Broadcom Inc. (AVGO) | The Buildout — AI Infrastructure
The Verdict
Broadcom designs and supplies custom AI accelerators, Ethernet switches, and networking silicon that power the hyperscale clusters of the world’s largest AI labs. Its chips are embedded in the training and inference infrastructure of Google, Meta, OpenAI, Anthropic, and other frontier developers, making it a non-optional supplier to the AI buildout.
| Market Cap | — |
| Revenue (TTM) | $75.5B |
| Revenue Growth | +32.3% |
| EBITDA Margin (TTM) | 55.3% |
| Net Debt | $45.3B |
| Earnings Beats | 7 of 7 |
| P/E (TTM) | — |
| EV/EBITDA (TTM) | — |
What We Like
- AI semiconductor revenue accelerated from +106% YoY in Q1 FY2026 to +143% in Q2, with Q3 guidance pointing to >+200%.
- Q2 AI bookings of >$30B against $10.8B shipped extended visibility to 2028.
- Management reiterated the FY2027 AI revenue target of in excess of $100B, backed by ~10 GW of contractual deployments across six customers.
- AI networking (switches, SerDes, optics) is share-agnostic, attaching to both Broadcom’s own XPUs and third-party GPUs.
- The AI XPU Platform with Apollo and Blackstone provides a $35B first tranche to fund customer compute capacity, derisking demand.
What We’re Watching
- Customer concentration: top-5 end customers reached 45% of revenue in H1 FY2026; the 10-Q warns of material adverse effect if orders are cut.
- Google diversification: CEO acknowledged Google will “have some diversity of sources,” which could reduce Broadcom’s share over time.
- Unexplained pause in share buybacks: no repurchases in Q2 after $7.8B in Q1, raising capital allocation questions.
- Gross margin compression: consolidated gross margin guided down to ~74% in Q3 from 77.1% in Q2, driven by mix shift to lower‑margin AI semis.
The AI-driven growth thesis has strengthened materially: bookings overwhelmed shipments, customer commitments expanded to binding multi-gigawatt contracts, and management provided a formal FY2026 AI revenue target of $56B. The open question is whether the concentration of revenue among six customers introduces a fragility that the current backlog masks.
Earnings Beat
Broadcom posted record revenue of $22.2 billion, up 48% year-on-year, with a gross margin of 67.2% and EBITDA of $13.0 billion. AI semiconductor revenue surged 143% to $10.8 billion, while quarterly AI bookings exceeded $30 billion — nearly three times shipments.
| Metric | Q2 FY2026 | Q1 FY2026 | Q2 FY2025 | YoY |
|---|---|---|---|---|
| Revenue | $22.2B | $19.3B | $15.0B | +47.9% |
| Gross margin | 67.2% | 65.6% | 68.0% | -80bps |
| EBITDA | $13.0B | $10.8B | $10.2B | +28.2% |
| EPS | $1.91 | $1.50 | $1.03 | +85.6% |
| AI semiconductor revenue | $10.8B | $8.4B | n/a | +143% YoY |
During the quarter, bookings for AI semiconductors were over $30 billion against the $10.8 billion we shipped.— Hock Tan, CEO, June 3, 2026
Management tone: Management shifted to a more confident and contractual tone, moving from aspirational language to specific gigawatt commitments and purchase-order references. The CEO called demand “simply insatiable,” forcefully clarified the chip-only business model, and candidly acknowledged Google’s expected diversification.
Management Guidance
For Q3 FY2026, management guided consolidated revenue of approximately $29.4 billion, AI semiconductor revenue of $16 billion, and non-AI semiconductor revenue of around $4.5 billion. Infrastructure software is expected to reach $8.9 billion. Consolidated gross margin is guided to roughly 74%, with operating margin holding at around 67%. The full-year FY2026 AI semiconductor revenue target is $56 billion.
Trajectory
Revenue accelerated from $19.3 billion in Q1 FY2026 to $22.2 billion in Q2, with EBITDA rising from $10.8 billion to $13.0 billion, driven by the AI semiconductor ramp. The non-AI semiconductor segment also showed signs of recovery, with bookings exceeding $6 billion, while infrastructure software saw ARR growth of 17% year-on-year.
The Model
The model projects FY+1 revenue of $106.1 billion with EBITDA of $62.5 billion (58.9% margin), and FY+2 revenue of $170.5 billion with EBITDA of $105.7 billion (62.0% margin). Near-term growth is anchored by the H2 FY2026 ramp and customer deployments; FY+2 captures the full-year effect of those ramps plus the AI XPU Platform’s capacity expansion.
| Metric | FY2025 | Next FY (E) | Following FY (E) |
|---|---|---|---|
| Revenue | $63.9B | $106.1B | $170.5B |
| YoY Growth | — | +66.1% | +60.7% |
| EBITDA | $36.4B | $62.5B | $105.7B |
| EBITDA Margin | 57.0% | 58.9% | 62.0% |
Projections are the median of 5 independent model runs. The model’s revenue sits 1.9% below analyst consensus.
For Q3 FY2026, management guided consolidated revenue of approximately $29.4 billion, AI semiconductor revenue of $16 billion, and non-AI semiconductor revenue of around $4.5 billion. Infrastructure software is expected to reach $8.9 billion. Consolidated gross margin is guided to roughly 74%, with operating margin holding at around 67%. The full-year FY2026 AI semiconductor revenue target is $56 billion.
What Could Go Right — and Wrong
- AI booking momentum sustains at 3:1 book-to-ship ratio, pulling FY2028 visibility forward.
- Content per gigawatt increases faster than expected, pushing FY+2 AI revenue beyond modeled levels.
- Non-AI recovery and software reacceleration add a second and third growth engine, boosting consolidated revenue and margins.
- AI XPU Platform attracts customers beyond the current six, expanding the addressable market.
- Gross margins within the AI semiconductor segment prove more durable than assumed, limiting blend compression.
- A major AI customer, particularly Google, significantly reduces its reliance on Broadcom silicon.
- Execution delays on one or more of the six custom XPU ramps cause revenue shortfalls.
- Supply-chain disruption at TSMC or among sole-source materials suppliers halts shipments.
- AI capex pullback by frontier labs, driven by model efficiency improvements or funding constraints.
- Gross margin dilution eventually pressures operating margin if R&D spending scales faster than revenue.
Looking Ahead
The next twelve months will test execution across multiple, simultaneous AI chip ramps. The Q3 FY2026 report is the immediate proof point, with AI semiconductor revenue guided to $16 billion. Beyond that, deployment of Anthropic’s 5 GW and Meta’s 1 GW in 2027, the first revenue from OpenAI’s Jalapeño processor late this year, and progress on the $35 billion AI XPU Platform will define whether management’s $100 billion-plus FY2027 target remains on track. Supply-chain preparedness and capital-allocation decisions will also be closely watched.
- Q2/Q3 FY2026Tomahawk 7 tape-out — 200 Tbps Ethernet switch silicon tape-out; critical for next-gen networking.
- Sep 2, 2026Q3 FY2026 earnings — Tests $16B AI semiconductor revenue and ~74% gross margin guidance.
- Late 2026FY2026 full-year results — Confirms $56B AI semiconductor revenue target and FY2027 setup.
- Late 2026OpenAI production — Jalapeño XPU volume production; validates 9-month design-to-production cycle.
- Beginning 2027Anthropic 5 GW TPU compute — Next-gen TPU deployment at scale; tests gigawatt ramp execution.
- 2H 2027Meta 1 GW delivery — Initial MTIA XPU and networking shipments; anchors 3 GW through 2028.
Financials
Annual Summary
| Metric | FY2025 | TTM |
|---|---|---|
| Revenue | $63.9B | $75.5B |
| Gross Margin | 67.8% | 67.0% |
| EBITDA | $36.4B | $73.9B |
| EBITDA Margin | 57.0% | 55.3% |
| Net Income | $23.1B | $29.3B |
| Free Cash Flow | $26.9B | $55.5B |
| Net Cash | — | — |
Key Ratios (Trailing)
- P/E TTM—
- EV/EBITDA TTM—
- EV/Revenue TTM—
- Price/FCF TTM—
- Gross Margin (TTM)67.0%
- EBITDA Margin (TTM)55.3%
- Net Margin (TTM)38.8%
- ROIC19.6%
- FCF Conversion78.6%
- SBC / Revenue11.6%
The Company
Broadcom designs and supplies a vast portfolio of semiconductor and infrastructure software products. Its semiconductor segment spans custom AI accelerators (XPUs/TPUs), Ethernet switches, SerDes, optical components, RF filters, and storage connectivity — chips that underpin hyperscale data centers, enterprise networking, and broadband.
The company outsources the majority of front-end wafer manufacturing to TSMC, with internal sole-source fabs in Colorado, Pennsylvania, and Singapore for niche processes. It sells through a concentrated distribution channel, with a single distributor accounting for 42% of revenue in H1 FY2026, and its top five end customers — all hyperscalers and AI labs — represented 45%.
Business Segments
Competitive Landscape
Broadcom competes in custom ASICs primarily against Marvell, and in networking against NVIDIA’s Spectrum switches and InfiniBand. In infrastructure software, VMware faces public-cloud alternatives. The company’s deep IP portfolio, sole-source products (Tomahawk 6, SerDes), and rapid custom-design execution create a wide competitive moat, though hyperscaler in-house teams and a potential Google diversification effort are notable threats.
- Marvell TechnologyDirect competitor in custom ASICs for AI, according to management and supply-chain intelligence.
- NVIDIACompetes in data-center networking with Spectrum switches and NVLink; Broadcom’s Ethernet is positioned as an alternative.
Supply Chain
Broadcom sits at the intersection of compute and networking in the AI infrastructure stack. It supplies custom silicon and switches to hyperscalers and AI labs, relying on TSMC for leading-edge wafers and a concentrated set of OSAT and memory suppliers.