Broadcom Could Be Above $1,000 In 2 Years

Broadcom management is guiding AI revenue to $230 billion by 2028, underpinned by multi-year custom XPU orders from Anthropic, OpenAI, and Meta. Q3 revenue rose 85% year over year while net margin expanded to 44%, signaling accelerating AI-driven growth and strong operating leverage. The results reinforce Broadcom's positioning as a major beneficiary of hyperscaler AI infrastructure spending.

Analysis

The investable implication is not simply higher AVGO revenue, but a potential re-rating of custom silicon from a lower-multiple semiconductor category into the scarce capacity layer of hyperscaler AI capex. If customer-funded ASIC programs move from design wins to recurring volume ramps, AVGO gains unusually strong visibility and operating leverage; the offset is extreme customer concentration, where a delayed platform launch or a single customer’s shift toward internally designed silicon can materially alter utilization. TSM is the clearest second-order beneficiary through advanced-node and CoWoS packaging demand, while ANET benefits if scale-out networking keeps pace with accelerator deployments.

Custom XPUs are strategically disruptive to NVDA at the largest customers because they lower inference cost and reduce vendor dependence, but they are not direct substitutes for general-purpose training systems. The near-term read-through is therefore more negative for merchant ASIC competitors such as MRVL than for NVDA: AVGO’s scale, IP portfolio and packaging access raise the hurdle for competing custom-chip programs. META is a beneficiary only if ASIC deployment converts into lower cost-per-inference; otherwise, the capex burden arrives before the efficiency gains and can pressure its free-cash-flow narrative.

The key risk is that the market capitalizes management’s long-duration AI target before independently verifiable evidence appears in backlog conversion, customer prepayments, and segment-level gross-margin durability. Over the next 1-3 months, scrutinize whether incremental AI revenue comes with stable or improving gross margin and whether non-AI semiconductor weakness is being masked by mix; over 6-18 months, supply constraints in advanced packaging and hyperscaler capex normalization are the principal thesis breakers. A material reduction in AI-related purchase commitments, slower CoWoS capacity additions, or a gross-margin decline despite AI mix gains would falsify the premium-multiple case.

Consensus may underappreciate that successful custom silicon can cannibalize some high-margin networking and merchant accelerator content per AI cluster, making aggregate revenue growth an incomplete measure of value creation. The bull case requires AVGO to capture enough proprietary interconnect, packaging and software content to offset that substitution; without this, revenue can grow while incremental returns compress.

Key Decisions for Investors

• Maintain or initiate AVGO overweight versus SMH on post-earnings confirmation of AI backlog conversion and stable segment gross margin; use a 6-12 month horizon. The favorable setup is continued custom-silicon volume realization, while a sequential gross-margin deterioration or reduced customer commitment is a clear exit signal.
• Pair long AVGO / short MRVL over 3-6 months: AVGO’s scale and customer access create a higher probability of winning incremental hyperscaler ASIC programs, whereas MRVL needs new design-win evidence to defend expectations. Cover the short if MRVL discloses a material incremental AI ASIC customer or guides accelerating data-center margin expansion.
• Add TSM exposure selectively on evidence that advanced packaging lead times remain tight through the next two quarterly capacity updates; this captures the manufacturing bottleneck regardless of which chip architect wins. Do not add solely on semiconductor demand headlines—falsification is easing CoWoS lead times or a meaningful cut to capex from major AI customers.
• Treat META as an operational-efficiency watch rather than a direct AVGO sympathy long: add only if management quantifies lower inference unit costs or moderating capex intensity. Absent that disclosure, rising infrastructure spend can outweigh the strategic benefit of greater silicon control.