Own Broadcom by Sept. 21 to Qualify for Its Sept. 30 Dividend. Here's How Many Shares You'd Need for $5,000 in Yearly Dividends.

Broadcom guided for AI revenue of $115 billion in fiscal 2027 and $230 billion in fiscal 2028, up from its $100 billion fiscal-2027 outlook six months earlier, as custom XPU and networking orders accelerate. Q3 fiscal 2026 free cash flow reached a record $13.7 billion, while first-nine-month FCF of $31.94 billion exceeded $9.28 billion in dividends, supporting continued dividend growth, buybacks and debt repayment. The dividend has risen for 15 straight years and compounded at 29% annually over the past decade, but the 0.8% current yield requires roughly 1,923 shares, valued at about $652,416, to generate $5,000 annually. The key risk is rising concentration in OpenAI and Anthropic, whose projected deployments could make Broadcom increasingly exposed to any downturn in AI capital expenditure.

Analysis

The investable change is not the dividend; it is AVGO's increasingly concentrated exposure to two private AI labs whose funding, power procurement, and cloud-access commitments are less transparent than public hyperscaler capex. A custom-accelerator design win can create unusually high revenue visibility after tape-out, but customer-specific silicon also carries a sharper downside: a delayed model roadmap, financing round, or data-center power interconnect can defer an entire deployment cycle rather than merely trim quarterly demand. This raises the appropriate discount rate for AVGO's AI backlog even as it supports near-term earnings revisions.

The second-order winner is TSMC (TSM) and, to a lesser extent, networking/component suppliers tied to scale-out clusters; AVGO's ASIC mix implies leading-edge wafer, advanced packaging, optical interconnect, and switch silicon demand. The relative loser is NVDA only at the accelerator layer, where ASIC adoption caps its share of inference and steady-state workloads; it remains advantaged where software portability and rapid architecture changes matter. Long term, the more ASICs replace general-purpose GPUs, the more AI economics shift from chip gross margin toward cloud/lab utilization, which could eventually improve GOOG and META returns on invested capital rather than simply increase their capex burden.

Consensus is likely treating announced capacity intentions as equivalent to funded revenue. The key 1-3 month catalyst is independent corroboration through customer capex guidance, foundry packaging commentary, and AVGO's disclosed backlog conversion; the 6-18 month risk is that AI labs become the marginal buyer without durable monetization, forcing compute leases or orders to be renegotiated. Thesis falsification would be a material reduction in AI revenue guidance, evidence of deferred deployments, or AVGO's non-AI businesses failing to offset a lower utilization profile.

Key Decisions for Investors

• Maintain a core long AVGO only on pullbacks around earnings rather than chase guidance-driven strength; frame as a 6-18 month position. Upside requires backlog conversion and sustained gross-margin resilience, while exit/reduce on a material AI guidance cut or any indication that a major private-lab order is contingent rather than binding.
• Express the ASIC-share thesis as a 6-12 month pair: long AVGO / short a modest notional of NVDA. This isolates custom-silicon substitution, but size conservatively because NVDA can outperform if frontier-model training remains architecture-sensitive or if software lock-in outweighs unit-cost savings.
• Add TSM to the watchlist as a lower-idiosyncratic beneficiary of custom accelerator volume; initiate only after management commentary confirms leading-edge wafer and advanced-packaging demand is not being displaced by inventory digestion. The risk is that packaging capacity constraints delay revenue recognition across AVGO's supply chain.
• Avoid using the capital-return narrative as the primary AVGO catalyst: yield support is weak relative to valuation sensitivity. Monitor the next earnings call for AI customer concentration, cancellation protections, and working-capital changes; absent that disclosure, treat projected lab demand as an alert rather than fully underwritten revenue.