Broadcom: Six customers, $100bn AI silicon revenue, and a stock still pricing merchant-GPU risk

Stevie AI on Broadcom Inc. (AVGO-USA | broadcominca)

9/1/2026

Summary

Broadcom is a two-segment franchise—custom AI accelerators (XPUs) plus high-performance networking silicon, layered on a stable, high-margin infrastructure software base (VMware Cloud Foundation, Spring Security)—that has become the primary alternative to NVIDIA's merchant GPU stack for hyperscalers building frontier AI clusters. The structural insight is that Broadcom does not compete with NVIDIA head-on; it co-designs bespoke silicon (TPUs, Trainium-class chips, and custom networking fabric) with a small number of the world's largest AI spenders, giving it revenue visibility (>$30B quarterly bookings, 10 GW of shipment commitments into FY2027) that is unusual for a semiconductor company at this scale. The financial trajectory validates the thesis: revenue nearly doubled from $51.6B in FY2024 to $63.9B in FY2025, with net income almost quadrupling from $5.9B to $23.1B and EPS rising from $1.23 to $4.77 as AI semiconductor revenue scaled and non-GAAP tax anomalies normalized. Management's guidance embeds a further step-change—FY2026 revenue of $97.5B (AI semis alone ~$56B, +180% YoY) and FY2027 revenue of $124.8B with AI semiconductor revenue guided to exceed $100B. Free cash flow is expected to compound from $39.3B in FY2026 to $66.9B by FY2029, flipping the balance sheet from $40.2B net debt to $36.9B net cash over that window. Applying a 26x multiple to FY2027 EPS of $9.44 yields a price target of $245—below the current $370.34 price on a one-year view, but the multiple compresses on a forward basis as EPS scales: by FY2028 ($11.27 EPS) and FY2029 ($12.60 EPS), the same growth-adjusted multiple framework supports $293 and $328 respectively, with a 30x multiple justified as visibility into the FY2028–2029 AI silicon ramp firms up (yielding $338 and $378). We use a 30x forward multiple on FY2027 EPS as the primary anchor, reflecting Broadcom's position as the structural, monopoly-adjacent supplier of custom AI silicon to hyperscalers with 40%+ EPS CAGR and net-cash conversion—above the semiconductor sector average multiple (~20-24x) but below hypergrowth AI pure-plays, given Broadcom's diversified software base and de-risking customer commitments.

Thesis

1. **Uncontested Position in Custom AI Silicon**: Broadcom is not a merchant GPU vendor; it is the design partner of record for hyperscalers building proprietary AI accelerators (Google TPU, and reportedly Meta, potentially others) plus the networking fabric (switches, SerDes, CPO optical) that ties these clusters together. This dual position—compute plus networking—is structurally different from NVIDIA's model and from Marvell's more narrowly scoped custom silicon business. Quarterly AI semiconductor revenue of $10.8B, up 143% YoY, with $30B+ in quarterly bookings, indicates this is not a speculative narrative but a backlog-driven business. 2. **Revenue Inflection Is Front-Loaded and Multi-Year**: Management's guidance of ~$56B AI semiconductor revenue in FY2026 stepping to >$100B in FY2027 represents one of the largest disclosed revenue inflections in semiconductor history. The 10 GW shipment commitment signals multi-year hardware deployment cycles already contracted, not merely modeled. Hock Tan's commentary on FY2028

Risks

1. **Extreme Customer Concentration**: Six customers account for substantially all AI semiconductor revenue and backlog, with Google's TPU/networking relationship being the single largest identified contributor. Loss of, or reduced capex from, even one of these customers would materially impair the growth trajectory; the dollar magnitude of Google's committed spend is undisclosed, creating a visibility gap for investors. 2. **In-House Silicon Substitution Risk**: Meta's MTIA and other hyperscaler in-house silicon efforts represent a structural threat to Broadcom's design-partner model. If large customers move design work in-house or diversify to Marvell/AMD for cost or supply-chain reasons, Broadcom's share of the custom silicon TAM could shrink even as the overall AI capex pool grows. 3. **Margin Compression from AI Mix**: Gross margin is guided down to ~74% in Q3 FY2026 from AI/custom silicon mix dilution, and management flags continued gross margin pressure as XPU revenue scales faster than higher-margin networking and software. If opex leverage fails to offset this dilution, operating margin (currently held flat at ~67%) could erode faster than modeled. 4. **Hyperscaler Capex Cyclicality**: The entire AI semiconductor growth thesis is a derivative of hyperscaler capital expenditure decisions, which are macro- and ROI-sensitive. Any pause or moderation in AI infrastructure spending—whether from AI monetization disappointment, macro tightening, or overcapacity concerns—would directly compress Broadcom's $100B+ FY2027 AI revenue guidance. 5. **Valuation and Multiple Compression Risk**: At $370.34, the stock trades at a premium multiple to forward EPS even before this framework's price targets are reached; a de-rating from AI-related sentiment shifts (as seen in the stock trading >23% off peak despite record results) shows the market is already discounting execution risk. Any guidance miss versus the Q3 FY2026 $29.4B revenue / $16B AI semis target could trigger further multiple compression independent of fundamentals. 6. **Tax Rate Normalization Drag on EPS**: The shift from a negative 1.8% non-GAAP tax rate in FY2025 to ~16% in FY2026 due to global minimum tax rules is a headwind to EPS growth optics; investors extrapolating FY2025's EPS growth rate without adjusting for this tax normalization may be modeling unrealistic near-term EPS acceleration.

📈 Price Targets