Broadcom FY25Q2 corresponds to the February/March/April 2025 period.
Broadcom FY25Q2 Earnings:
Revenue reached a record $15.004B, up 20% year over year and 1% sequentially.
GAAP gross margin 68%, up 6 percentage points year over year. Operating margin 39%, up 15 percentage points year over year. Non-GAAP operating margin 65%, up 8 percentage points year over year.
GAAP net income $4.965B, up 275% year over year. Non-GAAP net income $7.787B, up 48% year over year. Non-GAAP net margin 52% (flat vs. prior quarter).
Repurchased $4.2B in shares this quarter; paid $2.8B in dividends. Gross principal debt $69.4B.
FQ2 Segment Results:
Semiconductor revenue $8.408B, up 17% year over year, revenue share increased to 58%. Semiconductor gross margin 69%, up 1.4 percentage points year over year. Operating margin 57%.
Software revenue $6.596B, up 25% year over year, revenue share decreased to 44%. Software gross margin 93%, operating margin 76%. Most VMware contracts are ~3 years. Subscription transition progress over halfway complete (was 60% last quarter), with 1-1.5 years remaining for full transition. ~87% of top 10,000 customers now subscribe to full-stack VCF (was 70% last quarter), driving double-digit ARR growth in core software. Q3 software revenue guided at $6.7B, up 16% year over year. (Note: VMware revenue scale mentioned again this quarter; may not be disclosed separately going forward.)
Semiconductor segment detail for FQ2: (Note: segment revenue figures not disclosed again this quarter; may not be disclosed going forward. Only semiconductor AI vs. non-AI split provided.)
Semiconductor AI revenue $4.4B, up 46% year over year and 7% sequentially, in line with prior guidance, representing 52% of semiconductor revenue. Custom AI chip revenue $2.64B (Marvell custom AI chip revenue ~$700M this quarter), up double-digits year over year. Ethernet-based AI networking chip revenue $1.76B (NVIDIA AI networking chip revenue $4.96B this quarter, far exceeding Broadcom), up 170% year over year. AI networking vs. custom AI chip revenue split 40%/60%. AI networking share expected to decline to ~30% in FY26.
Semiconductor non-AI revenue $4.0B, down 2% year over year, at least the sixth consecutive quarter of decline. Management indicates approaching cyclical bottom. Enterprise networking, broadband, and server storage grew sequentially, but industrial and wireless declined sequentially.
FY25Q3 semiconductor AI revenue guided at $5.1B, up 16% sequentially. Non-AI semiconductor revenue guided at $4.0B, flat sequentially, with a slow recovery. Management states FY25 and FY26 AI revenue can both sustain ~60% growth, implying FY25 AI revenue ~$19-20B, FY26 AI revenue >$30B.
Broadcom's custom AI chip exposure is concentrated in three hyperscalers (Google, Meta, ByteDance). Four new partners added last quarter (OpenAI, rumored Apple, etc.). No new partners announced this quarter. Media reports suggest Google's next-gen TPU v7 will not be exclusive to Broadcom. Broadcom previously stated custom AI chip customer ramps typically take 6-12 months, with per-customer volumes no less than 5,000 units, and they do not engage startups as customers. CEO stated unable to determine if future export controls will impact the business (ByteDance).
Industry-first 1.6T Tomahawk 6 switch chip now sampling with customers; strong customer interest.
Outlook:
FY25 Q3 revenue guided at $15.8B, up 20% year over year. Semiconductor revenue $9.1B, up 25% year over year, of which AI revenue $5.1B, up 65% year over year, non-AI revenue $4.0B, down 4% year over year. Software revenue $6.7B, up 16% year over year. Gross margin guided down 1.3 percentage points sequentially, primarily due to higher custom chip mix. Adjusted EBITDA guided at $10.4B.
Capital allocation priority: return cash to shareholders via dividends, typically 50% of prior-year free cash flow. Remaining cash flow prioritizes debt repayment to bring net debt/EBITDA below 2x. Share repurchases at opportune times. Overall, cash primarily used for deleveraging, preserving borrowing capacity for potential large M&A. Deleveraging also supports M&A readiness.
Scale-up switch density is 5-10x higher than scale-out. Most scale-up interconnects still use copper cables because cluster sizes haven't reached the point where optics are mandatory. Once single clusters exceed 72 GPU/XPU interconnects, optical interconnects become necessary, making CPO relevant — but not the only option. Low-cost pluggable optics can serve initially. Optical interconnects expected to become increasingly prevalent over the next 1-2 years. CPO is one direction in development, but not the only solution.
Biggest question on the earnings call: Broadcom's prior SAM focused only on training. Now with inference demand exploding, the SAM should theoretically be larger, but management does not plan to raise it. Management believes there is no fundamental difference between training and inference when deploying general-purpose vs. custom AI chips.
Overall, judging purely by the numbers, this was a mediocre quarter for Broadcom. AI revenue merely met guidance (did not exceed), and next-quarter AI guidance was within prior expectations. Non-AI remains weak. Software growth is entering a plateau one year post-VMware consolidation. For over a year, many investors unwilling to accept NVIDIA's dominance rotated into AMD, Broadcom, and Marvell. The market's Broadcom crowding remains significant.
(Previously noted that many firms measuring AI chip share use incorrect denominators. For example, only ~60% of Broadcom's AI revenue this quarter is custom silicon. GPU is <50% of AMD data center. Marvell data center custom silicon may be ~50%.)
Previously, reverse-engineering from long-term SAM pie implied FY25 net income ~$30-32B; currently looks on track. Incorporating FY26 AI revenue details from this call, optimistic scenario puts FY26 revenue ~$80B; at 50% non-GAAP net margin that's $40B net income; 30x PE implies $1.2T market cap.