Hock Tan Just Guided Broadcom's AI Revenue to Reach $230 Billion by 2028. Should You Believe Him?

Broadcom projects annual AI semiconductor revenue of $230 billion in fiscal 2028, more than tripling from its approximately $65 billion annualized Q3 AI revenue run rate. The company also raised its 2027 AI-revenue outlook to $115 billion from its prior $60 billion-$90 billion range and says it has secured the supply chain needed to meet demand. The article estimates that, if AI targets are met and non-AI revenue grows 10% annually, Broadcom could generate roughly $280 billion in 2028 revenue and $140 billion in profit at a 50% margin, implying a $4.2 trillion valuation at 30x earnings.

Analysis

The investable issue is not whether custom silicon grows, but whether AVGO can convert design wins into recognized revenue at the implied scale without diluting economics through customer concentration. Hyperscaler ASIC programs shift compute spend away from NVDA's merchant-GPU model, but they also create lumpy, multi-year ramps tied to a handful of customers' capex budgets and model-roadmap decisions. AVGO's differentiation is the co-design ecosystem and networking attach, which can make each accelerator win more valuable than chip-only revenue; conversely, a delayed tape-out, lower utilization, or one customer internalizing more design work can produce a sharp quarterly guide reset.

The supply-chain assertion should be treated as capacity reservation, not proof of deliverable revenue. The binding constraints through the next 12-24 months are likely advanced packaging, leading-edge wafer allocation, HBM availability, and substrate/component yields; AVGO can secure inputs while still facing customer qualification delays or lower-than-planned system builds. TSMC is a cleaner volume beneficiary, while AVGO's custom-chip growth is a competitive headwind at the margin for NVDA and potentially Marvell (MRVL), though NVDA's software lock-in and faster time-to-market preserve a substantial segment of workloads.

Consensus may be over-extrapolating a revenue target into earnings by assuming both a stable mix and expanding margins. Custom ASICs can carry attractive gross margins, but the economics depend on NRE recovery, customer-specific pricing, and the proportion of lower-margin silicon versus higher-margin networking/software; the cited terminal-margin framework is not an underwriting basis. Near term, bullish positioning can persist on further customer-program disclosures, but the 1-3 month risk/reward is asymmetric if quarterly AI guidance merely meets expectations; over 6-18 months, the key catalyst is evidence that multiple programs move from qualification to repeat production rather than one-off deployments.

Key Decisions for Investors

• Maintain a core AVGO long only on a staggered basis around earnings; add after evidence of sequential AI revenue acceleration and unchanged gross-margin guidance, rather than chasing a guidance-driven gap. Falsify on a material reduction in AI backlog/conversion commentary, a customer concentration disclosure worsening, or gross-margin compression attributable to custom silicon.
• Express the custom-silicon substitution theme as long AVGO / short MRVL in equal semiconductor-beta dollars over 6-12 months. AVGO has broader customer and networking monetization, while MRVL is more exposed to execution expectations around a narrower set of AI programs; cover the short if MRVL discloses a major production-scale ASIC win or AVGO signals program delays.
• For a purer supply-chain read-through, monitor TSM (and advanced-packaging utilization disclosures) rather than assuming AVGO's reserved capacity is immediately monetizable. A long TSM position becomes actionable if leading-edge packaging lead times remain extended while hyperscaler capex guidance is raised; avoid treating this article alone as confirmation.
• Do not initiate a directional NVDA short solely on ASIC enthusiasm. Use AVGO/NVDA relative performance as an alert: a sustained AVGO outperformance alongside rising disclosed ASIC volumes, without corresponding NVDA inference-growth acceleration, would support a 6-12 month relative-value short NVDA versus long AVGO; NVDA software attach and new-platform demand are the primary invalidation risks.