AVGO

• Record Financial Performance: Q3 consolidated revenue hit a record $29.6 billion (up 86% YoY), driven by AI semiconductor revenue growth of 221% YoY to $16.7 billion. Operating income grew 92% YoY to $20.1 billion with a record operating margin of 68%.
• AI Semiconductor Momentum: Six XPU customers are accelerating custom accelerator adoption. Key shipments include Ironwood TPU v7 to Entropic and Google, TPU v8i production for Google, and Jalapeno for OpenAI. Q3 XPU shipments were over 3.5x YoY.
• Customer Roadmaps & Partnerships:
  • Google: Long-term agreement for future TPUs; planning multi-tens of billions in annual TPU supply.
  • Entropic: On track to become the largest XPU customer in 2027/2028, with 5GW deployment planned for 2027 and 10GW incremental capacity by 2028.
  • OpenAI: Jalapeno outperforms Grace Blackwell Ultra in inference; next-gen XPU approaching KBAR with line-of-sight for >5GW deployment by 2028.
  • Meta: Three generations of MTIA accelerators being delivered through end of 2027, with 3GW deployed by 2028.
• Networking Leadership: Tomahawk 6 (100/200 terabit) is seeing fast ramps and widespread adoption among hyperscalers. Tomahawk 7 (industry’s first 200 terabit Ethernet switch) has taped out. Tomahawk Ultra enables low-latency scale-up via Ethernet.
• Infrastructure Software: VMware Private AI Cloud launched to help enterprises run AI securely on-premises. ARR growth sustained at 15% YoY.
• Capital Allocation & XPV Platform: Established AI XPV platform with Apollo/Blackstone to finance compute infrastructure for OpenAI and Anthropic. Paid down $5.6B in debt post-Q3 and paid $3.1B in dividends.

• Q4 FY2026 Consolidated Revenue: Expected to be $34.8 billion, up 93% year-over-year.
• Q4 AI Semiconductor Revenue: Expected to be $21.7 billion, up 236% year-over-year.
• Q4 Non-AI Semiconductor Revenue: Forecasted at approximately $4.3 billion, up 5% sequentially.
• Q4 Infrastructure Software Revenue: Forecasted at approximately $8.7 billion, up 25% year-over-year.
• Margins: Q4 consolidated gross margin expected at ~73% (down from 78% YoY due to higher memory content in XPUs). Q4 operating margin expected at ~66%.
• Fiscal 2026 Full Year AI Revenue: Expected to be $58 billion (up 186% YoY), revised upward from prior guidance of $56 billion.
• Long-Term Trajectory: AI revenue expected to double to ~$115 billion in FY2027 and double again to ~$230 billion in FY2028.

Semiconductor Solutions: Revenue was $20.8 billion, up 127% year-over-year, representing 70% of total revenue. AI semiconductor revenue within this segment reached $16.7 billion (56% of total revenue). Gross margin was approximately 76%, and operating margin was 61%. Infrastructure Software: Revenue was $8.8 billion, up 29% year-over-year, representing 30% of total revenue. Gross margin was 94%, and operating margin was approximately 84%.

• Supply Chain Bottlenecks: Critical constraints exist in substrates, HBM memory, leading-edge wafers, and optical components (EML/CW lasers).
• Deployment Dependencies: Chip shipment volumes are contingent on customers securing land, power, and data center shells, which have long lead times.
• Gross Margin Dilution: Increasing proportion of AI revenue with higher memory content is diluting consolidated gross margins, though operating leverage mitigates impact on operating margin.
• Financing Risks: While the XPV platform structures financing off-balance-sheet, there are residual value guarantees and contingent liabilities associated with supporting specific lab deployments.

Q: Analyst asked about supply bottlenecks for the doubled AI revenue outlook ($115B in FY2027) and variables that could drive numbers higher. / A: Management stated demand exceeds supply but they provide conservative outlooks based on available supply chain (leading-edge wafers, substrates, HBM) and customer site readiness (land/power/shell). They believe $115B is achievable if supply scales, noting that deployment timing depends on customers' infrastructure build-out, not just chip delivery.

Q: Analyst questioned the gigawatt-to-revenue conversion ratio compared to competitors and whether all outlined gigawatts would deploy within the fiscal years. / A: CFO clarified that while 30GW is the total deployment target across key customers by 2028, not all may come online immediately due to data center construction timelines. The $115B/$230B figures represent confident shipping estimates rather than guaranteed immediate deployment of every gigawatt. Additionally, custom XPUs cost less than half of comparable GPUs.

Q: Analyst asked about the attach rate of Tomahawk Ultra networking to ASICs and adoption trends. / A: President of Semigroup confirmed that Tomahawk 6 is widely adopted for scale-out. Tomahawk Ultra, enabling low-latency Ethernet-based scale-up within racks, is seeing surprising adoption starting in Q3/FY27. Customers building XPUs or using GPUs are deploying both Tomahawk 6 and Ultra solutions.

Q: Analyst asked if CSP/funding partners dictate silicon choices for frontier labs like OpenAI/Anthropic, creating risk for Broadcom's certainty. / A: CEO explained that while short-term usage involves cloud services, these labs aim to become first-party hyperscalers running their own data centers. Broadcom provides financing vehicles (XPV) to help them bridge cash flow gaps, ensuring their silicon choice remains optimized for their specific workloads. This creates a strong economic moat and commitment.

Q: Analyst asked about other supply constraints beyond land/power/shell, specifically regarding substrates and memory. / A: CEO highlighted that substrates are a major bottleneck, prompting Broadcom to build its own substrate capacity in Singapore. Memory (HBM and system RAM) remains a critical constraint secured by customers. Optical components (lasers) are also tight, with Broadcom tripling capacity in US/Singapore factories to meet surging demand.