“2025 was another strong year for Broadcom. And we see the spending momentum by our customers for AI continuing to accelerate in 2026.”（CEO）
“This $73 billion in AI backlog to be delivered over the next eighteen months.”（CEO）
“Free cash flow grew 39% year on year to $26.9 billion.”（CFO）
“Q4 consolidated adjusted EBITDA was a record $12.12 billion, up 34% year on year.”（CEO）
“We are announcing an increase in our quarterly common stock cash dividend in Q1 fiscal 2026 to 65¢ per share, an increase of 10%.”（CFO）
Prepared Metrics
Metric
Value
Speaker/Context
2025财年总收入
639亿美元
首席执行官
2025财年AI收入
200亿美元
首席执行官
Q4收入
180亿美元
首席执行官
Q4调整后EBITDA
121.2亿美元
首席执行官
Q4半导体收入
111亿美元
首席执行官
Q4 AI半导体收入
65亿美元
首席执行官
Q4基础设施软件收入
69亿美元
首席执行官
Q4自由现金流
75亿美元
首席财务官
2025财年自由现金流
269亿美元
首席财务官
2025财年调整后EBITDA利润率
67%
首席财务官
Q4非GAAP毛利率
77.9%
首席财务官
Q1 2026 AI半导体收入指引
82亿美元
首席财务官
Q&A Batch (1-5 of 8)
Q1 — Hock Tan (analyst implied)
Topic: AI backlog clarification and XPU content evolution at largest customer
Key points:
Hock confirmed $73 billion AI backlog over eighteen months; analyst estimated ~$50 billion+ for fiscal 2026 alone (Hock did not directly confirm or deny that math).
Analyst asked about customer-owned tooling and XPU share evolution at largest customer over 1–2 years; Hock pivoted to backlog dynamics.
Mgmt stance: Neutral – Hock did not address the XPU content question directly; focused on accelerating backlog growth.
Q2 — Blayne Curtis
Topic: System vs. chip delivery model for customer #4 (follow-on order)
Key points:
Hock described delivery as a "system sale" (full rack) with Broadcom components, not just XPUs; responsibility for entire system.
For customer #4, the sale includes "key components" and is certified to run as part of the wholesale process.
No specifics on networking (e.g., Google’s own vs. Broadcom’s); Hock avoided naming customer or repeating prior model.
Mgmt stance: Bullish – Emphasized system sales as a scalable, responsible model for hyperscalers.
Q3 — Stacy Rasgon
Topic: Gross margin and operating margin impact from AI/system sales ramp
Key points:
Hock confirmed AI revenue has lower gross margin than rest of business (including software); system sales add pass-through costs (Kirsten: similar to XPU memory pass-through).
Gross margin dollars will rise, but margin percentage will fall; operating margin percentage will "come down a bit" but dollars grow due to operating leverage.
Kirsten said guidance for end of fiscal year will be provided; no explicit number (e.g., low 70s or high 60s) given.
Mgmt stance: Neutral-cautious – Acknowledged margin compression but highlighted dollar growth and leverage.
Q4 — Jim Schneider
Topic: AI revenue growth rate in fiscal 2026 and fifth customer identity
Key points:
Hock reiterated Q1 fiscal 2026 AI revenue doubled YoY; full-year growth originally (six months ago) guided ~60–70%, but backlog keeps rising.
$73 billion backlog over eighteen months is a "moving target" expected to grow; Hock declined to give precise fiscal 2026 guidance beyond Q1.
Did not confirm fifth customer as OpenAI; said "he's a real customer" on a multi-year journey to their own XPUs.
Topic: OpenAI contract timing and fifth customer order
Key points:
Hock clarified OpenAI agreement: ~10 gigawatts between 2027–2029, not starting in fiscal 2026 (Q1 start ruled out); press release referenced 2026–2029 but Hock said "more like 27, 28, 29."
Fifth customer order is separate; Hock did not confirm it is OpenAI, stated they will grow on their own multi-year XPU journey.
Mgmt stance: Neutral – Provided timeline clarification but avoided specific fifth-customer identity confirmation.
Q6 — CJ Muse
Topic: Custom silicon content growth generational and multiple XPU offerings per customer
Key points:
Hock said each of five customers can create "their version of chip" for training and inference, often on two parallel tracks per customer.
Differences include hardware variations (e.g., sparse call data routers vs. dense matrix multipliers in same chip; varying memory capacity/bandwidth for inference workloads like decoding vs. prefilling).
Hock rejected need to create more versions; existing pipeline has "plenty of different content."
Mgmt stance: Bullish – Emphasized breadth and differentiation of custom accelerators across customers.
Q&A Batch (6-8 of 8)
Q6 — Harsh Kumar
Topic: AI revenue growth drivers and optical component demand
Key points:
AI sequential growth guide is ~$1.7 billion, driven by a "mixed bag" of existing customers and existing XPUs.
Demand for 1.6 terabit per second DSPs (enabling optical interconnects) is "very, very strong"; demand for optical components (lasers, pin diodes) is "going nuts."
Of the $73 billion AI revenue backlog over the next 18 months, ~$20 billion is non-XPU (networking/optical), the rest is XPUs.
Mgmt stance: Bullish — growth feels "never ends," non-XPU backlog is still large ($20 billion) and valued.
Q7 — Harsh Kumar
Topic: (No content — analyst only said "Thank you"; operator interjection)
Key points: None.
Mgmt stance: Neutral — no management response provided.
Q8 — Carl Ackerman
Topic: Supply chain resiliency for AI components (co-ops, new custom compute processors)
Key points:
Hock Tan states AVGO has a "good handle" on bottlenecks through 2026, as it produces multiple leading-edge components: 1.6 Tbps DSPs, lasers (EMLs, VCSELs, CW lasers).
AVGO expands capacity early during design to match demand; it is often "part of the bottlenecks" and works to resolve them.
For AI system racks (excluding power, transformers, gas turbines), AVGO believes it has better visibility than other suppliers.
Mgmt stance: Bullish — AVGO proactively manages bottlenecks by co-developing capacity, feels well-positioned through 2026.