure commitments have materially exceeded analyst consensus. The catalyst: channel checks and earnings commentary from data centre equipment vendors indicating GPU order backlogs are growing, not shrinking, despite prior expectations of digestion following the 2024-2025 AI buildout surge.

The custom silicon thesis for Broadcom and Marvell is separate from — and potentially more durable than — the NVIDIA demand cycle. AVGO and MRVL are designing application-specific integrated circuits for Google (TPUs), Meta (MTIA), and Amazon (Trainium) that reduce dependence on NVIDIA for specific inference workloads. As these custom chips ramp to production volumes in 2026-2027, Broadcom and Marvell capture large-scale revenue that does not depend on NVIDIA's competitive position.

For semiconductor investors, Citi's note reinforces the thesis that the AI capex cycle has years of runway rather than being near a peak. The bearish alternative — that hyperscalers will cut capex once they determine they have over-built — is implicitly rejected by Citi's channel data showing continued order acceleration. The risk scenario is a demand surprise from a hyperscaler that pauses and triggers a cluster of analyst downgrades.

Synthesized from 1 source.