Earnings Highlights: Broadcom (AVGO) | FY25 Q4
Summary
Driven by its twin engines of AI and VMware, Broadcom delivered a historic set of results. FY25 full-year revenue was $64B (+24% YoY), including AI semiconductors at $20B (+65% YoY);
Q4 revenue was $18B (+28% YoY), with AI semiconductors at $6.5B (+74% YoY) for the quarter. AI order visibility has stretched out sharply, with an AI backlog of $73B (over the next 18 months),
covering XPUs, Tomahawk 6 switches, DSPs, lasers and PCIe switches. Management made it clear: XPU + AI networking + system sales (full systems/racks)
are pushing Broadcom into a new role as an AI data center system-level supplier.
💰 Financials (non-GAAP)
Q4 FY25
Revenue: $18.0B (YoY +28%)
Gross margin: 77.9%
Operating expenses (Opex): $2.1B (R&D $1.5B)
Operating income: $11.9B (+35% YoY)
Operating margin: 66.2%
Adjusted EBITDA: $12.12B (68% margin)
Free cash flow: $7.5B (41% of revenue)
Cash: $16.2B
Inventory: $2.3B (58 days DOI, improved QoQ)
FY25 full year
Revenue: $63.9B (YoY +24%)
Semiconductor revenue: $36.9B (+22% YoY)
Infrastructure Software: $27.0B (+26% YoY)
Adjusted EBITDA: $43B (67% margin)
Free cash flow: $26.9B (+39% YoY)
Shareholder returns: $17.5B (dividends + buybacks)
Q1 FY26 guidance
Revenue: $19.1B (YoY +28%)
Semiconductor revenue: $12.3B (YoY +50%)
AI semiconductors: $8.2B (YoY +100%)
Infrastructure Software: $6.8B
Adjusted EBITDA margin: ~67%
Tax rate: ~16.5%
Dividend: quarterly $0.65/share (+10% QoQ)
⚙️ Product updates
1️⃣ AI Accelerators (XPU / Custom Silicon)
FY25 AI revenue $20B (+65% YoY)
Q4 AI semiconductors $6.5B (+74% YoY)
XPU business more than doubled YoY
The existing four major customers continue to ramp;
a fifth XPU customer has been added ($1B order, delivering in 2026)
XPUs are no longer used only for internal training/inference; they are also starting to be offered commercially to outside customers (e.g., TPU)
Management stressed:
Custom XPU is a multi-year journey
It can implement Sparse Core / inference / reasoning in hardware
The "customers will build their own chips" thesis is seen as overstated
2️⃣ AI Networking (Switch / DSP / Optical / PCIe)
AI networking demand is "even stronger than XPU"
AI switch backlog > $10B
Tomahawk 6 (102.4T)
Currently the only 102T switch in volume production
Broadcom's fastest-growing switch product ever
Strong shipments include:
DSP (1.6T)
Lasers / EML / VCSEL / CW
PCIe switches
Total AI backlog: $73B (next 18 months)
About $50B+ to land in FY26
Of which roughly $20B is networking/optics, the rest XPUs
3️⃣ System Sale (Rack-level / System-level Delivery)
Broadcom formally enters the system sale (full system/rack) model
Selling not just XPUs, but:
XPU
Networking
Optical
PCIe
Memory (HBM pass-through)
Management noted:
System sales will dilute gross margin
but revenue scale × operating leverage still lifts absolute profit
System sales started showing up in Q4, ramping in the second half of 2026
4️⃣ Infrastructure Software (VMware / VCF)
Q4 revenue $6.9B (+19% YoY)
FY25 revenue $27B (+26% YoY)
VMware Cloud Foundation (VCF) is the main growth driver
Q4 TCV bookings $10.4B
Software backlog $73B (+49% YoY)
FY26 expected low double-digit growth
🔭 Outlook
AI becomes Broadcom's core growth engine
AI revenue keeps accelerating (FY26 YoY growth > FY25)
Non-AI semiconductors:
Broadband recovering
Wireless flat
Enterprise still weak
Infrastructure Software: VMware steady at low double digits
Supply chain strategy
Continuing to expand TSMC 3nm / 2nm
Singapore advanced packaging facility
Focused on multi-chip / CoWoS-type packaging
Strengthening supply-chain security and delivery capability
💬 Q&A highlights
$73B AI backlog
A "snapshot in time" that will keep growing
Lead times of about 6–12 months
XPU vs GPU
Commercial TPU sales to outside customers are a GPU substitute
They do not affect hyperscalers' long-term in-house XPU strategy
Gross Margin
AI / system sales lower gross margin
but operating income dollars still grow
OpenAI
The 10GW is a 2027–2029 alignment
Not a major revenue contributor in 2026
The XPU project is a separate, independent track
Silicon Photonics
Necessary in the long run, but not mainstream in the near term
Copper / pluggable optics come first to sustain scale-up
⚠️ Risks and watch points
AI's rapidly rising share of revenue → changing margin structure
HBM / advanced packaging as potential bottlenecks
System-level sales add execution complexity
Non-AI semiconductor recovery still unclear
Risk of AI CapEx volatility (2027 and beyond)
🧭 Bottom line
Broadcom has formally evolved from a high-end semiconductor supplier into a core system-level player in the AI data center.
Its moat is no longer just chips, but: XPU + AI Networking + Optics + PCIe + System Sale + VMware Software
Against a backdrop of AI training, inference and commercialization all accelerating, Broadcom's backlog, technology breadth and supply-chain control make it one of the most critical "invisible hubs" of the AI infrastructure era.


Comments